Two new Geekbench listings for Nvidia's upcoming RTX Spark laptop superchip have surfaced, confirming a second, cut-down SKU exists with 18 CPU cores, at least in the testing phases. Both parts score remarkably close to each other, posting impressive multi-core numbers, but their single-core performance isn't special. Still, these are preliminary results on unreleased hardware, so take these numbers with a grain of salt.

The model name for the 20-core (10+10) SKU was shown as "OEMQAJ 766_MIS Product Name DV" with its clocks reported at 4.0 GHz. This chip scored 2,570 points in the single-core test and 23,126 points in the multi-core test. The second listing showed up as "OEMQAJ OEMQAJ Product" with this SKU's 18 cores divided across 10+8 core clusters, clocked at 3.9 GHz. It scored 2,541 points in the single-core test, so within 1% of the 20-core part, while the multi-core was 21,776 points, which is about 6% faster than the full-fat chip.

When you compare these numbers to other Windows-running SoCs on the Geekbench database, these multi-core numbers are among the best. The Snapdragon X2 Elite Extreme currently holds the record for the highest multi-core score with 25,075 points, so both SKUs of the RTX Spark lose to it. But when we switch to the AMD AI Max+ 395, which scores 20,609 on average in the same test, even the 18-core Nvidia chip is 5.6% faster here, and the 20-core variant is about 12% faster.

Compared to Apple's lineup, the 20-core RTX Spark is 3.3% faster than the 14-core variant of the M3 Max, which scores 22,385 points in the multi-core bench. Considering that this is pre-release hardware with unoptimized drivers, we can expect it to even beat the maxed-out M4 Pro and the base M4 Max (which hover around 25,000 points) when it officially launches. Moreover, the 18-core SKU's 21,776 points are equal to, if not more than, the base M4 Pro, which also nets around 21,700 points in the multi-core test.

That being said, the swan song falls apart when we bring up the single-core numbers. Every Apple Silicon chip from the base M3 onwards beats the 2,570 points the 20-core RTX Spark achieved in this run. Only the M2 Max with its 2,477 points is bested by the Nvidia offering, and that too by just 3.75%. Apple's current-gen base M5 is a whopping 41% faster here. On the Windows side, the Snapdragon X2 Elite and X2 Elite Extreme score roughly 3,000 points, but AMD and Intel's top-end models take a beating.

The Ryzen AI Max+ 395 only achieves 2,429 points in the single-core test, making the 20-core RTX Spark 5.8% faster, and the 18-core variant 4.6% faster. The same goes for the Ryzen AI 9 HX 370 because its 2,432 points are within the margin of error of the Strix Halo chip. Intel's highest single-core score is achieved by the Core Ultra 9 275HX at 2,467 points, making it 3% slower than the 18-core RTX Spark and 4.2% slower than the 20-core SKU. However, Intel's actual fastest chip, the Core Ultra X9 388H, scores 2,767 points, making it about 7% faster than even the 20-core RTX Spark.

All in all, this is a much better showing for Nvidia's upcoming laptop Superchip than previous leaks. It's clear that driver optimization is slowly unlocking the full potential of the silicon, and by the time RTX Spark is in the hands of reviewers, we expect it to compete fiercely with every Windows machine. Nvidia is working hard to ensure a proper launch with a commitment to making DRM and anti-cheat work on day one.

From the outside, the Phantom Drive, created by Rootkit Labs, appears like a boring old USB. It has that classic, cheap plastic-feeling look to it that would never make you second-guess its intention; when you plug it in, it shows up as just an 8GB drive. However, under the hood, the Phantomdrive is powered by a microSD card that only unlocks itself when you specifically create a new text file, after which it mounts the secret partition, which comes with AES-256 encryption. It's made for situations and places where you might be forced to reveal the contents of a storage device.

The Phantomdrive is built using the CH569 microcontroller that has a bunch of different hardware blocks — this project uses the USB3, SD/eMMC, and the AES block. There are no NAND chips directly onboard the custom PCB because of the ongoing component crisis. Instead, the Phantomdrive relies on a simple microSD card (that you provide; it doesn't come with one) for both its 8GB decoy partition and the hidden partition. The creator does say they'll make a version with eMMC modules once prices come down.

