CHT

The generative AI boom is driving sales of servers used for artificial intelligence (AI) and high-performance computing (HPC), and dozens of companies will benefit from it. But one company will likely benefit more than others. Nvidia is estimated to sell over half of a million of its high-end H100 compute GPUs worth tens of billions of dollars in 2023, reports Financial Times.

Nvidia is set to ship around 550,000 of its latest H100 compute GPUs worldwide in 2023, with the majority going to American tech firms, according to multiple insiders linked to Nvidia and TSMC who spoke to Financial Times. Nvidia chose not to provide any remarks on the matter, which is understandable considering FTC rules.

While we don’t know the precise mix of GPUs sold, each Nvidia H100 80GB HBM2E compute GPU add-in-card (14,592 CUDA cores, 26 FP64 TFLOPS, 1,513 FP16 TFLOPS) retails for around $30,000 in the U.S. However, this is not the company’s highest-performing Hopper architecture-based part. In fact, this is the cheapest one, at least for now. Meanwhile in China, one such card can cost as much as $70,000.

Nvidia’s range-topping H100-powered offerings include the H100 SXM 80GB HBM3 (16,896 CUDA cores, 34 FP64 TFLOPS, 1,979 FP16 TFLOPS) and the H100 NVL 188GB HBM3 dual-card solution. These parts are sold either directly to server manufacturers like Foxconn and Quanta, or are supplied inside servers that Nvidia sells directly. Also, Nvidia is about to start shipping its GH200 Grace Hopper platform consisting of its 72-core Grace processor and an H100 80GB HBM3E compute GPU.

Nvidia dies not publish prices of its H100 SXM, H100 NVL, and GH200 Grace Hopper products as they depend on the volume and business relationship between Nvidia and a particular customer. Meanwhile, even if Nvidia sells each of H100-based product for $30,000, that would still account for $16.5 billion this year just on the latest generation compute GPUs. But the company does not sell only H100-series compute GPUs.

There are companies that still use Nvidia’s previous generation A100 compute GPUs to boost their existing deployments without making any changes to their software and hardware. There are also the China-specific A800 and H800 models.

While we cannot make any precise estimates about where Nvidia’s earnings from the sale of compute GPUs will land, nor the precise number of compute GPUs that the company will sell this year, we can make some guesses. Nvidia’s datacenter business generated $4.284 billion in the company’s Q1 FY2024 (ended April 30). Given the ongoing AI frenzy, it looks like sales of Nvidia’s compute GPUs were higher in its Q2 FY2024, which ended in late July. The full 2023 fiscal year is set to be record-breaking for Nvidia’s datacenter unit, in other words.

It’s noteworthy that Nvidia’s partner TSMC can barely meet demand for compute GPUs right now, as all of them use CoWoS packaging and the foundry is struggling to boost capacity for this chip packaging method. With numerous companies looking to purchase tens of thousands of compute GPUs for AI purposes, supply isn’t likely to match demand for quite some time.

When we think of Raspberry Pi robotics we typically think of wheels and motors but with the advancement of technology we now have quad and bipedal robots powered by our favorite single board computer, the Raspberry Pi. 

The Elecfreaks CM4 XGO Robot Kit is a $749 quadruped robot that looks like a dog, or a cat if you prefer. But this creature has a sting in its tail, a grabber arm that protrudes from its back. The face of the robot is a 2-inch screen with a camera placed above. The plastic shell contains the Raspberry Pi Compute Module 4, in this case a model with 2GB RAM and a 32GB micro SD card. 

When we think of Raspberry Pi robotics we typically think of wheels and motors but with the advancement of technology we now have quad and bipedal robots powered by our favorite single board computer, the Raspberry Pi. 

The Elecfreaks CM4 XGO Robot Kit is a $749 quadruped robot that looks like a dog, or a cat if you prefer. But this creature has a sting in its tail, a grabber arm that protrudes from its back. The face of the robot is a 2-inch screen with a camera placed above. The plastic shell contains the Raspberry Pi Compute Module 4, in this case a model with 2GB RAM and a 32GB micro SD card. 

Can we train this dog to respond to our commands, or are we barking up the wrong tree? To learn this and more we’ll need to grab the leash and take a trip to the dog park.

Elecfreaks CM4 XGO Robot Kit Specifications 

Getting Started with Elecfreaks CM4 XGO Robot Kit 

Unboxing doesn’t form part of the review, but we have to say that the kit comes in a heavy-duty. Inside the case we find plenty of padding and a robot dog taking a nap. In the upper lid, hidden under the padding, are a series of balls, charger and tools for the robot.

The packaging is exceptional, and something we would expect given the $750 price tag. Taking the robot out of the case and we are extremely impressed with the build quality. This is an all-metal robot dog, no plastic joints and everything is held in place with machine screws. The battery needs to be charged before use. It may have some charge in it, but the instructions specify that we shouldn’t let the battery drop below 20%. 

A DC jack on the belly of the robot provides a means to charge the battery which is safely encased inside of an aluminum case. We did notice that the wall wart got a little hot when charging the robot. In fact, we would say it was uncomfortably hot at around 45 degrees Celsius. The wall wart is an 8.4V 1A unit, an unusual voltage choice but there are units available. Just make sure that it is center positive. 

We also note that the green LED of the wall wart remains lit when the power supply is connected to the robot, but not the wall. This means that there may not be a diode to prevent the battery from discharging when left in this state. So after charging, remove the power supply!

Once charged up we are ready to power on the robot, the button for which is back the back legs. Keep your fingers clear of the legs otherwise you get a nasty nip. Powering up will take a minute or so. The Raspberry Pi Compute Module 4 will take a moment to boot and run the main menu. 

The onboard screen has three options. Program, RC and Demo. Using the top left and top right buttons we can navigate the menu, then press the bottom right to select. The demo section should be your first port of call. It has a selection of demo projects that show off what the robot can do: face tracking, color tracking, dancing etc. 

The robot can also understand simple voice commands; it takes a little practice to get it to understand my accent though. Face recognition can detect a person’s face, obviously, but the “Emotion” demo uses OpenCV and AI to determine your mood. It said that I looked sad, but rarely did I see “happy”. The detected emotion triggers the robot dog to whimper and shake, eliciting concern from the target human. 

We played around with the demos and this also proved that the robot was working to spec, ready for us to start programming. We noted that the servos were jitter free, the result of using serial servo motors over cheaper servos that use PWM (Pulse Width Modulation) to generate the pulses necessary to set the position.

Before we jumped into programming the robot we decided to have a little fun with an RC app. We installed the Android app and then set the robot to RC mode and connected via Bluetooth by shaking the device over the robot. This works 80% of the time, sometimes it fails to connect and this means we have to reset the connection, a common occurrence with Bluetooth devices. 

The RC app has basic controls for moving the robot, an advanced controls scheme for precision, arm control and a demo mode. The demo mode is largely the main demo broken down into user selectable options. After exercising our robot dog, it was time to start training our cyber pooch.

Programming Electreaks CM4 XGO Robot Kit

There are two supported ways to write code for the robot. Both of which are via a web browser. The first is via XGO-CM4, a block based editor which provides an easy introduction to writing code. The second is Python 3. The steps to reach this part of the review took a little back and forth with Elecfreaks. The instruction manual and website links pointed us to resources that were either incomplete or incorrect. 

After receiving this link from Elecfreaks we opened the link, chose the XG0-CM4 block editor for our first coding projects and this is where we hit another stumbling block. The code editor is hosted on an external IP address, http://47.252.22.82:8088/ and after selecting the blocks editor we have to give it the IP address of our robot. This is no problem, as selecting Prog from the robot’s screen will connect the robot to our network. Or it will do if you use Mozilla Firefox. Apparently “other browsers need to set up cross domain settings separately, which is quite troublesome”. We were able to use Firefox and everything worked but It just feels a little clumsy considering the quality of the hardware and its price tag.

The Python library has an unusual level of abstraction. Sometimes it abstracts the task, other times it does not. That said, once you get used to it, the XGO Python library is easy to use. The Educational library is a bit of a mixed bag when used outside of the web based editor. It does work, but you need to get your head around how it works, and a modicum of patience is required as it loads. 

