Graphics Cards - Page 136

NVIDIA has been slammed by a cyberattack that is causing all sorts of issues, with another one of them being GPU-related leaks to its next-gen GPU architectures... the next one, is Ada Lovelace.

The hacking group has infiltrated NVIDIA's servers, and now more information on Ada Lovelace is here. We have the full specs of NVIDIA's next-gen Ada Lovelace-based GeForce RTX 40 series GPUs including the AD102, AD103, AD104, AD106, and AD107 GPUs.

NVIDIA's next-gen Ada Lovelace GPU architecture is expected to be deployed later this year, with the flagship AD102 GPU with the full 18432 CUDA cores and 384-bit memory bus. Next down on the list we have the AD103 GPU which should have 10753 CUDA cores, and the AD104 GPU with 7680 CUDA cores. After that there's the AD106 and AD107 GPUs with 4608 CUDA cores, and 3072 CUDA cores respectively.

We might not ever see the next-gen graphics card released this year that actually uses PCIe 5.0 power connectors, but the PSU market is gearing up for the big launch of next-gen PCIe 5.0-ready GPUs.

Well, now the official specifications of the new PCI-Express 5.0 cable for next-gen GPUs has been leaked, with data approved by Intel who are part of defining the ATX specs. We now know there should be up to four different power settings, with the PCIe 5.0 power cable supporting 150W, 300W, 450W, and a huge 600W over a single 16-pin PCIe 5.0 power cable.

It looks like the binary configuration of the Sense0 and Sense1 sideband signals, but when both of the signals are grounded, the GPU will have up to 375W of power pumped into it, with a maximum of up to an insane 600W of sustained power.

The ongoing NVIDIA issues are surely causing massive meetings inside of Team Green, with the cyber attack now seeing the DLSS (Deep Learning Super Sampling) source code being leaked... and that's not the worst of it.

NVIDIA has kept its magic DLSS technology close to its chest, and while this is good for the company, its competitors -- AMD and Intel -- have their similar upscaling tech being open source. NVIDIA hasn't pushed into the world of open-source with DLSS, but hackers might actually force their hands.

DLSS beats the pants off of FSR (AMD's FidelityFX Super Resolution) and even more so with DLSS 2.2 and DLSS 2.3 if the games support it, because when they do... NVIDIA reigns supreme by a (very) long shot. But now that the DLSS source code is out in the wild, it could end in a ball of blames for NVIDIA as their secret sauce is out in the nude.

NVIDIA's next-gen Ada Lovelace GPU architecture could have a big surprise up its sleeve, with new information that continues to flow from the leaked cache of content that hackers gained access to when they cyberattacked NVIDIA.

Inside of some of that information are some files that refer to the "Ada" GPUs with 16MB of cache per 64-bit memory bus, up from the 512KB per 32-bit memory on Ampere GPUs. This is a huge change, so the fact that the flagship AD102 GPU has 96MB of cache means it'll have a whopping 90MB more L2 cache than the GA102 Ampere GPU.

AMD has up to 128MB of Infinity Cache on its RDNA 2-based Radeon RX 6000 series GPUs, with the likes of the Radeon RX 6800 XT and Radeon RX 6900 XT both rocking 128MB Infinity Cache. NVIDIA's next-gen AD102 GPU should power the GeForce RTX 4090, GeForce RTX 4080 Ti, and GeForce RTX 4080 graphics cards if the company keeps its naming system the same.

NVIDIA's systems have reportedly been compromised, with hackers now threatening to leak out confidential NVIDIA data, information, and even crypto gimping unlocks.

Things continue to go from bad to worse, where now the hackers are threatening to leak out information -- and unlocks to the LHR-based GeForce RTX 30 series GPUs. The LHR V2 bypass for NVIDIA's gimped GA102 + GA104 GPU, and now those unlocks are being sold.

This means that the hacking collective has found a solution to the algorithm that gimps crypto mining on the RTX 30 series GPUs that are LHR-gimped, should be un-gimped soon enough. NVIDIA has admitted it has been attacked now, but I guess we'll see how much deeper this goes in the coming hours, days, and weeks.

UL has just announced its next-gen 3DMark Speed Way benchmark, which will stress test GPUs with ray tracing, and other DirectX 12 Ultimate features.