Of course, if someone were to open up the drive, they'd be able to physically rip the SD card out. However, it still wouldn't be a dead giveaway since many cheap USBs are powered by microSD cards, and SD card adapters are also a thing. Anyhow, the finished drive looks pretty simple from the inside; it comprises a USB port, two buck supplies, a button for firmware update, the microSD card, UART test points, and some support components. The custom firmware now comes into play.

The name of the game is interception. The firmware doesn't know it's being used to handle file transfers; it just reads raw packets of data being sent over the USB bus as is. It's coded to always look for a "password" string, so as soon as you create the unlock.txt file with "password:xyz" inside, it detects the "xyz" bit and copies it over to SRAM, while overwriting the raw data with a bunch of zeroes before it's written to the SD card. The operating system thinks it's created the new file, but the firmware intercepted it midway and sent essentially nothing to the physical storage cells.

The firmware uses the unique internal hardware ID of the microcontroller as the salt, combines it with the password you just typed, and runs the pair through PBKDF2-HMAC-SHA-256 hashes 100,000 times (you can also select 600,000 times) to create a key. Each run induces an intentional 2-3 second delay to deter brute-force attacks that would try to crack the password. The key is used to unlock and lock the hidden partition, and since it's tied to the silicon itself, putting the SD card into another Phantom Drive will not unlock it.

Even if you try to access it directly through a microSD card slot, it will stay hidden, and the 8GB decoy will actually appear like a broken partition. The creator has also provided the option to recompile the firmware on your own to spoof specific vendor IDs, so that the USB appears like it's from different brands. You can make the Phantomdrive pretend to be a cheap 16GB Kingston Datatraveler at the hardware protocol level, at which point no scanning software will flag it as anything more than a cheap USB.

There are Linux-based C tools you can use in case of emergency, however, to recover your data, such as physical damage to the board. The transfer speeds are also very limited because the project is built over USB 2 — 20 MB/s reads and 9 MB/s writes in AES-CTR mode. The speeds fall to 10 MB/s reads and 6 MB/s writes in AES-XTS, but at least you have the freedom to choose between the two according to your needs.

You can buy the Phantomdrive from Rootkit Labs for $50, but you can also build your own, since this is an open-source project. The prebuilt device comes with two extra shells because the pins wear off and break quickly. Otherwise, the GitHub repo provides all the resources, including the PCB layout files and schematics, the BOM for every part, and the custom firmware, alongside its related tools.

The creator of Phantomdrive, Ryan Walker, explains the goal behind the project as both utilitarian and political. It's built using entirely open-source software and with inexpensive, easily accessible components. They wanted to create something that could "slip past authoritarian government representatives, corrupt police, and anyone else that doesn't respect basis encryption rights."

Valve's second-gen Steam Controller has been a major hit with the community thanks to the company's open approach to its hardware. Modders have already used the haptic motors inside to move it across a desk and charge automatically, while others have built their own open-source pucks. Now, a new project from Pixel1011, called "Steam Haptics Player," can turn these motors into a stereo speaker capable of playing music both wired and wirelessly.

For context, the Steam Controller doesn't have a speaker; there are two linear actuators under each trackpad that vibrate to provide haptic feedback. Previous projects have leveraged these components to make the controller vibrate in a certain way. Since vibrations are the basis of sound, Valve itself built a secret easter egg into the controller that randomly makes it do a Wilhelm scream when dropped from certain heights.

Pixel 1011 discovered that the haptic motors inside can be targeted with low-level HID (Human Interface Device) feature commands using HIDAPI. As such, Steam Haptics Player takes an audio file, downsamples it through FFmpeg, splits it into chunks of a PCM stream, and sends those packets directly to the controller, bypassing high-level software like Steam Input. The controller's firmware then feeds this raw data to the motors, at which point they vibrate to produce sound.