On the whole, the software side of things is “good enough” but it brings down the quality of the product. It could be down to our workflow, but the stumbling blocks that we encountered will likely catch others out. In the end we chose to SSH into the robot and write our code in a text editor, referencing the documentation to create a simple demo project which involved our robot dog dancing to Britney Spears’ “Toxic,” which we can’t share for copyright reasons, and for shame.

Boston Dynamics’ robots are famed for navigating the world around them. From simple stairs to assault courses, these robots can’t be stopped. That can’t be said about our robot dog. We constructed a simple staircase from audio CD cases and using the blocks editor we set the robot to walk forward for five seconds. Using the normal walking settings, the feet of the robot cleared both sets of steps (with audio CD cases measuring 10mm on average).

Changing the gait to “trot” or “highWalk” meant that the robot dog was unable to clear even the first row of CD cases. You can manually set the height for each foot, by tweaking the x,y,z servos. This means that in order to navigate a rough terrain we would either have to run through the terrain, changing settings on the fly, or create a loop that will exaggerate the leg motion of the robot. Navigating a ramp is possible; we found that a plastic ramp, raised 30mm tall (using CD cases) was possible, but 40mm was not. The reason for this is grip. The ABS plastic feet of the robot have no grip on plastic / tiles / wooden desks. Putting some rubber boots on the feet will aid grip, we wonder why this wasn’t done in the factory?

Calibrating the legs isn’t possible via the Blocks editor (that we can see). But they can be calibrated using the remote app, well sort of. Under “Enhanced options” we can set the gyroscope status and use that to “right” the robot according to the terrain it is on. This helps RC controllers to keep the robot dog upright according to the limitations of the servos for each leg. 

Absolute precise control is possible using the Python module and it is best used when starting up the robot. The Python module’s calibration function will return the robot to the default position, legs folded and the robot laying on the ground. From there the user can either follow a pre-scripted sequence, or write their own Python code to control the robot. We looked around the documentation pages for more calibration options but sadly found many pages were not created (404 errors). Still this is far superior to the situation with the Petoi Bittle robot dog where we had to do a lot of manual leg calibration and had to repeat the process many times.

The Dog With Two Brains

Inside of the robot dog there are two brains. The face of the dog contains the Raspberry Pi Compute Module 4 and a breakout board which provides a camera, microphone and sensors used to keep our robot dog upright. The other brain lies deep in the chassis and is an ESP32, solely there to control the many servo motors.

The ESP32 is no stranger to being a co-processor. Arduino’s latest board, the Uno R4 WiFi also features this chip, a chip that is an incredible microcontroller in its own right. The CM4 and the ESP32 communicate using a serial connection, a four pin cable that runs from the face into the body of the robot. This connection does mean that there is a small delay when we start our code as the serial connection is made.

Who is Electreaks CM4 XGO Robot Kit for?

At $749 this is a big investment and probably prices a few people out, but let’s put this kit into context. For the price we get a quadruped robot with a similar form factor to Boston Dynamics’ robot dog. So our small dog comes with a smaller price tag than the big dog! We can see the robot kit being a big hit in education but not so much in the home.

Could this be your faithful robot dog telepresence unit? Well sure it could, but Elecfreaks doesn’t directly support this out of the box. The challenges faced in this endeavor are that we need to stream live video and audio from the dog. This is possible using the Picamea Python library, and performance is decent.

Don’t expect crystal clear video as you move around the office. We’ve used streaming video from a Raspberry Pi 4 in a project and sometimes the video can corrupt (think cable TV levels of image distortion) but it generally corrects itself. Audio pickup via the microphone is also possible, but that would require another Python module to handle audio. The XGO Python module has a function for audio playback, which we used to make the robot dog dance to Britney Spears.

The biggest obstacle to using it as telepresence unit would be controlling the robot while all of this is happening. Face tracking would handle locating the person that you wish to speak to, but navigating the world would be slow and require some form of manual control. It is possible, you could make an Anvil app which remotely interacts with the robot dog, but the effort versus reward would be imbalanced. Lastly, the strain of streaming, tracking, audio, walking would be immense on the battery. How long would it last before you need a recharge?

Bottom Line

You’ll need deep pockets to purchase Electreaks CM4 XGO Robot Kit, but for the money you get a lot of great hardware in an incredibly well made package. The software is decent, but falls below the quality of the hardware. That said, the software is relatively easy to use, both the Python and block coding.

We’re not especially keen on the Mozilla Firefox only browser restriction. Sure we can install another browser, they are free after all. But it is another barrier to entry, and may be troublesome for educators with locked down equipment.

Lenovo has started selling its latest Yoga Slim 7 Gen 8 laptops in Europe and Yoga Air 14S in notebooks in China based on customized versions of AMD’s Ryzen 7040-series accelerated processing units (APUs). According to IT Home, the new Zen 4-based processors come in FP8 packaging to enable better connectivity and performance while retaining their low power consumption. 

The new 14-inch Lenovo Yoga machines (which are called differently in China and Europe) use AMD’s Ryzen 7840U processor in the FP8 packaging (instead of the FP7r2), which is why it is called the Ryzen 7 7840S but retains eight Zen 4 cores clocked at 3.30 – 5.10 GHz and Radeon 780M RDNA3-based GPU with 768 stream processors. At this point configurable thermal design power of the Ryzen 7 7840S is unclear, but it could be different from the original version. 

Meanwhile, VideoCardz reports that some of Lenovo’s Yoga Slim and Yoga Air machines can also use AMD’s custom Ryzen 5 7640S in the FP8 form factor. In fact, the website says that Lenovo will exclusively use the Ryzen 7 7840S and Ryzen 5 7640S, though we would expect this exclusivity to last for a few months only.

AMD’s Ryzen 7040-series processors come in three types of packages: compact FP7 (LPDDR5X) and FP7r2 (DDR5) optimized for thinner designs that need a high-performance APU as well as larger FP8 that is designed to deliver the highest functionality and performance. The FP8 package is notably larger than its FP7 counterparts. It is designed to support higher-performance interfaces, such as AMD’s MIPI CSI, a high-speed interface to connect advanced cameras to host SoCs.

Indeed, Lenovo’s Yoga Slim 7 Gen 8 laptop is not the most compact and lightweight system around: they are 13.9 mm thick and weigh 1.35 kilograms, but it offers quite formidable performance and feature set. The machine comes with a 14.5-inch OLED display with a 2944×1840 resolution with up to 400 nits luminance, 16 GB of LPDDR4X-6400 memory, and a 512 GB M.2-2280 SSD with a PCIe 4.0 interface. The notebook has everything one expects from a high-end PC in 2023, including Wi-Fi 6E 2×2 and USB Type-C connectivity.

If you’re on the hunt for a replacement controller for your games console or perhaps a new gamepad for your PC, then take a look at the Sony PlayStation 5 controller at $49— that gives you an awesome $20 discount. With haptic feedback and a built-in microphone, the Sony controller is a great choice for gaming if a mouse and keyboard doesn’t suit the game.

Pick up 2TB of storage with this excellent price on Crucial’s P3 SSD for $74. This Gen 3 SSD can reach read/write speeds of 3500/3000 MB/s and is a great option for a budget drive for your games library. It’s an older SSD and may not be the best SSD on the market, but its price-to-performance ratio is a winner. 

And for only a few dollars more than the above offering, you can upgrade to a 2TB Gen 4 Crucial P3 Plus SSD for just $79. This newer drive offers an increase in performance with speeds of 5,000 MB/s sequential read and 4,200 MB/s write. This is a great price for an SSD for your games, or if you’re looking to upgrade the storage in a laptop.

For more deals keep on scrolling.  

TL;DR — Today’s Best Deals

Today’s best deals in detail

Looking for more deals?

Glorious is mostly known for its hole-filled, ultra-lightweight mice, such as the 2.05oz (58g) Model O Minus. Its newest mouse, the Model I 2 Wireless, might actually be its heaviest mouse to date, even if it’s still ultra-light at just 2.65oz (75g). 

The Model I 2 Wireless is an interesting pitch: a lightweight, hole-filled mouse with bright RGB light strips and nine programmable buttons — including four side buttons, two of which are swappable. Glorious says the mouse is ideal for MOBAs and MMOs, thanks to its many buttons — and while the Model I 2 Wireless has significantly more buttons than Glorious’ other mice, it’s hardly breaking records. Still, if you’ve been looking for a great gaming mouse that’s lightweight without being extremely stripped down, the Model I 2 Wireless — which features Glorious’ latest BAMF 2.0 sensor, 2.4-GHz wireless and Bluetooth connectivity, and onboard storage for 3 profiles — might be one of your only options.  

