Graphics Cards - Page 117

NVIDIA has detailed its next-gen Hopper H100 GPU at the Hot Chips 34 event, with the new Hopper H100 GPU being made on TSMC's fresh 4nm process node.

TSMC's new 4N process node was optimized and designed exclusively for NVIDIA and its new Hopper H100 GPU, where NVIDIA flexes its silicon muscle with its 4th Gen Tensor Core architecture, the world's first use of HBM3 memory, and so much more.

The new NVIDIA Hopper H100 GPU is an absolute behemoth: made on TSMC's new 4N process node exclusively for NVIDIA, with 80 billion transistors, and the world's first use of HBM3 memory technology. NVIDIA has built the Hopper H100 onto the PG520 PCB board, which packs 30+ power VRMs and a gigantic integral interposer that uses TSMC's new CoWoS technology: this combines the NVIDIA Hopper H100 GPU with a 6-stack HBM3 design for a very potent mix of GPU + VRAM dominance.

We're seeing some juicy silicon news coming out of the Hot Chips 34 conference, where AMD showed off the GPU block diagram for its upcoming MI250X MCM GPU.

AMD's upcoming MI250X GPU was detailed at Hot Chips 34, where we get the GPU block diagram that gives us all the good stuff in terms of specifications and details. The new MI250X features not 1 but 2 x GPUs for a total of 58 billion transistors, with AMD getting TSMC to make their MI250X GPU on their 6nm process node.

We also have 128GB of HBM2e memory at a huge 3.2TB/sec memory bandwidth, in-package Infinity Fabric with 400GB/sec of bandwidth between the two GPUs. All-in-all, AMD has 10 chips that makes up the MI250X "Aldebaran" GPU, with a mix of GPU dies, and HBM2 memory dies (2 x GPUs, 4 x HBM2e stacks per GPU for 8 x HBM2e stacks total + 2 x GPUs = 10 chips in total).

Intel has detailed a little more performance data on its upcoming Ponte Vecchio GPUs on its upcoming Sapphire Rapids HBM server platform during Hot Chips 34.

The next-gen Intel Ponte Vecchio GPU was detailed during a presentation by Intel Fellow and Chief GPU Compute Architect, Hong Jiang, where the new Ponte Vecchio will arrive in 3 configurations with a single OAM that has up to an x4 Subsystem with Xe Links, in a single or dual-socket Intel Sapphire Rapids platform.

Performance-wise, Intel said a 2-Stack Ponte Vecchio GPU configuration on a single OAM is capable of up to 52 TFLOPs of FP64/FP32 compute, 419 TFLOPs of TF32 (XMX Float 32), 839 TFLOPs of BF16/FP16 and 1678 TFLOPs of INT8 horsepower. Intel said its maximum cache sizes and peak bandwidth for each of them: the Register File size on Intel's new Ponte Vecchio GPU is 64MB with a huge 419TB/sec of memory bandwidth.

Intel's entry-level Arc A380 desktop graphics card has reached the hands of an overclocker, who has pushed it above 3.0GHz, but the overclocker said that the experience and process were far more complicated than it should be.

SkatterBencher secured an Intel Arc A380 graphics card from a Chinese website and has been playing with it ever since: the first problem that he ran into was that the official Arc Control software is the one and only overclocking tool to push the Arc GPUs. But it was so bad that the overclocker had to analyze driver code to see what the click OC percentage slide actually does: that's when he discovered that there is a limit of 2450MHz for the default Voltage-Frequency curve.

Intel has two overclocking modes on its Arc discrete GPUs: offset mode and lock mode. Offset mode increases the maximum frequency with higher voltage, similar to NVIDIA, while the lock mode provides you with the ability to adjust voltage and frequency manually, where you can push up and above the maximum 2450MHz GPU frequency.

Intel's entry-level Arc A380 graphics card is kinda out in the wild, where it's available in China and coming to PCs in the US inside of pre-made systems... but now the Arc A380 has been tested with the gold standard of PC gaming over a decade ago: Crysis.

