Graphics Cards - Page 134

NVIDIA is now hours away from launching its new flagship GeForce RTX 3090 Ti graphics card, with NVIDIA's in-house GeForce RTX 3090 Ti Founders Edition to become the first graphics card with a nifty new 16-pin PCIe power connector.

NVIDIA's upcoming GeForce RTX 3090 Ti Founders Edition will have a PCIe Gen5 16-pin power connector, replacing the 12-pin Molex MicroFit 3.0 plug that was on the RTX 30 series Founders Edition GPUs. The new 16-pin PCIe power connector can drive up to 450W on the RTX 3090 Ti, while the 12-pin PCIe power connector could drive up to 350W on the RTX 3090.

But on some of the custom GeForce RTX 3090 Ti graphic card design, we can expect up to 480W of power... or up to 516W with some tweaking. The upcoming ASUS ROG Strix LC GeForce RTX 3090 Ti (liquid-cooled with a dual-fan AIO cooler) has up to a 480W TDP, and can be tweaked for up to 516W. At this point, you're reaching the maximum power -- 525W -- of the PCIe 5.0 power connector and PCIe slot.

AMD is gearing up to its next-gen RDNA 3 GPU architecture, and during GDC 2022 we've got game developer Arkake Studios -- the team behind DEATHLOOP -- presenting at GDC 2022 with a talk called "A Guided Tour of Blackreed: Rendering Technologies in Deathloop".

In this chat, Arkane talked about the graphics engine advancements in DEATHLOOP, with the end of it addressing AMD's next-gen FSR 2.0 upscaling technology.AMD Developer Technology Engineer Lou Kramer talked about the comparisons between the edges of the wooden pattern being better-reconstructed in FSR 2.0 (Quality Mode, shifting from 1440p to 4K) that it evended up looking even better than native 4K... with TAA... and sharpening enabled.

But when it came to DEATHLOOP, we saw Kramer talking about the input sources of FSR 2.0 into the game -- including color buffer, depth buffer, and motion vectors -- DEATHLOOP also provides exposure textures for input, but FSR 2.0 can also compute one of those on its own.

AMD's new Radeon Adrenalin 22.3.2 drivers have been released, with support for Ghostwire: Tokyo and Tiny Tina's Wonderlands.

If you've got a Radeon RX 6900 XT, Radeon RX 6800 XT, or Radeon RX 6700 XT graphics card and were wanting to play Tiny Tina's Wonderlands, then you will enjoy up to 23% more performance, up to 21% more performance, and up to 20% more performance at 4K on those GPUs, respectively.

There's the usual bug fixes, with some issues squashed in Cyberpunk 2077 on the Radeon RX 570, issues with Stranger of Paradise: Final Fantasy Origin on the RX 6900 XT, and more. You can read more about that below, and download the new AMD Radeon Adrenalin 22.3.2 drivers right here (539MB download).

NVIDIA just unveiled its next-gen Hopper GPU architecture, AMD just teased its next-gen FSR 2.0 upscaling technology, and now Intel has detailed its upcoming XeSS upscaling tech at the Game Developers Conference (GDC 2022).

Intel had its software engineers discuss its new XeSS technology, saying that upscaling and anti-aliasing should be fixed as a single issue. Both older games and previous-gen GPUs would see performance improvements and quicker implementation with spatial upscalers.

On the other side, games that have ray tracing and other high-end graphics technologies that are only supported with the very latest GPU architectures, will obviously see bigger improvements in performance and IQ through temporal upscalers. As for GPU support, Intel plans to support as many GPUs as it can with XeSS -- supporting all GPUs that support Shader Model 6.4+ and DP4a instructions -- so most GPUs should be fine.

We are only a few days out from NVIDIA's new flagship GeForce RTX 3090 Ti graphics card launching, and with it will be a beefy new PCIe 5.0-ready 16-pin power connector.

But don't worry: you'll get a 16-pin to 3 x 8-pin PCIe power adapter, so you can plug your existing PSU cables into a turbocharged GeForce RTX 3090 Ti. But there aren't any PSUs on the market with high-end 16-pin PCIe power connectors just yet, which is why there will be a power adapter included.

The power adapter will provide 450W of power on its own, with the PCIe interface providing another 75W we have 525W total. This means we should expect custom variants of the GeForce RTX 3090 Ti to have a maximum TDP of 500W, up from the 400W on the custom RTX 3090 cards.