The Model I 2 Wireless is available now for $99.99, and is available in both black and white colorways.

Design and Comfort of the Model I 2 Wireless

The Model I 2 Wireless is a right-handed wireless mouse with a lightweight, hole-filled chassis and nine programmable buttons. It has an ergonomic, contoured shape with split main buttons that curve lightly to hug your fingers, and a shallow thumb rest. It’s on the larger side, measuring 5.08 inches (129mm) long by 2.83 inches (72mm) wide at its widest point — minus the thumb rest, it’s only 2.44 inches (62mm) wide — and 1.65 inches (42mm) high.

It’s slightly smaller than the similarly-sized Razer Basilisk V3 Pro (5.11 x 2.96 x 1.67 inches / 130 x 75.4 x 42.5mm) and the Logitech G502 X Plus (5.17 x 3.12 x 1.62 inches / 131.32 x 79.25 x 41.15mm), though this is still a mouse that’s best-suited for someone with medium- or large-sized hands. The DPI switch button is located closer to the body of the mouse than it is on the Basilisk V3 Pro and the G502 X Plus, however, so someone with smaller hands at least won’t be straining for any buttons on the Model I 2 Wireless.

The Model I 2 Wireless is also significantly lighter than both the Basilisk V3 Pro and the G502 X Plus, weighing just 2.65oz (75g) (both the Basilisk V3 Pro and the G502 X Plus weigh 3.74oz / 106g). I’m not sure how much of the Model I 2 Wireless’ diminutive weight can be attributed to its chassis (as opposed to internals — both the Basilisk V3 Pro and the G502 X Plus have quite a few features the Model I 2 Wireless lacks), which is made of plastic and is full of holes. I’m not a particularly big fan of the holes, but at least they let some of the mouse’s RGB shine through. The mouse has two RGB lightstrips on either side of its body, which are semi-programmable via the Glorious Core software.

The Model I 2 Wireless has nine programmable buttons: left/right click, scroll wheel click, two buttons above the scroll wheel, and four side buttons (three thumb buttons and a DPI shift button). Two of the four side buttons (the third thumb button and the trigger button) are swappable; the mouse comes with five alternate button shapes — three for the DPI shift button and two for the thumb button. It also comes with a sim card tray eject tool, which you can use to pry the buttons out for swapping (this is pretty easy, if not quite as intuitive as using the tool for ejecting a sim card tray).

While swappable side buttons are an interesting concept — albeit, not at all unique — Glorious’ implementation is… underwhelming. There are only two alternate buttons for the third thumb button — one juts out a little further (to make it easier to press for those with less flexible thumbs, I assume), while the other is a button delete. There are three alternate buttons for the DPI shift button: one flips the button so the raised portion is closer to your thumb, one is lower and less pronounced, and one is a button delete.

While I did find the buttons to be different enough that they each gave the mouse a distinct feel, that feel wasn’t necessarily good. The alternate button for the third thumb button and the flipped alternate for the DPI shift button both jutted out from the mouse far enough that I found them uncomfortable to use.

The only alternate I can see users readily using is the DPI shift button delete — since the DPI shift button is right under your thumb and will almost certainly get accidentally clicked if you ever pick up your mouse. The third thumb button delete felt a little unnecessary because of the button’s location (although I did accidentally click that button several times when I had the alternate button in — but it seems like you could just get rid of both the alternate and the delete and be fine).

It’s not that big of a deal, but it does feel a little messy, especially since you end up with a bunch of tiny, easy-to-lose parts. I definitely prefer the way Logitech implements its button-swapping on the Logitech G502 X Plus, which has a 2-in-1 reversible DPI switch button that doesn’t require a tool to remove, as well as a button delete.

On the bottom of the mouse you’ll find six small 100% PTFE skates, a power switch that toggles between 2.4 GHz and Bluetooth, and a DPI light. The DPI light simply changes color to indicate which DPI step you’re on (they’re color-coded in the software), and would probably be more useful if it weren’t on the bottom of the mouse.

The mouse comes with a 2.4-GHz wireless USB-A dongle, a USB extender, a 6.5-foot (2m) USB-C to USB-A charging cable, five swappable button alternates, and a button swapping/sim card ejection tool. The mouse’s charging cable is lightweight, flexible, and relatively drag-free, so you can use the mouse in wired form without losing too much movement freedom.

Specs

Performance of the Model I 2 Wireless

The Model I 2 Wireless sports the Glorious BAMF 2.0 sensor, which has a maximum sensitivity of 26,000 DPI and a maximum speed of 650 IPS, and can handle up to 50 G’s of acceleration. This is on par with other high-end mouse sensors, and the mouse was precise, accurate, and responsive in my testing — and it worked well on a variety of surfaces, including hard-surface, soft-surface, and glass mouse pads. The main buttons feature Glorious’ own switches, which are tactile and clicky (if a little on the loud side), and are rated for 80 million clicks. 

Glorious keeps calling this an ideal mouse for MOBA and MMO games, and while it does have more than the standard two extra buttons, I think that’s a stretch. The nine programmable buttons include the two buttons behind the scroll wheel, which are located pretty far back on the mouse and not very easy to press, as well as things like the scroll wheel click (at least they’re not including the scroll wheel directions as programmable buttons, which is what Razer likes to do). 

But I love mice with millions of buttons, and the Model I 2 Wireless is definitely on the lower side, button-wise — it doesn’t even have a tilt wheel. I don’t think this mouse is “ideal” for MOBAs/MMOs — I’d call it an all-purpose gaming mouse that’s probably better for gamers who lean toward having fewer, rather than more, buttons.

The Model I 2 Wireless is extremely lightweight, and that helps it fly across your mouse pad. It’s not the lightest mouse on the market, but it’s much lighter than most similarly-sized (and similarly-shaped) mice. Of course, if it’s designed for MOBA and MMO players, that light weight isn’t necessarily going to be as appreciated as it would be if the target audience were FPS and eSports players. But the light weight definitely makes for a comfortable gaming experience, regardless — I’m not sure if the mouse had a particularly ergonomic design or if it was just so lightweight that it didn’t matter, but I definitely didn’t feel any wrist or arm fatigue while playing approximately 700 hours of Baldur’s Gate 3.

I did find the scroll wheel to be a little stiffer than I preferred, however. Those of you playing Baldur’s Gate 3 know that the default controls involve a lot of scroll wheel action, and the Model I 2 Wireless’ scroll wheel is slightly on the tough side for both scrolling and clicking.

Features and Software of the Model I 2 Wireless

The Model I 2 Wireless works well out of the box, but you’ll need to download Glorious’ universal peripheral software, Glorious Core, to customize lighting, program buttons, and adjust the mouse’s DPI presets and performance settings. The mouse has onboard memory and can store up to three profiles.

Glorious Core isn’t the worst peripheral software I’ve used, mostly because it’s relatively simple. Of course, this means the mouse is not quite as customizable as mice from Razer or Logitech, especially when it comes to lighting. Glorious only allows you to pick from eight lighting effects, most of which are either spectrum or single color. There are two lighting effects that let you customize beyond a single color, but they’re both transitioning effects (a dual-color flashing effect and a six-color breathing effect), so you’re essentially working with single-zone RGB.

This may not be a big deal to some (especially since the Model I 2 Wireless’ lighting is a pretty big battery drain), but I think it’s worth knowing that you won’t be able to create multi-zone effects like you can with the Logitech G502 X Plus.

The key mapping part of the software is much more robust. It allows you to program the mouse’s buttons with various functions, multimedia settings, shortcuts (such as opening a program or a website), and has a built-in macro recorder. It also lets you program a second layer of functionality (you’ll need to program a button as the layer shifting button, of course). In the performance menu, you can adjust the DPI stages, as well as lift-off distance, debounce time, and polling rate.

Wireless Experience and Battery Life of the Model I 2 Wireless

The Model I 2 Wireless offers three forms of connectivity: low-latency 2.4-GHz wireless, Bluetooth, and wired via USB-C. I didn’t have any issues with the 2.4-GHz wireless connection, though the mouse did start to act a little weird when it got to below 10 percent battery life (which is not unexpected). Glorious rates the Model I 2 Wireless’ battery life at 210 hours over Bluetooth and 110 hours over 2.4-GHz wireless, which is on the higher end of average for premium gaming mice these days. 