Because of the DIY nature of this process, bandwidth is an issue, especially wirelessly, since the puck isn't designed to carry audio data over its radio signal. This means the audio has to be compressed, and there's a limit to the overall quality the motors can transmit — 8-bit 8kHz (μ-law) over the puck and 16-bit 8kHz PCM over a wire.

Still, the 16-bit audio is enough to produce a surprisingly loud and balanced result that can deceive you as if it's coming from a speaker. The video at the start features Pixel1011 playing the song "Want You Gone" from the end credits of Portal 2, and you can hear the quality for yourself. There's a genuine low-end that you would never expect from a jerry-rigged setup like this, even if it's sitting on top of a towel on a flat surface to give it its best chance.

The wireless audio quality is rough in comparison; it crackles and pops and even cuts off in between because of the bandwidth limitations. Pixel1011 is using 8-bit mu-law companding here, which actually offers dynamic range similar to 14-bit audio while keeping the byte footprint of an 8-bit stream. Even through this compression, the puck is struggling, but the fact that this is even possible is still very impressive, and perhaps there's room for improvement down the line.

You can try Steam Haptics Player yourself on your own second-gen Steam Controller with pretty much any audio file in any format. Just download the archive from the GitHub repo, install FFmpeg (for transcoding, if you don't have it already), and run the program via a PowerShell window in the original directory. Detailed steps are on the linked GitHub page, but we don't know how this affects the controller's battery life. Just don't go all out with your Spotify playlist on these haptic motors.

AMD is expected to unveil its next-gen Zen 6 platform at CES 2026 with Ryzen 10000 series processors. While we're looking forward to IPC and clock speed improvements, it seems like more subtle, under-the-hood changes could end up upgrading the gaming experience in a big way. According to a tip received by Videocardz, AMD is implementing various per-core optimizations to ensure foreground tasks get priority over background applications. Each new feature is supposed to more smartly manage the power and thermal budgets at the silicon's disposal, so let's go over each of them.

First up, we have CPPC Performance Priority — CPPC stands for Collaborative Processor Performance Control, and it's responsible for communication between the silicon and the OS. It lets the firmware sitting in between decide the performance of each core separately. This feature has actually existed since the Ryzen 3000 series, but it doesn't work perfectly. Zen 6 CPUs are apparently supposed to apply a band-aid fix and improve their effectiveness.

Secondly, we have FloorPerf, which acts as a dynamic, targeted power delivery system. It sets a minimum clock speed for all cores that the silicon can clock down to in case of thermal throttling. For instance, imagine you're running a game in the foreground while Discord and Spotify sit in the background. The temperatures on your Zen 6 CPU rise to the point of throttling, but instead of reducing the speed of all cores, FloorPerf will target the cores running the background tasks first, in order to maintain the performance of your game.

We covered a recent fix like this for the Linux world as well, where a sudden spike in the background can cause the foreground task to choke. Linux has an issue with priority allocation across tasks, so that was a different situation, but similar logic applies here. On a Zen 6 CPU, instead of your game suddenly experiencing microstutters because of aggressive throttling, FloorPerf will try to limit whatever's running in the background first. The aforementioned CPPC Performance Priority should work hand in hand with this feature to maintain clocks on the important cores.

CPPC will get another boost in the form of HighestFreq, which will allow the OS to access more granular chip data to make better core management decisions. As the name suggests, it pertains to maintaining high clock speeds on the cores running a foreground task. The OS will be aware of exactly which core can boost the highest and maintain that speed the longest. This will allow it to assign, for instance, a game's main rendering thread to the fastest cores, while pushing background apps to more power-efficient ones.

In conjunction with this, Zen 6 is also seemingly getting per-core EPP boost. EPP stands for Energy Performance Preference, and it's supposed to fix core parking issues caused by a momentary bottleneck. For example, if the CPU is waiting for the GPU to render the next frame, the clock speeds for the active cores will drop in that moment and struggle to ramp back up in time, forcing all cores to boost when the frame is ready. EPP boost is supposed to identify the active cores and individually keep them in high-performance mode to eliminate any wait times.