NVIDIA has unleashed its next-gen Hopper GPU architecture, its new AI accelerator that packs some serious horsepower, HBM3 memory, and is made on TSMC's new N4 process node.

The new NVIDIA H100 GPU has a huge 80 billion transistors, is made on the TSMC N4 process node (which NVIDIA and TSMC worked together on), and the H100 GPU itself comes in two forms with two counts of CUDA cores: 16896 FP32 CUDA cores for the SXM variant and 14592 cores for the PCIe-based model.

We also have a huge 80GB of ultra-fast HBM3 memory with an insane 3TB/sec of memory bandwidth with the Hopper-based H100, this is a change from the 40/80GB of HBM2e memory on the Ampere-based A100. This is a huge 1.5x increase in memory bandwidth alone, up from 2TB/sec.

Intel will be unleashing a monster HPC GPU in 2023 that will succeed its current-gen Ponte Vecchio GPU, with the first teases of Intel Rialto Bridge arriving aka Xe-HPC2 (or something like that).

But now that NVIDIA has burst through the gates with its Hopper GPU architecture and blisteringly-fast NVIDIA H100 GPU, it gives Intel and AMD time to make some adjustments to their respective HPC GPUs, schedules, and everything in between. Intel's new codename Rialto Bridge is reportedly launching next year according to Tom at Moore's Law is Dead.

We already know that the current Intel Ponte Vecchio -- which is based on the Xe-HPC GPU -- packs 100 billion transistors and two compute dies (each 640mm2). There are actually 47 different tiles doing different things, made on different process technologies (from Intel, and some outside of Intel). It's impressive to say the least.

NVIDIA has released their new GeForce Game Ready 512.15 WHQL drivers, which include the best day one experience for Ghostwire: Tokyo... but as always, there are more goods hidden inside NVIDIA's very latest GPU drivers.

Ghostwire: Tokyo can enjoy its beautiful ray-traced reflections and shadows, and super-boosts performance through NVIDIA DLSS. Ghostwire: Toyko isn't the only game to benefit, with Evil Dead: The Game launching on May 13 and will also include DLSS but there's also support for DLSS with these new drivers in Tiny Tina's Wonderlands, Chorus, Jurassic World Evolution 2, and No Man's Sky.

NVIDIA also recently teased its driver-based spatial upscaler -- NVIDIA Image Scaling (NIS) -- which boosts performance in ALL games. How does it do this? It lowers the rendering resolution of your game, and then upscales and sharpens the game to match the resolution of your monitor.

NVIDIA unleashed its next-gen Hopper GPU architecture and new flagship NVIDIA H100 GPU, but also its next-gen Grace CPU design that has been unveiled alongside two new Superchip platforms.

The new NVIDIA Grace Hopper Superchip has been designed for giant-scale AI and HPC (high-performance computing) use, we have NVIDIA's in-house Grace CPU + Hopper GPU combo, with an insane 600GB of GPU memory, 900GB/sec of coherent NVLink interface, and it runs the NVIDIA Computing Stack software suite.

There's a huge 30x system memory bandwidth to the GPU improvement over the DGX A100 here, with a CPU and GPU that were designed for giant-scale AI and HPC. The 900GB/sec of coherent interface is 7x faster than PCIe 5.0 which is only just dropping into gamers' desktops now, with PCIe 5.0-enabled GPUs launching this year.

NVIDIA has unleashed its next-gen Hopper GPU architecture and with it are new GPUs, HBM3 memory, the fastest silicon on the planet and now NVIDIA Eos: the world's fastest AI supercomputer.

The new NVIDIA "Eos" supercomputer will be the world's fastest AI supercomputer when it is turned on later this year, packing 576 x DGX H100 systems with a total of 4608 x DGX H100 GPUs. NVIDIA is expecting its new Eos supercomputer to have a huge 18.4 exaflops of AI computing performance: this is 4x faster than the AI processing of the Fugaku supercomputer in Japan.

The Fugaku supercomputer in Japan is the world's fastest system for AI processing right now, with NVIDIA's new Eos supercomputer expected to act as a "blueprint for advanced AI infrastructure from NVIDIA, as well as its OEM and cloud partners".