This is with the lighting turned off, of course. I found the battery drained shockingly quickly with the lighting turned on high — it went from 100% to flashing red as a low battery warning in less than a day — and I wasn’t even using the mouse at the time. This was with the “Glorious Mode” lighting effect turned up to full brightness. I’m not sure what I expected, but it seemed exceptionally quick considering it was just sitting on my desk and not even moving. There is an option in the software to set separate brightness levels for wired versus wireless modes, which is a nice touch.

Bottom Line

The Model I 2 Wireless isn’t a bad gaming mouse — it’s got a lot going for it, and it’s one of few lightweight mice on the market with more than two extra buttons. But it seems to be sort of in between categories: it doesn’t really have enough programmable buttons for hardcore MOBA/MMO players, nor is it light enough or small enough for hardcore FPS players. 

It looks like some of the best all-purpose gaming mice, but it lacks a lot of the features found in mice like Razer’s Basilisk V3 Pro and Logitech’s G502 X Plus (such as programmable tilt wheels, dual- or multi-mode scrolling, and game-specific profiles). And while it has great battery life with the lighting off, it has awful battery life with the lighting on (case-in-point: the battery just dropped from 100% to 93% in the 20 minutes it took me to write this last section — and I wasn’t even using the mouse, I was typing). 

If you’re looking for a lightweight mouse that’s larger and has more buttons than something like the Razer DeathAdder V3 Pro or the Logitech G Pro X Superlight, then the Model I 2 Wireless might be the perfect mouse for you. But considering one of its stronger points is how pretty the lighting is, you may want to stick with one of Glorious’ wired mice — like the Model I, which is almost half the price.  

Tesla, Inc., the clean-energy company that’s about more stuff than just some of the world’s most desired electric cars, is seemingly on the lookout to improve its datacenter infrastructure. According to a new job listing on Tesla’s corporate website (spotted by Elektrek), the company is looking to hire a “Sr. Engineering Program Manager, Data Centers.” That hire is usually a good step for any company planning to operate datacenters built on custom or proprietary silicon – perhaps Tesla is looking to build a dojo for its Dojo AI accelerators?

“This role will lead the end-to-end design and engineering of Tesla’s 1st of its kind Data Centers and will be one of the key members of the factory engineering team.”

The position is based in Austin, Texas, where the company has several facilities focused on manufacturing and R&D. However, that doesn’t mean that its effects will only take effect in Austin – especially considering reports that Tesla has taken over some number of X’s (formerly Twitter) datacenters back in Sacramento. It’s also unclear whether Tesla is just being hyperbolic regarding how “first of their kind” these datacenters will be. There are multiple ways to fit that definition that doesn’t involve as much engineering work as one would expect.

Tesla originally announced its Dojo D1 (it seems the product name has since changed to Dojo V1) in 2021, even as it promised to increase the amount of processing power available for the company to train its self-driving AI systems. At the time, these Dojo ASICs (Application-Specific Integrated-Circuit) were meant to carry up to 50 billion transistors each, delivering around 362 TeraFLOPs of power per custom chip.

It’s unclear whether Tesla has toyed with the design in the meantime (although we’d say that’s likely). What is unclear, however, is that Elon Musk has clarified that the first generation ship won’t include general AI processing and will instead focus on accelerating “video training” for the firm’s computer vision systems. According to Musk, V2 of Dojo will address these limitations and eventually become a full-fledged, general AI processor not unlike NVIDIA’s hot-of-the-presses A100, H100, and its DGX GH200 supercomputing system. 

Tesla has already purchased a number of GPU accelerators from Nvidia. The company already bragged about it even when it “only” deployed 7,360 A110 accelerators. But that number has already increased; the company aims to have as many as 100 ExaFLOPs on-hand by October 2024 (already counting the deployment of its Dojo supercomputer).

Tesla designing and ordering its own custom ASICs and general-purpose AI accelerators should give the company increased control over feature-sets and reduce the Total Cost of Ownership (TCO) for its datacenters. But then again, all companies designing silicon these days mostly take honey from the same TSMC-branded pot; and there’s only so much capacity to go around. 

So while Tesla is likely to enjoy several benefits by designing and integrating its own High-Performance Computing (HPC) systems, it’s unlikely that Elon Musk will find fewer reasons to complain that “everyone and their dog” is buying GPUs. There are only so many wafers to distribute across the cadre of TSMC clients, and most would also love to stay at the cutting edge of fabrication.

A new PC builder website has cropped up that helps gamers and professionals build a potent bang-for-buck system with the power of artificial intelligence. Known as pcbuilderai.com, the site aims to take the guesswork out of building an effective PC parts list and give users the best possible parts list with the best components for the price. But, despite the site’s modern and simplistic UI, some of the parts it recommends are out of whack with today’s market prices.

The creators behind PC Builder AI is a company known as AE Studios. According to the company’s Indeed page, the company has no venture capital, no outside shareholders, and is apparently focused on creating tech for the development of humanity. The company has already built several AI-generated programs, including games, image generators, and text-based AI generators.

To test the effectiveness of the PC Builder AI parts list generator, we grabbed three of our Best PC Builds for Gaming parts lists, priced at $500, $1500, and $3600, respectively, and compared them to the three parts lists the AI-accelerated site generated with the same prices.

Starting at the $500 price class, the PC Builder AI site generated a moderately decent gaming machine featuring a Ryzen 3 3200G. Still, it’s nothing special, and some components are overpriced. The biggest offender is the memory, which is $40 more expensive than the kit we recommended, which has the same capacity and speed. The storage configuration is also bizarre, consisting of a separate 1TB HDD and a tiny 120GB SATA 3 SSD reminiscent of pre-2020 era systems. With the Best SSDs priced at an all-time low right now, you can find 1TB M.2 drives for the same price as the SSD and the HDD combined. This would yield a far better gaming experience and significantly increase the storage speed overall.

The lack of a graphics card is also unfortunate for the AI generator. In our build, we could squeeze in a significantly faster Core i3 CPU and an Arc A310 discrete GPU, which will be much more capable than what the 3200G can provide.

Sadly, the $500 system was the best machine the PC Builder AI website could generate. The $1500 price bracket reveals many pricing problems with the AI system, leading to terrible component choices.

The AI generator’s most absurd component choice was the GPU, choosing a $600 RTX 3060 Ti. If you know anything about the GPU market today, you’ll know that paying $600 for a previous-generation mid-range 60 series card is utterly ridiculous. You can get a current generation RTX 4070 for that price, which is substantially faster than the 3060 Ti. If you want to step it up further, you can get an RX 6950 XT for that same price. It is AMD’s flagship graphics card from the same generation as the 3060 Ti and one of the Best GPUs for gaming.

The CPU choice is also horrible. For some reason, the AI generator chose another previous generation product, a Ryzen 5 5600X, priced at $280. In today’s market, that CPU is going for nearly half that price, and a modern Ryzen 5 7600 is still $40 cheaper. For $280, you can get a current generation i5-13600KF — one of the Best CPUs —that can run circles around the 5600X in gaming workloads and heavily multithreaded applications.

The rest of the system is also not great; the motherboard, memory, cooler, and storage options could be much better. Even though the CPU and GPUs are wildly overpriced, we could put in more storage capacity, memory capacity, a higher wattage PSU, and a better cooler in our $1500 build recommendation while spending the same amount of money on the CPU and GPU.

The same pricing errors also appear in the $3600 build, where the CPU and GPU are incredibly overpriced. The PC Builder AI generator opted for a $800 Ryzen 9 5950X and a $1200 RTX 3080 Ti in this price class. Right now, a 5950X is going for as little as $450, and a 3080 Ti class performance is going for as little as $600 in the form of a Radeon RX 6950 XT. If you want an Nvidia counterpart, the RTX 4070 Ti is roughly $800 and generally outperforms the 3080 Ti.

Thankfully, the rest of the components are not too bad for a $3600 build and have decent quality/performance for the money, but it’s not perfect. The SSD choice, in particular, is very strange, with the AI opting for a PCIe Gen 3 970 EVO Plus 1TB SSD. For just $10 more, you could get a 990 PRO that is PCIe Gen 4 capable with 2TB of storage, or opt for a 1TB Crucial T700 with PCIe Gen 5 support. The 64GB RAM option is also strange but somewhat appropriate for a $3600 build. We suspect the AI is targeting the $3600 price point with streaming in mind, which would explain the capacity choice.