Lastly, Zen 6 is reportedly introducing PQOS Global Bandwidth Enforcement and an updated IBS Memory Profiler. Both of these target memory bandwidth with the same goal of maximum stability as everything else discussed so far. The former is responsible for allocating RAM to background tasks, ensuring their usage stays under the limit for foreground priority. The latter will limit L3 cache access to background tasks when it detects that those tasks are slowing down the foreground application, which should help with frametime consistency in games.

As you can tell, all of these optimizations are supposed to work together and harmonize into a more efficient processor at the end — one that maximizes the silicon's potential as much as possible. Of course, none of this is confirmed and there's a lot that could be gated behind product segmentation. We could see some features only on high-end parts, while some are limited to mobile; there's no telling at the moment, but it's exciting stuff, nonetheless. Seems like a battle for the ages is brewing between Nova Lake and Zen 6 next year.

Buying a new GPU is always a joyous occasion, but the celebration was cut short for an unfortunate customer in China. Right after buying an AX Gaming RTX 5060 Ti, they got into a serious car crash that ended up totaling the car — and the card along with it. Fortunately, the driver was safe, and somehow the GPU survived, too, with no major electronic damage. Despite being bent in half, the card ended up only requiring minor repair work and was brought back to life quickly by Brother Zhang.

Visually, the card didn't look salvageable, and midrange GPUs like these often don't make for feasible repair jobs either because of the high labor cost. The backplate and cooler were completely destroyed, so Brother Zhang separated them from the actual PCB. Luckily, this variant of the 5060 Ti used a very short PCB that just barely touched the point of curvature, so the internals were mostly fine.

The technician checked all the power rails for a short circuit but saw no abnormal resistance, which confirmed the core and memory weren't dead. He then proceeded to quickly test the GPU with its bare PCB, and it booted successfully. Brother Zhang devised a makeshift cooler that consisted of the bent fin stack straightened back to the best of his abilities, and two 120mm fans strapped on top via zip ties. This seemed enough to keep the card cool.

However, as soon as he tried to install drivers, the PC would black screen and hard crash, showing consistent signs of instability. At this point, Brother Zhang reopened the card and inspected the PCB. He inspected the flex point and decided to resolder the memory module closest to the edge of the board since it experienced the most torque. A simple removal and reapplication fixed the issue, and the card was detected correctly as a 5060 Ti with the full 8GB of VRAM.

Brother Zhang rechecked for any short circuits to ensure no solder bridges were created during the BGA work before moving on to benchmarking. In Furmark, the temps peaked and stabilized at 64°C with no visual artifacts. He then ran a full 3DMark loop to test the PCIe 5.0 x8 link speeds — the card passed with a high 98.9% stability score. It drew around 92% of its maximum power limit during the run, with core temps peaking at around 67°C. This 5060 Ti was officially a survivor.

The PC hardware industry is facing an unprecedented challenge at the moment: a component crisis. During times like these, if you wanted to build or upgrade your own PC, you'd be stuck with overpriced parts unless you find deals. Well, worry not, as we've spotted exactly that — a combo bundle featuring a Ryzen 5 5500 with 16GB of DDR4 RAM and a B550M motherboard for just $280.99 on Newegg right now.

• Check out this deal on Newegg

The Ryzen 5 5500 is not exactly the newest or fastest processor on the block, but it's plenty for gaming on a tight budget. It's a six-core CPU with 12 threads that can boost up to 4.2 GHz. It has 16MB of L3 cache and 2MB of L2 cache, along with a 65W TDP. Since it's based on the Zen 3 architecture, it uses the AM4 platform, which means more affordable motherboards and, more importantly, access to less-expensive DDR4 RAM.

The AM4 socket also gives you a direct upgrade path to the Ryzen 7 5800X3D down the line because the motherboard is strong enough to support it. You're getting a Gigabyte B550M Gaming X board with Wi-Fi 6 that costs $155 on its own. It has strong rear I/O thanks to its four USB 3.2 ports on the back, along with two USB 2.0 ports and Gigabit Ethernet for wired connectivity. Internal I/O includes two ARGB headers and a Type-C header for your case's front panel.