The new 3DMark Speed Way benchmark sees UL teaming with Lenovo for branding, stressing out your GPU with ray tracing, and real-time global illumination to render realistic lighting and reflections. Microsoft's new DirectX 12 Ultimate features will be stressed, including mesh shaders, and variable rate shading (VRS) to optimize performance and visual quality.

Lenovo's partnership with UL sees the company's Legion gaming brand having product placements in the Speed Way benchmark, including Legion gaming products (gaming PC + monitors + laptop) as well as merchandise. As for the benchmark, 3DMark Speed Way will be launching on Steam later this year -- but we don't know if it'll be an add-on available for free -- or a separate DLC.

NVIDIA should hopefully be unveiling its next-gen Hopper GPU architecture at its upcoming GTC (GPU Technology Conference) in March 2022 -- but now we're hearing about its successor -- the next-next-gen Blackwell GPU architecture.

Where is this news coming from? Well, that would be from NVIDIA being recently hacked, where one of the files refers to "ARCH_BLACKWELL" which would see NVIDIA's next-gen GPU architecture named after David Harold Blackwell, an American statistician and mathematician specializing in statistics, game theory and information theory.

The first mentions of Blackwell turned up in July 2021, with leaker "Kopite7kimi" tweeting out the Blackwell GPU architecture tease. NVIDIA has only mentioned "Ampere Next" and "Ampere Next Next" but we should see those GPUs as Hopper and Ada Lovelace, but now we have Blackwell thrown into the next-gen GPU mix.

NVIDIA hasn't confirmed whether it was hacked or not, but it looks like there's a bit of a conspiracy theory building around what happened to Team Green. Supposedly, a South American hacking group called "LAPU$" are responsible for the cyberattack on NVIDIA.

LAPU$ reportedly attacked NVIDIA's internal servers through a ransomware attack, taking over 1TB of NVIDIA's precious data. NVIDIA hasn't confirmed this, obviously... but another tentacle of this conspiracy is that NVIDIA reportedly hacked the hackers, in an attempt to encrypt their 1TB of data (that wasn't stolen, right?).

NVIDIA aren't hackers and I'm sure they might have a team of really smart people, but they wouldn't be anon-level digital soldiers like LAPU$ or Anonymous. The group made a copy of the 1TB of NVIDIA data in a virtual machine environment meaning that NVIDIA's attempt to out hack the hack will fail.

The Pentagon has some harsh words for Generation Z -- those aged between 18 and 25 years old -- with the Defense Visual Information Distribution Service's explaining in a press release that this generation is literally not built as strong as the last.

Major Thibodeau, a clinical coordinator and chief of the medical readiness service line at Fort Leonard Wood in Missouri actually said: "The "Nintendo Generation" soldier skeleton is not toughened by activity prior to arrival, so some of them break more easily".

The weaker skeleton of Gen Z is because of the sedentary, you know... too much time sitting around like I am right now writing, or when you're playing games... and this isn't something that previous generations did. At least not as much, as there wasn't the allure of gaming surrounding them. This has led to more skeleton-related injuries as the "Nintendo Generation" bodies aren't used to the stress that soldiers are put through... even in basic training.

NVIDIA has just silently launched its new GA103-200 GPU-based GeForce RTX 3060 Ti, at least with ZOTAC China launching its new custom RTX 3060 Ti... meanwhile, the flagship GeForce RTX 3090 Ti is nowhere to be seen. Righty-o.

ZOTAC's new custom GeForce RTX 3060 Ti graphics cards come in three new flavors: Apocalypse, X-GAMING, and PGF which all pack the fresh new GA103-200 GPU. So what's the difference here? Well, ZOTAC has added an "X" to the end of the product name of these GPUs.

So for example the GeForce RTX 3060 Ti non-LHR (GA104-200) is joined by the RTX 3060 Ti LHR (GA104-202) which is why ZOTAC is bringing the "X" in. So we would then have:

NVIDIA has reportedly been slammed by a cyber attack, seeing parts of their systems taken offline including their mailing system. A spokesman for NVIDIA said: "We are investigating an incident. We don't have any additional information to share at this time".