Yes, the Intel Arc A380 graphics card can run Crysis... and not too badly I might add. We're talking about 1080p 60FPS+ with the "Very High" preset and it even had 4x MSAA enabled. GUNNIR's custom Intel Arc A380 Photon OC graphics card was tested, with a 92W TBP (75W TBP on the reference Arc A380) and higher GPU clocks: 2450MHz (up from 2250MHz on the reference Arc A380).

Crysis was one of the very best-looking games when it was released, and hell... it still is. It looked better than most flagship AAA games on either of the consoles at the time, and for many years after Crysis was released. In the new Intel Arc A380 graphics card benchmarked with Crysis, it's using the DX10 API, 1080p resolution, Very High preset and 4x MSAA enabled for 60FPS+ without an issue.

Slowly but surely the world finds out the mess that the Intel Arc GPU situation is, with enthusiasts and readers of TweakTown and any other tech press (written or video) should know by now... but now Intel is thanking Gamers Nexus for exposing 43 separate issues within the Arc GPU software.

In a new blog post on the official Intel website, Intel's Vice President of the Visual Computing Group, Lisa Pearce, explains some of the recent happenings within the Intel Arc graphics software development department. There's not much new here, but there are some interesting things like saying the reviews of their Arc A380 graphics card by Gamers Nexus (who said it wasn't a review) helped them file 43 issues with their engineering team.

One of the questions that Lisa answers is "Question #3: I've read reviews that say Intel's driver is not ready yet. What's the status?" to which she answers: "We appreciate the feedback we are getting in early reviews of our Arc software stack. And it has been bumpy. We have received frank feedback from press during recent reviews, and we have taken it to heart. For example, we filed 43 issues with our engineering team from a review of the A380 by Gamers Nexus".

Intel has just pushed out its new Arc graphics driver "30.0.101.3268" with some new titles getting Game-On support, a list of games that have been optimized, and a bunch of the usual bugs being squashed.

The new Intel Arc graphics driver "30.0.101.3268" has Game-On support for both Saints Row and Madden NFL 23, on the Intel Arc A-series graphics cards, while there are optimizations for the just-released Marvel's Spider-Man Remastered.

Along with the new Spider-Man Remastered optimizations, Intel's new Arc GPU drivers also fix Shadow of the Tomb Raider from crashing when setting the Ray Traced Shadow Quality to High. Intel notes that a workaround to this is changing it to "Medium" or "Off" instead.

NVIDIA will be detailing its next-gen Hopper GPU architecture at Hot Chips 2022, where NVIDIA engineers will be speaking during four talks across two days at the event.

NVIDIA will have a senior distinguished engineer who will lift a bit more of the veil behind the NVIDIA Hopper H100 GPU, with 14-year veteran Jack Choquette. NVIDIA's new Hopper H100 is the world's first to use the super-fast new HBM3 memory system, which pumps away at 3TB/sec (3000GB/sec).

The virtual Hot Chips event will see the annual gathering of process and system architects, where NVIDIA teases that they will disclose performance numbers and other technical details on another interesting NVIDIA product: their first server CPU, as well as the Hopper GPU, the latest version of the NVSwitch interconnect chip, as well as the NVIDIA Jetson Iron system on module (SoM).

Intel's new Arc Alchemist graphics card is nearly here... nearly... with ASRock's new custom Challenger Arc A380 6GB graphics card now up for pre-order on Newegg.

ASRock's new Challenger Arc A380 6GB graphics card is up for backorder on Newegg, with the company saying that the "ETA date" is the 22nd of August... just a few days from now. The listing for the card does detail that it has the new DisplayPort 2.0 connector, but the older HDMI 2.0b standard (and not HDMI 2.1, which is capable of 4K120).

Inside, the card is based on Intel's full ACM-G11 GPU on the TSMC 6nm process node with 8 Xe-Cores or 1024 ALUs, with a max GPU clock of 2450MHz. 6GB of GDDR6 is here at 15.5Gbps on a 96-bit memory bus that pumps 186GB/sec of memory bandwidth, with a 92W TBP.