You get two M.2 slots running at PCIe 3.0 speeds, which can be a blessing in disguise since you don't have to look for the latest SSDs anymore, helping you save a bit of money. There are four RAM slots with support for up to 128GB of memory at DDR4-3200. Speaking of which, the final product in this bundle is a 16GB (2x8GB) kit of TeamGroup's T-Force Delta Force RGB RAM in black. It's running at DDR4-3200 with a CAS latency of CL16.

Even though DDR4 is a generation old now, it's not outdated. It's sufficient for gaming and even some professional workflows. You're only getting 16GB in this combo, but the four-slot motherboard leaves room for an easy upgrade to 32GB later when things normalize. Overall, this is a fantastic budget bundle to get you started whether you're building a new PC or coming from a really old platform.

When you purchase all these items separately, the total bill comes out to around $420, so you'll be saving 32% upfront by going with the bundle instead! For just $281 on Newegg, you'll get a few essential components knocked off the list without compromising on performance or breaking the bank, all the while leaving room for upgrades that make sense in the current state of the industry.

If you're looking for more savings, check out our Best PC Hardware deals for a range of products, or dive deeper into our specialized SSD and Storage Deals, Hard Drive Deals, Gaming Monitor Deals, Graphics Card Deals, gaming chair, or CPU Deals pages.

The state of PlayStation 5 emulation has been rapidly evolving, partially in response to Grand Theft Auto 6's release later this year. The latest milestone is the first 3D games being playable at very respectable framerates of around 30 frames per second.

The two emulators at the forefront of all this are KytyPS5 and SharpEmu, the former of which is intensely focused on expanding compatibility. KytyPS5 was already able to boot 3D titles and go until their loading screens, with at least two smaller games actually rendering graphics. Now, the same emulator can run two 3D games at around 30 FPS, though there are some glitches.

The first game is XIII, the 2020 remake of the 2003 classic, and it's "playable" at up to 30 FPS with real-time gameplay that accepts player input. We put playable in quotation marks because there are very obvious visual glitches that manifest in the environment not rendering correctly. The textures seem to be broken in multiple scenes, though they don't entirely kill the immersion. XIII is a relatively lightweight game but its cell-shaded aesthetic might be adding to emulation complexity.

The same person who tested XIII also tested Teenage Mutant Ninja Turtles: Wrath of the Mutants and managed to run it at just under 25 FPS with real-time gameplay. The game crashes on the first launch, but once initialized, it's stable and does not seem to have any bugs or glitches like XIII. The framerate looks visibly choppy but is undeniably impressive, on a technical level. Kyty can run other 3D PS5 games as well, but not nearly at this level of performance.

KytyPS5 is advancing at a faster pace than SharpEmu, and that's true for the most part, but not out of stagnation. The team behind SharpEmu is busy tackling a different challenge: accuracy and infrastructure setup rather than game-specific compatibility, per its GitHub repo. In other words, it means fixing low-level issues that are preventing games like Astro Bot from currently reaching the gameplay state.

Both emulators are now part of an unexpected emulation race. With GTA 6 just a few months away and GTA 5 already booting in Kyty, it clearly has the better support for now, and its demos are very promising. SharpEmu's work will undoubtedly help the emulation scene at large, along with RPCSX and other independent developers. The craziest part is that you can try all of this at home for free, and just by doing that you're already contributing to the scene.

AMD officially launched its cheapest RDNA 4 graphics card, the Radeon RX 9050, a few days ago, and we already have some performance numbers for it. Preliminary testing from Japanese outlet GDM puts the GPU up against its closest Nvidia rival, the RTX 5050, in a matchup that sees the AMD offering lose by a big margin. The RX 9050 is fine on its own, especially if you're limited to 1080p gaming, but in the head-to-head it leaves a lot to be desired based on its roughly 30% slower framerates averaged across various titles.

The test bench consisted of a Ryzen 7 9800X3D and specifically the ASRock Challenger variant of the 9050, alongside the Gigabyte Windforce OC variant of the RTX 5050. Starting with synthetic benchmarks, the Red Team's GPU is almost 20% slower in the 3DMark suite. The biggest difference lies in Port Royal, where the RTX 5050 scores 6,181 points but the RX 9050 only manages 4,531 points. Time Spy sees the smallest delta, where both cards are separated by about 500 points.