Overall, the results AE Studio’s PC Builder AI is generating is very bizarre, and it is not a site we can recommend you use for sound guidance on PC part selection. The site never recommended a current-generation CPU or GPU and priced the graphics card options from the GPU shortage in 2021. We are unsure what AI system the site uses, but it possibly uses 2-3-year-old information, which would explain the bizarre component choices.

If you want good advice on good parts to choose for your next computer, check out our forums or our AI Chatbot, which will generate far better results.

Intel’s proposed acquisition of Tower Semiconductor has been approved by regulators across the world except in China, which is why the deal has not closed nearly 19 months after its announcement in mid-February of 2022. The deadline for Intel to close the deal after another extension is tomorrow, August 15, 2023, but the two companies cannot close it without approval from China. Meanwhile, it is unclear whether the companies will extend the closure period and refile with other regulators.

When Intel originally announced plans to take over Tower Semiconductor, it said that it would close the transaction in 12 months, by mid-February, 2023. But because China’s State Administration for Market Regulation (SAMR) had not approved the deal, the companies extended the acquisition period until mid-June and then extended it again to August 15. So far, SAMR has not formally greenlit the transaction and if it does not OK it by tomorrow, Intel and Tower will either have to extend the closure period one more time, or call the deal off.

Pat Gelsinger, chief executive of Intel, recently flew to China in a bid to convince Chinese regulators to approve the deal, but the outcome is currently unknown. Since Chinese semiconductor sector is suffering from major sanctions from the U.S. government, China’s government is not really inclined to support Intel’s plan to acquire Tower and its dozens of customers and mature fabrication processes. Tower competes against China-based contract makers of chips, such as SMIC and Hua Hong, whereas Intel Foundry Services (IFS) division is barely competing against these foundries. Yet, once Tower becomes a part of Intel, the new entity will be a formidable rival for Chinese chipmakers.

To that end, SAMR is slowing down approval of the deal, even though it does not create a monopoly of any kind. In theory, companies can agree to extend the closure period of an acquisition for as long as they mutually decide. However, in practice, there are several factors that limit the feasibility of indefinitely extending the closure period.  

For example, in many jurisdictions, regulatory bodies approve mergers and acquisitions for certain periods. If companies keep extending the closure period, it might raise red flags or lead to additional scrutiny. It is possible that to prolong the closure periods, Intel and Tower will have to refile their proposals with regulators in the USA or Europe if the transaction is not approved in China by August 15.

Intel has one day left to close the Tower Semiconductor acquisition, extend the closure period, or cancel plans to take over the contract maker of chips. 

 

Intel CEO Pat Gelsinger spoke at the Aspen Security Forum 2023 in July; The topic under the spotlight was that of semiconductors and national security, and the issues that have arisen from the passing of the CHIPS Act. One of Gelsinger’s most eyebrow-raising assertions was that Intel deserves a bigger slice of the $52 billion US CHIPS Act pie than foreign rivals TSMC and Samsung. It was a bold claim, and he put forward a convincing case in what could be a high-stakes / high-reward gambit (h/t to EETimes).

Security, and China export controls particularly, is a thorny topic for successful global semiconductor businesses with their roots in the US. As US sanctions on China tighten it is making business for the likes of Intel (and Nvidia, and Qualcomm, and AMD) more difficult. Last month we reported on US semiconductor industry bosses having negotiations with senior US government economics and national security figures.

At Aspen, Gelsinger put forward the case for Intel to face fewer restrictions from trade regulation and get a larger portion of the US CHIPS Act cash. His multi-pronged argument appeared to be designed to hammer home the following points:

Firstly, China represents 25% to 30% of Intel’s market, and its current $30 billion US expansion plans were partly put in place to address this demand, according to Gelsinger. If trade restrictions tighten and further cut this geographic revenue stream, then it could mean Intel may not want to continue to make a significant investment in US projects, which have been touted to be good for US jobs, the US economy, and even US national security. Overall, the point was that China export controls unfairly target Intel business and that is bad for the US.

Secondly, Gelsinger talked about the current set of trade restrictions already being excessive and badly targeted. “Today, we have over 1,000 companies on the Entity List, many of which have nothing to do with national security, and nothing to do with security concerns in China,” noted the Intel CEO. So, in essence, Gelsinger was arguing for a loosening of the restrictions now in place, to refocus tightly on areas of undeniable national security.

Lastly, and probably most controversially, Gelsinger made his case that Intel should get a bigger slice of the US CHIPS Act’s $52 billion pie. The Intel CEO’s argument was quite clear and reasoned. He said the US should celebrate the fact that TSMC and Samsung were investing in building in the US, but there is a glaring difference in their operations and Intel’s. “All of my essential R&D is done here. Most of their work is done overseas,” Gelsinger underlined. “We should benefit more.”

If you care to watch the video embedded above, you can see Gelsinger underline the necessity for the CHIPS Act, and provide some of the history leading up to its passing. Moreover, you can listen to him make Intel’s key arguments for loosening the current trade restrictions, while being more generous to Intel – and his reasoning.

As well as the interviewer Steve Clemons, an editor at Semafor, you will see Penny Pritzker, the US Commerce Secretary, sitting alongside Pat Gelsinger. Pritzker as a representative of US government, discussed the federal government’s thinking and strategy. For the US to reinvigorate the semiconductor industry, it isn’t all about throwing money at chipmakers. For example, there is the widening chasm in skilled worker requirements compared to actual labor availability.

It seems that AMD’s issued patch for its Zen 1 “Division by zero” bug wasn’t the end-all, be-all the company wanted it to be. While the company was fast in issuing a patch, there’s now the suspicion that they might’ve been just a bit too fast: according to Michael Larabel with Phoronix, AMD Linux Engineer Borislav Petkov published a new patch that fixed an issue with the original solution (also published by him). It’s just another datapoint on the difficulties of hardening against possible attack vectors.

The original bug related to how Zen 1 processed an integer calculation divided by 0 in certain circumstances: according to the findings, there was the possibility that AMD’s CPU kept “stale quotient data” within its registers even after the operation was fully finished, which could give attackers a window to retrieve sensitive information. The original workaround was to perform a final “dummy division 0/1 before returning from the #DE exception handler”. The idea is simple: whatever old data was still stored would be wiped upon the completion of the 0/1 division (whose result is always, well, zero).

The issue with that solution, as Petkov explained, was that by the time that security provision kicked in, the speculative execution attack would have already advanced too far: there would already be some amount of old data on AMD’s divider, which the attackers could get at before the dummy division kicked in. As Petkov explained it, his new solution now forces that same division in a number of scenarios:

“Initially, it was thought that doing an innocuous division in the #DE handler would take care to prevent any leaking of old data from the divider but by the time the fault is raised, the speculation has already advanced too far and such data could already have been used by younger operations.

Therefore, do the innocuous division on every exit to userspace so that userspace doesn’t see any potentially old data from integer divisions in kernel space.

Do the same before VMRUN too, to protect host data from leaking into the guest too.”

It’s already been a busy month for vulnerabilities in the CPU realm, with both AMD and Intel both having been hit with disclosures. From Intel’s more extreme Downfall vulnerability (affecting Skylake through Tiger Lake/Rocket Lake) through AMD’s SQUIP and Inception vulnerabilities (and the now re-fixed “divide by zero” vulnerability, researchers have been hard at work. It still doesn’t compare to the storied history of Meltdown and Spectre days (although these new bugs are also related to speculative execution vulnerabilities. Speculative execution refers to the way modern CPUs try to pre-empt calculation steps before they’ve even become necessary, so that the required data is already available in case it’s called to the execution. Yet while the fixes to some of those vulnerabilities have carried (sometimes severe) performance penalties, it’s at least a good sign that AMD’s 0/1 dummy division doesn’t come with additional overhead.

At the same time, it’s heartening to see that at least in this case, the security patch wasn’t issued in a sort of “set it and forget it” manner – with the sort of merry-go-round work that blue team experts have to carry, there were other ways this could have gone (the deficient patch could’ve been believed to fully work, leaving the door open to further hacking explorations down the road (with whatever impact those might carry).