The mystery that is Intel's GPU marketing plan continues, where it seems that the company is playing around with the desktop Arc A580 graphics card with the first benchmark showing up.

Intel's upcoming Arc A580 graphics card sits right in the middle of the Arc A380 and Arc A770 graphics cards, where it uses the Intel ACM-G10 GPU with 16 Xe Cores (2048 FP32 cores) versus the Arc A380 with the ACM-G11 GPU and 8 Xe Cores (1024 FP32 cores) and the flagship Arc A770 with the ACM-G10 and its 32 Xe Cores (4096 FP32 cores).

As for the benchmark run, the Intel Arc A580 desktop graphics card was run with an Intel engineering sample CPU that rocks 16 cores, meaning it could be the new 13th Gen Core "Raptor Lake" CPU, or something else entirely. I'm sure it would be a new Raptor Lake CPU + Arc A580 graphics card and ran Ashes of the Singularity (in the Vulkan API, not DirectX 12) at 1080p on the Minimum graphics preset: 9300 points and 95FPS average.

We heard more details on AMD's upcoming RDNA 3-based Navi 3X GPUs last week: including the Navi 31, Navi 32, and Navi 33 GPUs, and a month ago we saw some fan-made renders... but now we have some updated renders, with the very latest rumored specs on the new Navi 3X GPUs.

The new renders are from Twitter user "Wild_C" who made some new renders of the Navi 31, Navi 32, and Navi 33 GPUs after his original renders a few weeks ago. The updated die sizes are interesting to note, while some more details on the GPUs themselves (GPU specs + memory + memory bus) are provided by 3DCenter.org.

AMD's new Navi 31 GPU is looking dapper in the render, with a die size of 533 mm², that's including 308 mm² GCD (Graphics Die) and six 37.5 mm² MCDs (Memory Dies). This is up from the previous specs, where it was estimated that Navi 31 would have a die size of 633mm². Some of these specs are "pure speculation" just to note, so don't get your hopes up too much... still, it's exciting to see.

NVIDIA is reportedly resuming production of its GeForce RTX 3080 12GB SKU, after production stopped some two months ago now.

The company released the newer GeForce RTX 3080 12GB graphics card earlier this year in January, an upgraded version over the GeForce RTX 3080 10GB graphics card. The biggest two differences between the GeForce RTX 3080 10GB and GeForce RTX 3080 12GB is that firstly, the obvious difference in 12GB of GDDR6X memory over 10GB of GDDR6X memory. Secondly, the RTX 3080 12GB is an LHR "Lite Hash Rate" model, while the GeForce RTX 3080 10GB has full-speed crypto mining performance.

But now, NVIDIA faces warehouses of current-gen Ampere-based GeForce RTX 30 series graphics cards, in the months leading up to its next-gen Ada Lovelace-powered GeForce RTX 40 series graphics cards. This is why we're hearing that NVIDIA will reportedly launch the flagship GeForce RTX 4090 graphics card first, and the GeForce RTX 4080 and GeForce RTX 4080 later... because there are too many higher-end GeForce RTX 30 series graphics cards left over in warehouses, and now more GeForce RTX 3080 12GB graphics cards are being made.

Intel has officially dropped native DirectX 9 hardware support for its integrated Xe GPUs on its 12th Gen Core "Alder Lake" CPUs, as well as its upcoming Arc A-series desktop GPUs.

In its place, Intel will be using a DX9 to DX12 emulator called "D3D9On12" from Microsoft, where it will send 3D DirectX 9 graphics commands to the D3D9On12 layer, instead of directly to the D3D9 graphics driver. As soon as the D3D9On12 layer gets the command from the D3D9 API, it will convert all of the commands into D3D12 API calls.

D3D9On12 is pretty much a GPU driver on its own, instead of a GPU driver with DX9 support from Intel. Microsoft says that the performance out of its DX9 to DX12 emulator should be nearly, if not as good as native DX9 hardware support. I don't see how native DX9 hardware support inside of Intel Xe and Arc GPUs isn't here, but DX9-based games should be able to run even on the mid-range first-gen Arc A-series GPUs.