The Telegraph reports that NVIDIA's internal systems were possibly compromised, forcing the company to take them offline -- mailing systems, developer tools could've been compromised -- so they were the first to be taken offline.

As for the cyber attack, of course, they're being connected to the ongoing conflict between Russia and Ukraine, with many countries around the world preparing for more cyber attacks. As for the cyberattacks, they're nothing new as they've always happened, but now it's hit a company closer to home in the gaming world. I don't like the speculation and rumor that Russia is behind it, it's like what's happening overseas gives everyone a boogeyman to blame (like usual) as we don't know who is responsible for the cyberattack on NVIDIA.

In a news post yesterday I reported about the "NVIDIA RTX LHR v2 Unlocker" tool that would reportedly un-gimp an LHR-based GeForce RTX 30 series GPU... but man, we could not have been more wrong.

The tool is actually made by a hacker and is actually filled with malware, and most likely connects to a remote server somewhere and the world and siphons data from your PC. I reported on it but didn't include a link to it, just the articles that had reported on it, so I won't include a link here.

In the original article, I stated: "The way this new tool works is that the software implements a modification extracted from BIOSes that are detected by mining operating systems like HiveOS. The modification is then added into the BIOS of cards that have limited hash rate performance, while the new Unlocker software modifies, flashes, and gives you a backup of your BIOS. It's safe, and reversible, says Sergey".

If you thought the 300W+ that your GPU uses now, or even the 450W+ of past dual-GPUs and even NVIDIA's maybe-it-might-release-one-day GeForce RTX 3090 Ti and its 450W power... but man, Ada Lovelace is power-hungry in the latest rumors.

NVIDIA's new Ada Lovelace GPU will debut later this year with the GeForce RTX 40 series graphics cards, with new rumors from industry leaker "Greymon55" that there are 3 x TDP numbers of AD102: 450W, 650W, and a monster 850W.

We should expect the 450W TDP with the GeForce RTX 4080, the 650W TDP for the GeForce RTX 4080 Ti, and the bonkers 850W TDP for the GeForce RTX 4090. THis is all rumor mill right now, so I've included some salt in the Amazon affiliate link just in case. You can never be sure that you don't have enough salt, especially with these rumors. But then again, Ampere is a power-hungry GPU monster... and it seems NVIDIA is really having to dial things to 11 for Ada Lovelace and the GeForce RTX 40 series GPUs.

NVIDIA's next-gen Ada Lovelace GPU architecture is in the headlines again, with the flagship GeForce RTX 4090 looking like it could chew through up to 850W of power.

In a new series of tweets, industry leaker "Greymon55" tweeted that NVIDIA's next-gen Ada Lovelace GPU would "be available in September". Greymon55 replied to his own tweet, adding: "I am not clear at the moment whether one model has three TGP ranges or whether it has three models but the TGP number of the AD102 is 450W-650W-850W, of course this is not the final specification and there may be some deviation".

NVIDIA's upcoming-but-who-the-hell-knows-when-it-will-launch GeForce RTX 3090 Ti is meant to ramp up power consumption considerably, with up to 450W at its disposal and up to 600W through the new 16-pin PCIe 5.0 power connector. But the next-gen Ada Lovelace GPU architecture and GeForce RTX 40 series looks to be a power-hungry, and with an expected doubling in performance over the RTX 3090, the power consumption numbers on the RTX 4090 might not matter at the end of the day.

I don't know why this is 'new' news but I'm still reporting on it anyway: Intel's new Arc Alchemist GPU family will have multiple performance tiers (that I reported on back in November 2021) in the form of the A300, A500, and A700 series GPUs.

The news is coming from SiSoftware which has a roundup of OpenCL benchmarks where we get a look at the higher-end A500 and A700 series Arc GPUs which should see the flagship Intel Arc A700 featuring the DG2-512 GPU with the full 512 Execution Units.

Intel Arc A500 should see the DG2-384 GPU, while the Arc A300 will find its way into our motherboards with the DG2-128 GPU. We should see 16GB of GDDR6 memory on the Arc A700, 12GB of GDDR6 memory on the Arc A500, and 6GB of GDDR6 memory on the Arc A300. Intel has already confirmed that its upcoming first-gen Arc Alchemist GPUs will be made on TSMC N6 process technology.