Your keyboard is your primary connection to your PC, the part you touch the most. Getting a better keyboard can deliver a more immediately-noticeable difference than nearly any other upgrade — including upgrading your CPU. 

To a serious gamer, a keyboard is even more critical. Gamers need keyboards with instant response time and tactile feedback, but they also need it to be comfortable and easy to type on for long gaming and chatting sessions. And an attractive design with per-key RGB that can be customized to accompany your rig doesn’t hurt, either.

Whether you’re ducking, strafing, and sniping your way through a first-person shooter, wearing out your W-key exploring a vast open world, or chatting furiously on Discord and Twitch, your keyboard will be your main sidekick. So you don’t want to make this choice lightly.

Below, you’ll find the best gaming keyboards we’ve tested, in every size, color, and style. These keyboards hit the perfect combo of low-latency responsiveness, convenient, gamer-oriented features, high levels of customization options, and, of course, attractive RGB lighting. 

The Quick List

Best Gaming Keyboards You Can Buy Today

Best Mainstream Gaming Keyboard

The HyperX Alloy Origins’ compact and quality build makes it the best gaming keyboard for mainstream gamers who just want to get down to gaming. It’s comfortable, with a premium look and feel from its keys to its frame and vibrant RGB lighting. If you’re familiar with linear switches, which are quick but offer little feedback, you’ll feel right at home with this keyboard’s red switches. We’ve also tested the clickier version, which uses HyperX’s aqua switches and is also available  at Amazon and HyperX’s store. 

For over $100, you can find gaming keyboards with more luxuries, such as media control buttons, a USB pass-through port (for easily plugging in another accessory, like your best gaming mouse), and more advanced software. The Alloy Origins is definitely a no-frills keyboard — but what it does offer, it executes very well. 

If you’re not looking for a full-size keyboard, HyperX does offer smaller (and cheaper) versions of this keyboard. We’ve also tested the HyperX Alloy Origins 65 and the mini HyperX Alloy Origins 60, and HyperX also offers a TKL version — the HyperX Alloy Origins Core. 

Read: HyperX Alloy Origins review

Best Feature-Packed Gaming Keyboard Splurge

The Razer BlackWidow V4 Pro is the ultimate, full-size, feature-packed customizable gaming keyboard to outfit your battlestation. It’s got everything you could need or want: 8 dedicated macro keys, a programmable multi-function “Razer Command Dial,” four dedicated media keys and a volume roller, and a plush, padded detachable wrist rest that lights up when you connect it to the keyboard. 

The BlackWidow V4 Pro is a wired keyboard that comes with Razer Green (clicky) or Razer Yellow (linear) mechanical switches. It features an aluminum alloy top plate, doubleshot ABS keycaps, and has bright, per-key RGB as well as underglow with a total of 38 zones when the wrist rest is connected. It’s not our favorite gaming keyboard for typing on (that’s up next), but the typing experience is decent: the Razer Green switches are tactile, clicky, and loud, and the case has two layers of dampening foam. 

The main downside to this keyboard is its size: it will take up a large portion of your desk, especially if you use it with the detachable wrist rest. At $230, it’s also pretty pricey — those extra buttons and dials are only worth it if you actually end up using them. 

Read: Razer BlackWidow V4 Pro Review 

Best Budget Gaming Keyboard

Patriot Memory is better known for its RAM and storage than its peripherals, but the Viper V765 provides a key feel that’s second to none — in an attractive, affordable package. It’s been a while, but the Patriot Viper V765 is still a good budget option with plenty of functionality. It’s one of the rare keyboards to come with Kailh Box White switches, which actuate faster than regular Blue or Green switches thanks to their slightly reduced travel (3.6mm versus 4mm). The keys are the most responsive we’ve tested and make one of the most pleasant click sounds you’ll find anywhere.

If Kailh Box White switches aren’t enough to convince you the Viper V765 is worth buying, it’s got plenty more going for it. This full-sized keyboard has an aluminum surface with tapered edges that give it a futuristic look. The vibrant RGB keys offer dozens of different light combinations — and you don’t even need to install the software to access most of the RGB effects. The keyboard also features media keys, which are more often seen on pricier boards, as well as IP56 water and dust resistance. 

The Viper V765 currently sells for $90,  cheaper than most competitors with similar features. We wish it had taller flip-out feet and a slightly better app, but the V765 is still our favorite budget-friendly gaming keyboard.

Read: Patriot Memory Viper V765 review

Best RGB in a Gaming Keyboard

The Roccat Vulcan II Max is designed to dazzle you: This full-size wired keyboard not only features bright, per-key RGB lighting, which is on maximum display thanks to the keyboard’s flat, shallow keycaps; 24 of its switches also have dual-LEDs. The Vulcan II Max comes with a detachable silicone wrist rest that serves as a conduit for the keyboard’s lighting (it’s beautiful), and features dedicated media keys and a clickable volume knob. 

This is an attractive, unique-looking keyboard even without lighting — it has a slim, lightweight chassis with an aluminum alloy top plate, and flat, chiclet-y keycaps that leave the switches entirely exposed. The keyboard has 24 pre-programmed multi-function “smart” dual-LED keys, which light up to indicate secondary keybinds and can also change colors to display real-time info, such as headset or mouse battery life. The keyboard features Roccat’s Aimo lighting experience, which is an “intelligent” lighting system that interacts with, and reacts to, the user (and adapts over time). It’s definitely a dazzling alternative to the typical spectrum cycling, and it’s even more dazzling when paired with the Vulcan II Max’s translucent wrist rest. 

If you’re looking for a hypnotizing light show, look no further. The Vulcan II Max comes in both black and white colorways, with Roccat’s Titan II optical switches in red (linear) or brown (tactile). If you’re looking to save desk space, there’s also the 65 percent Vulcan II Mini.

Read: Roccat Vulcan II Max Review  

Best Wireless Gaming Keyboard

If you want the same experience as your typical full-sized mechanical gaming keyboard but without wires, the Razer BlackWidow V3 Pro is the best wireless gaming keyboard we’ve tested. While many wireless keyboards are built for traveling, the BlackWidow V3 Pro maintains all the size, bulk, and functionality of a full-sized mechanical keyboard — including a premium tactile volume dial, media keys, an aluminum top plate, and doubleshot ABS plastic keycaps.

Razer doesn’t skimp on the BlackWidow V3 Pro: It’s packed with up to four profiles of onboard memory and vibrant per-key RGB lighting. However, RGB doesn’t stick to profiles — advanced RGB fanatics will have to rely on software for pairing RGB, including custom effects, to profiles.

The BlackWidow V3 Pro connects to your computer via 2.4 GHz dongle (which we had no issues with during testing) or Bluetooth (up to three computers simultaneously). With RGB at max settings and no dimming effects, our test unit averaged about 14 hours of battery life — but you’ll get up to 25 hours with RGB using power saving settings. You’ll want to plug in before the battery hits 3%, as power saving settings kick in at that point and the keyboard starts acting a little wonky. 

If you’re looking for a more portable full-sized wireless gaming keyboard, the Logitech G915 Lightspeed is fantastic and reliable with low-profile switches that some will enjoy for gaming and others will enjoy for travel. But for a full-sized mechanical gaming keyboard experience, the BlackWidow V3 Pro is the premium choice. 

For more cable-free recommendations, visit our Best Wireless Keyboards page. 

Read: Razer BlackWidow V3 Pro review  

Best TKL Gaming Keyboard

SteelSeries Apex Pro TKL Wireless gives you back some desk space as well as wireless freedom, thanks to built-in dual wireless (low-latency 2.4GHz and Bluetooth 5.0). The 2022 redesign of the Apex Pro TKL features a simple, elegant chassis with an aluminum alloy top plate, double-shot PBT keycaps, and SteelSeries newest OmniPoint 2.0 linear optical switches.

Optical switches technically have mechanical parts, but they’re not the same as mechanical switches — optical switches are actuated via light, while mechanical switches are actuated via physical force. Although this makes for a less-than-ideal typing experience, it’s good news for gamers: Optical switches are speedier (and less prone to failure) than their mechanical counterparts. The Apex Pro TKL Wireless’ OmniPoint 2.0 switches offer both adjustable actuation (37 levels) and dual-action actuation, which lets you assign two actions to one key (at different actuation levels).