We all know that NVIDIA founder and CEO Jensen Huang is a little obsessed with AI... I mean, his company has shifted focus onto artificial intelligence so hard I wonder if Jensen is even alive anymore... especially when they DeepFake him at a virtual GTC 2021 and no one freaking notices (no, I wasn't joking... they really did fake the CEO of NVIDIA last year).

Anyway, NVIDIA's next-gen Hopper GPU architecture has been unveiled and it is built for the future of AI, now. The new Hopper GPU architecture led by the new flagship NVIDIA H100 GPU is helping turn data centers into something Jensen is now calling "AI factories".

Jensen explained from a virtual environment in the NVIDIA Omniverse real-time 3D collaboration and simulation platform -- see, he isn't even real -- that "companies are processing, refining their data, making AI software ... becoming intelligence manufacturers," describing how AI is "racing in every direction".

NVIDIA's new Hopper GPU architecture is official, and with it comes a huge wave of information: including the Hopper-based NVIDIA H100 GPUs being up to 40x faster with new DPX instructions.

The new NVIDIA Hopper GPU architecture accelerated dynamic programming -- which is a problem-solving technique used in algorithms for genomics, quantum computing, route optimization, and more -- by up to a huge 40x with new DPX instructions.

But wait, what is Dynamic Programming? Dynamic Programming was developed back in the 1950s, becoming a popular technique for solving complex problems with two key techniques: recursion and memoization. These might not sound like much, but man... does Hopper have some gigantic upgrades with them.

NVIDIA has just announced its next-gen Hopper GPU architecture, with the new flagship NVIDIA H100 GPU rocking 80 billion transistors, next-gen HBM3 memory technology, PCIe 5.0 connectivity, and enough power that if you use 20 of NVIDIA H100 GPUs you'd be able to sustain the entire world's internet traffic.

Yeah, you read that right: 20 x NVIDIA H100 GPUs have almost 5TB/sec of external connectivity, as well as 3TB/sec of memory bandwidth thanks to its HBM3 memory, which is enough that in the words of NVIDIA that 20 x NVIDIA H100 GPUs "can sustain the equivalent of the entire world's internet traffic".

In the middle of this is the new Transformer Engine -- which, when compared to the previous-gen A100 -- the new NVIDIA H100 speeds up natural language processing (Transformer being one of the most important deep learning models ever invented) by up to a gigantic 6x performance without losing accuracy. Beautiful work, NVIDIA... as always, hopping over the competition (which is yourself at this level).

The day is finally here: NVIDIA has unleashed its next-gen Hopper GPU architecture. NVIDIA announced the first Hopper-based GPU -- the new NVIDIA H100 GPU -- at its GPU Technology Conference (GTC 2022).

NVIDIA's new Hopper GPU architecture was named after pioneering US computer scientist, Grace Hopper, and succeeds the Ampere GPU architecture. The new NVIDIA H100 GPU has 80 billion transistors, and now becomes the world's largest and most powerful accelerator.

NVIDIA and TSMC (Taiwan Semiconductor Manufacturing Company) worked on the TSMC 4N process node, tweaking it for the Hopper-based H100 GPU. There's PCIe 5.0 support, next-gen ultra-fast HBM3 memory technology with an absolute blistering 3TB/sec (3000MB/sec). Insanity.

NVIDIA is hosting its annual GPU Technology Conference (GTC 2022) this week, where we should be introduced to their next-gen Hopper GPU architecture.

We should see NVIDIA introduce a next-gen MCM-based (multi-chip module) GPU with the flagship GH100, which should be a monster GPU with up to 140 billion transistors -- yeah, that's 140,000,000,000 transistor -- around 2.6x times the largest GA100 GPU from NVIDIA.

The full GH100 GPU should have 140 billion transistors, 144 Streaming Multiprocessors, and an insane 18,432 CUDA cores. The dual GPU chiplet design would have 2 x 144 SMs (288 SMs) and a mind-boggling 36,864 CUDA cores. It would be absolutely bonkers, and the 1000W+ power figures in rumors of the last few months come into play... and man, I can't wait. Bring it on, NVIDIA.

NVIDIA is just over a week away from unleashing their new flagship GeForce RTX 3090 Ti graphics card, with a new spotting of the custom ASUS ROG Strix LC variant for over $4000 at a Canadian retailer.