Moving onto gaming, the testing was conducted across multiple resolutions, so we've chosen to show the most representative results possible. When talking about esports titles such as Apex Legends and Rainbow Six Siege, the RX 9050 holds its ground well, especially as you scale up the resolution. At 4K Ultra+ settings, both GPUs actually match at 64 FPS. The biggest difference was 20%, reported at 1440p medium, where the RTX 5050 pushed 150 FPS while the RX 9050 could only manage 125 FPS.

The rest of the games show a more underwhelming result for the RX 9050, with DOOM: The Dark Ages acting as an outlier where both GPUs take turns destroying the other. The game has multi-frame gen support at up to 6x for the Nvidia card, while the FSR frame gen is limited to just 2x. Regardless, the RX 9050 wins at higher resolutions in all settings due. As mentioned, however, this is the only test where the 9050 seemed to beat the 5050.

The remaining titles show rather boring numbers where the RX 9050 loses every single time, no matter the resolution or preset. It comes the closest in Final Fantasy XIV at 1080p, where the internal benchmark separates the two GPUs by 23%, whereas the biggest difference lies in Onimusha, where the RTX 5050 is almost twice as fast. Across these games, on average, the RX 9050 is 31% slower at 1080p, 34% slower at 1440p, and 42% slower at 4K.

The RX 9050 scores a win in power consumption, drawing about 8W less when idling and about 35W less under heavy load compared to the RTX 5050. The card seems to run cool and quiet, and can be manufactured to be rather small, making it ideal for SFF builds. Overall, it's an okay experience when it comes to gaming, performing amicably in a vacuum, but disappointing against competition. Thankfully, the 4GB variant of the RX 9050 seems to be limited to OEMs. As always, these are preliminary testing results from one outlet, so take them with the customary grain of salt.

Mechanical pressure serves as a better system to design computers around than natural buoyancy, but a new experiment from der8bauer is attempting to challenge that notion. The legendary modder recently built a chimney to exploit the "stack effect," which uses temperature differences to catalyze natural convection to the point where it becomes effective again. As a result, the CPU in his test bench saw temperatures drop by 19 degrees Celsius without any fans.

The video above explains the concept of the stack effect pertaining to its foundation, natural convection, in great detail. Also known as the chimney effect, it relies on differences in indoor-to-outdoor air density, which are dictated by differences in temperature and humidity. The bottom of the chimney becomes a low-pressure zone, sucking in cold air that escapes through the low-pressure exit at the top.

In the experiment, der8auer uses a fog machine to show how a radiator can't "pull" air without fans attached to it; the hot air dissipating from the other side is forced to fall back on natural convection. The test bench was water-cooled and consisted of an AMD Ryzen 7 9800X3D, so he added a temperature probe to the fittings to monitor the water temp as well. The experiment then begins with the first 3D-printed chimney that's 10cm tall — it barely works, but temps already start to drop by 0.5 degrees.

At this point, the water was at 61°C and der8auer added another 20cm chimney atop the previous one, making the enclosure 30cm tall in total. This made the temps drop by another 5 degrees over a 30-minute stabilization period. The fog machine didn't really show any difference in the airflow, however. Now, it was time for the big boy: an 80cm tall chimney made up of 4x 20cm parts, bringing the total height to 110cm.

Right away, temperatures began dropping below 55°C before the modder could even set up his camera for the ultrawide framing this setup warranted. Just five minutes later, the water had hit 50°C, but the visual demonstration really sealed the deal. Turning on the fog machine underneath the radiator showed air being pulled through the chimney as if there were fans facilitating airflow on top.

The CPU temperatures had peaked at 90°C previously, but the "big tower" had now brought them down to just 71°C, as reported by HWiNFO. The experiment was a raging success. The video ends with der8auer actually tying the stack effect back to how we use fans in modern PCs to synthesize air pressure, since they achieve the same goal — a pressure differential — just more reasonably. So, unless you want to redefine what a tower PC means, sticking to regular, boring old fans is enough.