The main downside of the Apex Pro TKL Wireless is its $250 asking price (the keyboard also comes in a wired version, for $180), which makes it one of the priciest gaming keyboards on this list. While the Apex Pro TKL Wireless comes with plenty of premium features and accessories — including a customizable OLED smart screen and a detachable magnetic wrist rest with a soft-touch finish — these might not be enough to justify its expense. If you’d like something a little less customizable (and you don’t mind being tethered), SteelSeries’ Apex 9 TKL has an almost-identical form factor with hot-swappable optical switches — and will only set you back $140. 

Read: SteelSeries Apex Pro TKL Wireless Review 

Best Gaming Keyboard for Typing

Most gaming keyboards are designed for gaming, not typing. That doesn’t mean they’re bad for typing — they’re usually still pretty good (better than a non-mechanical keyboard, anyway), but they focus on gaming first and prioritize features like flashy RGB over tactile feel and sound. The Asus ROG Azoth, however, is one of the best keyboards for both gaming and typing that we’ve ever used — and we’re pretty impressed that it comes from a mainstream gaming company and not a boutique keyboard supplier. 

The ROG Azoth is a gasket mount mechanical keyboard with a 75 percent form factor (that’s slightly smaller than a TKL). It’s wireless, with both 2.4GHz wireless and Bluetooth connectivity, and it features doubleshot PBT keycaps, a metal top plate, and three layers of sound dampening foam inside its plastic chassis (plastic so as not to interfere with the wireless signal). It comes with Asus NX switches in Red (linear), Brown (tactile), or Blue (clicky), and Asus packages it with a full accessory kit, which includes keycap and switch pullers, extra switches, and a DIY lube station so you can have ultimate control of the keyboard’s sound and feel. 

It’s an excellent keyboard for typing, but it’s also great for gaming — thanks to its reliable, low-latency 2.4GHz wireless connection, snappy responsiveness, and incredible battery life (up to 2,000 hours over 2.4GHz, with both lighting and the OLED screen turned off). It also offers decent customization via a 2-inch OLED screen that can be programmed to display everything from a random GIF to current system info.

Read: Asus ROG Azoth Review 

Best Low Profile Gaming Keyboard

If you want the speed and slimness afforded by low-profile mechanical switches, the Logitech G915 Lightspeed is the best gaming keyboard you can buy. This is a premium gaming experience through and through. That means a row of G macro keys for leveraging during battle, dedicated media controls, including an epic volume wheel roller, and even multiple connectivity options. You can go tried-and-true wired, use Logitech’s reliable dongle connection, or store the dongle in the keyboard’s built-in compartment and use Bluetooth. This makes connecting to multiple devices a little easier too. 

The G915 Lightspeed has earned a popular reputation but comes at a price. Despite its $250 MSRP, there’s no wrist rest, USB passthrough or premium keycaps resistant to smudging. As mentioned in our Logitech G915 TKL review, the smaller version of this keyboard is more affordable but foregoes even more luxuries, most noticeably G keys. 

But in addition to being a top-notch gaming peripheral, the G915 Lightspeed (and G915 TKL) offers a surprisingly good typing experience for a low-profile keyboard. We’ve tested it with Logitech’s tactile low-profile switches for days and haven’t felt much extra exhaustion or like we were typing through sand, like we’ve suffered through on other low-profile keyboards. 

Read: Logitech G915 Lightspeed review 

Best Mini Gaming Keyboard

Ultra-compact keyboards aren’t for everyone, but if you have limited desk space or you’re looking for a gaming keyboard that’s travel-friendly, a 60 percent keyboard can be a godsend. You’ll have to give up a lot — number pad, navigation keys, function keys, and even arrow keys — but you’ll barely notice these sacrifices with the right keyboard.

SteelSeries’ Apex Pro Mini Wireless (also comes in a wired version) is the right keyboard. 

The Apex Pro Mini Wireless is a sleek, sturdily-built 60 percent keyboard featuring SteelSeries OmniPoint 2.0 switches, which feature both adjustable actuation and dual actuation (much like the Razer Huntsman Mini Analog). Because it’s so small, the Apex Pro Mini Wireless has a slew of default secondary keybinds, which are printed on the sides of the keycaps and activated using SteelSeries SS function key. These secondary keybinds are so intuitive (and can also be remapped, if you want) that you won’t feel limited by the 60 percent layout. The keyboard is highly customizable: Almost every key can be remapped (first and secondary keybinds), have its actuation point adjusted, and have its RGB set – individually. 

If anything, the Apex Pro Mini Wireless might have too much going for it — maybe you don’t need an ultra-compact keyboard with every single feature shoved into it (especially considering its premium price point). If you don’t need all of these features and you’d like to spend a little less, try the SteelSeries Apex 9 series, which comes has adjustable actuation, hot-swappable switches, and comes in both Mini and TKL layouts.

Read: SteelSeries Apex Pro Mini review 

Quick Shopping Tips

• Mechanical or bust? Most gamers probably aren’t even considering a non-mechanical keyboard — for good reason. Only mechanical keyboard switches offer the tactile feedback, precision, and accuracy most gamers need, and membrane switches feel gummy and unresponsive by comparison. Recently we’ve been seeing more and more optical keyboards, which are technically mechanical, but use light — not physical force — to actuate. Optical switches still feel a little less satisfying to type on, but they actuate faster (nothing is faster than light, after all) and last longer than fully-mechanical switches.
• RGB or not? All RGB lighting is not equal. Per-key RGB lets you program each key’s color individually, while zone lighting limits your customization to just a few areas of the board. You can save a little money by getting a keyboard with a single-color backlight, but you’ll miss out on a spectacular light show. You can also forego lighting altogether, so long as you’re a touch-typist (or gaming in a well-lit area).
• Full-size, tenkeyless, or smaller? Tenkeyless boards drop the numpad, 65 percent boards eliminate navigation keys, and 60 percent boards also cut the arrow keys. Some users, such as MMO players, want every possible key (and more), while others prefer a smaller keyboard to clear up desk space.
• Pick Your Switch The best gaming keyboards use a number of different mechanical switch types that determine the feel and sound of each key press. The type of switch you choose depends on your personal preferences for typing and gaming. For more on the ins and outs of mechanical switches, check out our guide on mechanical switch spec ranges.

☑ Clicky Tactile: Blue, Green, White
☑ Quiet Tactile: Brown, Clear
☑ Linear (quiet and go straight down): Red, Silver

If you don’t want to commit, a hot-swappable keyboard will let you swap out switches easily.

Savings on the Best Gaming Keyboards

Whether you’re shopping for one of the best gaming keyboards we listed above or a different model, you may find some discounts by checking out our list of Corsair coupon codes, Newegg promo codes or Razer coupon codes.

The Raspberry Pi fans at Tom’s Hardware love a Cyberdeck build and the pinnacle of this project has yet to be reached, despite the best efforts of makers like  Brian Corteil. Fellow maker Mike Horne (co-founder of Pi Wars) has decided to make their own Cyberdeck using off the shelf components and an old aluminum case.

Bill of Materials (BoM)

• Raspberry Pi Zero 2 W
• 5-Inch HDMI Display (800 x 480 resolution)
• Pimoroni Enviro+ HAT
• Wireless Keyboard and trackpad
• LM2596S DC to DC Buck Converter (with seven segment display)
• 2200mAh LiPo battery
• Aluminum case

We’ve been following Horne’s progress since July and we just had to write about it. The Cyberdeck is currently a work in progress and so things are liable to change. Powering the project is a Raspberry Pi Zero 2 W, yes they do exist, but are hard to get hold of. The Pi Zero 2 W appears to be in an injection molded case (that doesn’t feature on our list of best cases!). On top of the Pi Zero 2 W is a Pimoroni Enviro+ HAT board. This hat provides a power input from the 2200 mAh LiPo battery (via the LM2596S buck converter) to power the Raspberry Pi Zero 2 W. It also provides a slew of environmental sensors.

Pimoroni Enviro+ Sensor List

• BME280 temperature, pressure, humidity sensor
• LTR-559 light and proximity sensor
• MICS6814 analog gas sensor
• ADS1015 analog to digital converter (ADC)
• MEMS microphone
• 0.96-inch color LCD
• Connector for particulate matter (PM) sensor
• Pimoroni breakout-compatible pin header (Stemma QT / QwST / Qwiic)

Horne isn’t too happy with the keyboard right now and is awaiting a mini Bluetooth keyboard to reduce the amount of cables squeezed into the case. This keyboard measures around 150 x 80 x 15 mm and also has a handy laser pointer for when we need to get corporate! Horne’s original idea involved a folding Bluetooth keyboard / trackpad combo that was a fraction too wide for the aluminum case.