The upcoming ASUS ROG Strix LC GeForce RTX 3090 Ti and TUF Gaming GeForce RTX 3090 Ti graphics cards have been listed by a Canadian retailer for $5239 CAD ($4148 USD) and $4649 CAD ($3681 USD) respectively. If you consider the same retailer sells the ASUS ROG Strix GeForce RTX 3090 White Edition OC for $2998 CAD ($2378 USD) which means the RTX 3090 is over 70% cheaper than the RTX 3090 Ti... why would you buy the RTX 3090 Ti, at all?

We should expect to see ASUS unleash its new family of custom GeForce RTX 3090 Ti graphics cards on March 29, which is when the huge NDA lifts from NVIDIA. NVIDIA will be launching its own flagship GeForce RTX 3090 Ti Founders Edition while AIB partners can release their custom RTX 3090 Ti offerings on the same day.

Uh, so there are some warehouse workers that are in some serious trouble: staffers at Russian company "Wildberries" tried to steal 20 x GeForce RTX 3070 Ti graphics cards... but they were busted.

Like, I know GPUs are hard to buy right now but this is at a different level -- in Russia, well most countries now -- graphics cards are a "luxury good". But in Russia, AMD and NVIDIA both imposed embargoes on product shipments into Russia... so warehouses were hollowed out by either customers buying GPUs before they couldn't, or guys like this... stealing them.

The warehouse workers of Wildberries in Russia stole up to 20 x NVIDIA GeForce RTX 3070 Ti graphics cards from their warehouse, but they were busted. The warehouse workers stole the cards and stupidly tried to sell them at a local pawnshop... because the pawnshop owner was wondering why you'd have a brand new graphics card, and try to sell it right away. It seems those Russian memes aren't always right. In Russia, graphics cards can't buy pawnshop.

NVIDIA is dropping the price of its GeForce RTX 30 series GPUs across the world, with the current flagship GeForce RTX 3090 dropping to under 2000 EUR in Germany and Austria... for the first time since August 2021.

3DCenter.org has been keeping track of the prices of the GeForce RTX 3090, where the RTX 3090 was a bonkers 3199 EUR back on May 16, 2021 -- floating down to around 2500 EUR in the first few days of 2022 -- and now under 2000 EUR for the first time since August 29, 2021 (when they dropped down to 1999 EUR, but freaking blew up to 2800 EUR on November 21).

It's not just in Europe, but Down Under here in Australia my fellow Aussies at HardwareUnboxed have seen the ASUS TUF Gaming GeForce RTX 3080 drop by 35% in a single day. The GPU dropped from $2299 AUD ($1685 USD or so) down to $1499 AUD ($1100 USD or so). That's still expensive, but hey... a huge 35% price drop can't be sneezed at.

AMD has officially launched its new Radeon Super Resolution (RSR) technology, teased its next-gen FidelityFX Super Resolution 2.0 (FSR 2.0) and now launched its new Radeon Adrenalin 22.3.1 driver... which packs RSR tech support, right now.

The new Radeon Adrenalin 22.3.1 driver has its star: Radeon Super Resolution (RSR) technology, which is a new "in-driver spatial upscaling feature that is built on the same algorithm as AMD FidelityFX™ Super ResolutionGD-187 technology. It delivers near-native resolution and increased performance across thousands of games that run in exclusive full screen mode on AMD RDNA-based and newer discrete graphics".

AMD has also updated its AMD Link Play feature, which now lets 4 users connect to an AMD Radeon GPU-powered PC from an Android- or Windows-based PC, phone, tablet, or TV... so you can play local multiplayer games together. Very cool.

AMD has just announced that its FidelityFX Super Resolution 2.0 technology will be launching in Q2 2022, using temporal upscaling.

The new FSR 2.0 technology will shift away from spatial technologies that FSR 1.0 uses, and into temporal upscaling -- but while NVIDIA DLSS and Intel XeSS use machine learning algorithms -- there is Tensor Core or XMX acceleration required.

AMD explains: "Like FSR 1.0, FSR 2.0 does not use Machine Learning (ML) in its upscaling algorithm. While ML is one vehicle to solve a number of problems, it is not a requirement to achieve good quality upscaling. Therefore, FSR 2.0 does not require dedicated ML hardware which means more gamers can benefit from it. FSR 2.0 will work across a wide range of products and platforms, and like FSR 1.0, that includes select competitor graphics cards".