NVIDIA refreshed its Ampere cards last year with the LHR variant, otherwise known as Lite Hash Rate, gimped for crypto mining... but not anymore, well, it seems so at least.

Now we have Sergey releasing his new "NVIDIA RTX LHR v2 Unlocker" tool that will unlock the crypto mining performance on NVIDIA's GeForce RTX 30 series and Workstation RTX Ampere GPUs. There have been previous 'solutions' to un-gimping LHR cards, but this reportedly unlocks the "full mining performance" according to MyDrivers.

The way this new tool works is that the software implements a modification extracted from BIOSes that are detected by mining operating systems like HiveOS. The modification is then added into the BIOS of cards that have limited hash rate performance, while the new Unlocker software modifies, flashes, and gives you a backup of your BIOS. It's safe, and reversible, says Sergey.

Intel has scooped up yet another member of AMD's core Radeon GPU team, with AMD Senior Fellow and Lead Radeon Discrete GPU Architect of AMD Technologies, Rohit Verma, joining Team Blue.

Verma began his journey at AMD as a Fellow and Lead SoC Architect inside of the Semi-Custom Business Unit, where he called home for close to 5 years. In that role, he matured into a Senior Fellow and Lead Senior SoC Architect for Discrete GPUs at AMD for over 3 years.

But now that Verma is at Intel, he has become the new Lead Product Architect of Discrete GPU SoCs -- something he actually did, with Intel back in 1999-2013 in a 15-year stint. Intel's new CEO Pat Gelsinger has been turning things around for the company in record time, with his vision for 2022 pretty clear: "The brain drain, well that's changed. We now have brains coming back".

Intel might not have even launched its new Arc Alchemist GPU but it is confident in announcing "Project Endgame" which is shaping up to be a GeForce NOW-style cloud computing competitor.

The new Project Endgame was announced during a flurry of announcements and reveals at its recent Investors Meeting -- where we had many more details on the upcoming and future-gen Arc GPUs, 13/14/15th Gen Core CPUs, and so much more -- but Project Endgame is big for Intel, but is it too early?

Intel explains Project Endgame: "Project Endgame will enable users to access Intel Arc GPUs through a service for an always-accessible, low-latency computing experience. Project Endgame will be available later this year".

Intel's flagship Arc Alchemist GPU has been benchmarked once again, with its GPU boost clocks hitting 2.4GHz in a new Geekbench OpenCL test.

The new Intel Arc Alchemist desktop GPU was tested with a 2.4GHz boost clock, alongside an Intel Core i9-9600K processor (with 6 cores, and 6 threads). As for the GPU, it's rocking the DG2-512 GPU so we're looking at 512 Compute Units -- the flagship Arc Alchemist GPU.

In the Geekbench OpenCL test, the Intel Arc Alchemist with 512 EUs and 2.4GHz GPU boost clock, hits 85448 points which see it competing against NVIDIA's previous-gen Turing-based GeForce RTX 2070/2060 graphics cards.

GIGABYTE has formally announced its next-gen UD1000GM PCIe 5.0 power supply, which will support the next-gen wave of PCIe 5.0 graphics cards. GIGABYTE joins the ranks of ASUS and just recently, Cooler Master, with next-gen PCIe 5.0 ready PSUs.

The new GIGABYTE UD1000GM PCIe 5.0 PSU of course has the nifty, and very powerful, 16-pin PCIe 5.0 power connector that can drive up to 600W to a next-gen GPU. GIGABYTE admits that "traditional power supplies need a three 8-pin to 16-pin adapters to support the latest PCIe Gen 5.0 graphics cards" but their new UD1000GM PCIE 5.0 power supply "needs only a single 16-pin cable to directly supply power to the PCIe Gen 5.0 graphics cards".

A single 16-pin PCIe 5.0 power cable to your next-gen GPU... delicious. I'm sure those next-gen 16-pin PCIe 5.0 power connectors will be exclusive to NVIDIA's new flagship GeForce RTX 3090 Ti (and their new flock of Ada Lovelace-powered GeForce RTX 40 series GPUs later this year and into 2023).