As Horne mentions, the cabling inside the case is a mess. But it affords us a glimpse into the thinking of the project. We can see the HDMI and USB connection for the HDMI screen. We think that the USB connection provides a touch interface as well as power (we have a similar unit). There is another USB port broken out for the keyboard, and there appears to be a push button (top right of the case) and a switch. The switch must be for turning off the connection between the battery and the LM2596, and the push button for soft powering the Pi on / off. 

The aluminum case appears to be Innox branded and could be the Varicase 1 (which is currently selling for $3!). Horne has shoehorned the parts into a small case and was looking for ways to fix the HDMI display to the top lid. After seeking the knowledge of the Pi community (hot glue, epoxy etc), Horne was forced to break out the drill and mount the screen.

We’re vicariously enjoying Horne’s adventure as they work out the many problems of a build as complex as this. Keep an eye on Horne’s Twitter / X account for updates on this interesting project. 

Apple’s M3-series system-on-chips is expected to get a major performance upgrade compared to predecessors since they are projected to be made on TSMC’s N3 (3 nm-class) fabrication processor and use all-new CPU and GPU microarchitectures. Indeed, the highest-end M3 Ultra will feature 32 CPU cores and an 80-cluster GPU, but the entry-level M3 will retain eight cores, a Bloomberg report suggests.

M3: Up to Eight Cores

Apple’s vanilla M1 and M2 SoCs are used for Mac Mini, MacBook Air, MacBook Pro 13, and iMac systems, which are quite popular. The company’s upcoming M3 ix expected to retain eight general-purpose cores (four high-performance and four energy-efficient cores) and an integrated GPU with up to 10 clusters. The first systems on their base are expected to hit the market earlier this year. 

Apple’s M3 will reportedly be the first SoCs from the company’s third-generation PC processors and will also be the developers N3 chip designed for desktops and laptops. Apparently, the company decided not to increase core count in this SoC compared to M2, perhaps because it wanted to ensure the lowest possible power consumption or was sure that architectural advantages coupled with higher clocks will ensure a noticeable performance boost, or just wanted to maximize yields and keep costs down.

M3 Pro: Up to 14 Cores

The situation will get substantially better with M3 Pro that is projected to get 14 general-purpose cores (eight performance cores, six efficiency cores) in its top-end configuration, but its range-topping GPU will get 20 clusters, up from 19 clusters in M2 Pro. The M3 Pro in its maximum configuration will be more powerful than M2 Max in general-purpose workloads, but the latter will still have an edge in graphics applications.

Apple’s M3 Pro and M3 Max-based machine will likely hit the market sometimes in 2024.

M3 Max: Up to 16 cores

When it comes to M3 Max SoC, it is rumored to get 16 general purpose cores (12 performance cores and four energy-efficient cores) as well as up to 40 GPU clusters. Getting four additional high-performance cores is a big deal and will certainly bring substantial benefits to demanding software that M3 Max is architected to run. These will likely end up in high end MacBook Pros, the Mac Studio and Mac Pro.

M3 Ultra Up to 32 Cores

Meanwhile, the M3 Ultra system-in-package consisting of two M3 Max chips will therefore get 32 CPU cores and up to 80 GPU clusters. While 32 cores sounds quite massive when we talk about desktops, workstation-grade processors from AMD and Intel already have 56 – 64 cores and it remains to be seen what they are going to offer when Apple’s M3 Ultra-based Mac Studio or Mac Pro systems are available in the second half of 2024. 

While Bloomberg’s Mark Gurman tends to be accurate with his reports about future Apple’s products, he is still an unofficial source and information from him should be taken with a grain of salt.

An urgent pull request was sent out this weekend,to extend the Linux 6.5 kernel’s AMD Zenbleed vulnerability mitigation to Valve’s Steam Deck. It turns out that when the kernel was patched last month, mitigations weren’t implemented for Valve’s custom AMD SoC, sometimes known as Aerith, Phoronix first reported.

AMD’s Zen 2 product stack is pretty complicated, so it is hard to be critical of the Linux dev team. Zen 2 cores debuted with the well-known Ryzen 3000 chips, but the Zen 2 architecture can be also found in some Ryzen 4000 and 5000 processors, as well as Epyc server chips, the Sony PlayStation 5, in the Microsoft Xbox Series X/S consoles, and Valve’s Steam Deck.

It has been three weeks since news of the ‘Zenbleed’ vulnerability broke. The threat to AMD Zen 2 processor users seemed to be relatively severe. Zenbleed was shown to open up simple remote attack vectors, via JavaScript embedded in a webpage, for example. After successfully exploiting the Zenbleed vulnerability, an attacker could potentially learn your encryption keys or web logins.

If you care to head on over and look at the notes of the pull request you will see that a developer notes that “the AMD Custom APU 0405 found on Steam Deck was not listed, although it is clearly affected by the vulnerability.” The dev then goes on to request the Steam Deck’s CPU is added to the Zenbleed list “in order to unconditionally enable the fallback fix until a proper microcode update is available.”

This last point is important to note. AMD hasn’t provided any of its official patches for Zenbleed as yet, except to the EPYC 7002 ‘Rome’ processors. Everyone else will have to wait until November / December this year. The chipmaker said that it is not aware of any Zenbleed exploits outside of the research environment. Some assume that microcode updates will be able to fix the Zenbleed vulnerability with little or fewer performance penalties than a software patch would, but that remains to be seen.

A performance decrease isn’t ideal for a gaming machine, but with access to web browsers from the Steam side and a full desktop on the KDE Plasma desktop side, the Steam Deck is a full-on Linux computer.

It seems like the favorite processor families of PC DIYers have been plagued with newly uncovered bugs recently. In addition to Zenbleed, last week we reported on both (Intel) Downfall and (AMD) Inception.

A Chinese firm called Erying Technology has released its first compact motherboards offering a choice of Intel Raptor Lake Mobile chips pre-installed, VideoCardz reports. It calls this PC DIY product category ‘Mobile on Desktop’, or MoDT. The new choices open to Taobao customers in China are Erying Micro-ATX form factor motherboards with either an Intel Core i5-13420H or i7-13620H pre-installed. Prices of these foundational system components start at 1299 Chinese Yuan ($180).

Once you buy the Erying motherboard, you will be stuck with your CPU, as it is soldered in place, being a laptop BGA package CPU. Other than that, these motherboards are almost as good as their thoroughbred desktop counterparts for expansion and upgrades.

Reports suggest that the chipset that is used by Erying for these refreshed MoDT components is the B760M. However, Intel’s mobile platforms have fewer PCIe lanes. There could be a little impact on a powerful gaming system based upon these boards, with the primary PCIe Gen 4 slot offering just 8 lanes for a graphics card. Such a limit isn’t thought to have serious implications in 2023, though. It is estimated that a GeForce RTX 4090, for example, will only lose low-single-digit percentage performance due to this restriction.

Beyond that primary PCIe slot limitation, the boards offer a PCIe Gen 4 x4 slot, plus two PCIe Gen4 x4 M.2 NVMe slots for your SSDs. To help you get a better overall grasp of what is on offer, check out the specs and feature table below.

It is interesting to note that RPL mobile chips have better iGPUs than their desktop counterparts, which are limited to 32 EUs. However, we think that these MoDT products will be most frequently selected by those wishing to equip a discrete GPU – otherwise you could use one of many extremely compact alternatives.

If the above sounds like the foundation for a compact system that would work for you, these MoDT platforms aren’t prohibitively expensive, assuming you can get your hands on them. You are limited with regard to CPU upgrades, but if you are gaming on a modern system the CPU is rarely the bottleneck.

The Chinese retail listings show the Raptor Lake Core i5 Mobile version is 1299 Chinese Yuan ($180), and the Raptor Lake Core i7 Mobile version of the system board is 1699 Chinese Yuan ($234). Buying these products in the west might be a little tricky, but we shall have to wait and see how widely they get distributed. Previous gen Erying MoDTs are available on AliExpress, for example.