Intel had its Arc A770 discrete graphics card at SIGGRAPH 2022 recently, with the company running its Arc A770 graphics card in Blender Cycles with ray tracing.

Intel's Director of Graphics Community Engagement, Bob Duffy, tweeted out that he was running Blender Cycles with ray tracing and live denoising running on an Intel Arc A770 graphics card, inside of an Intel NUC 11 Extreme PC. CPU-wise, it is powered by an Intel 11th Gen Core CPU.

As for the Intel Arc A770: we have the full ACM-G10 GPU with 32 Xe-Cores and 16GB of GDDR6 memory. Interesting to note, given SIGGRAPH is a professional event and not for gamers, but the gaming-focused Intel Arc A770 graphics card was used... and not the recently announced Intel Arc A50 and Arc A40 workstation cards.

AMD and NVIDIA are both about to announce and release their next-gen graphics cards in the coming months, but before then we've seen pricing on both sides of the GPU camp drop significantly... thanking the GPU gods.

We've already seen EVGA slice and dice the price of its high-end GeForce RTX 3090 Ti FTW3 by $1000, dropping it from $2149 down to $1149... but now the AMD's second-fastest Radeon RX 6900 XT graphics card, and NVIDIA's third-fastest GeForce RTX 3080 Ti are both experiencing healthy price cuts which are now dipping their toes in -30% below MSRP territory.

Our friends at Wccftech have noticed that the ASRock Radeon RX 6900 XT Phantom Gaming D graphics card is now just $699, down -30% over its $999 MSRP. If you don't want to go with ASRock, Newegg has a listing for the GIGABYTE Radeon RX 6900 XT GAMING OC graphics card, also $699... down from its usual $899 pricing on the site. Not bad, not bad at all...

Intel kicked off its Xe HPG Scavenger Hunt in early 2021, and a rather large 18 months later we have no Arc graphics cards in the wild... meaning no Intel Arc graphics cards to give away to Arc Scavenger Hunt winners.

The company is now offering an alternative prize because there are simply no Intel Arc GPUs, so instead of an Intel Arc A7-series graphics card (Intel was giving Grand Prize winners an Intel Arc A770, while the first place prize was an Intel Arc A750) winners can now opt for an Intel Core i7-12700K and Core i5-12600K processor.

Intel sent an email to its winners, explaining to a Xe HPG Scavenger Hunt Top 300 prize winner that they would be provided with an alternative to the original hardware prize, in addition to the standard merchandise and swag prizing. The company confirms that the winners only have a few days to decide -- up to August 19 -- if you want to switch over your Arc GPU to an Intel 12th Gen Core "Alder Lake" CPU.

AMD's next-gen Radeon RX 7000 series graphics cards will begin their journey into our lives, benchmark charts, and hearts later this year... but in flagship Navi 31 GPU form first, and then comes Navi 32 aka "Wheat Nas".

The next-gen Navi 32 GPU is codenamed "Wheat Nas" with some new purported details on the smaller version of Navi 31, which will be re-using the same MCDs (Memory Chiplet Dies) as the Navi 31. AMD will be using its flagship RDNA 3-based Navi 31 GPU for desktop Radeon RX 7000 series graphics cards (Radeon RX 7900 XT, and others) while the Navi 32 GPU will be offered in mobile, and high-end desktop GPU form as well.

Angstronomics reports that Navi 32 was planned with 128MB Infinity Cache (as a 1-hi version) but AMD is reportedly going with just 64MB of Infinity Cache for Navi 32, down from the 128MB on the upcoming Navi 31 GPU. While the current-gen RDNA 2-based Navi 21 has 128MB of Infinity Cache, and Navi 22 has 96MB of Infinity Cache, meaning Navi 32 comes with less Infinity Cache than Navi 22... interesting.

A new report from Angstronomics says that AMD's next-gen RDNA 3-based Navi 3X GPUs were finished in 2019, finalized sometime in 2020, and have had "no changes since".

AMD's upcoming Navi 31 is the flagship GPU based on the RDNA 3 graphics architecture, which will reportedly feature 1 x GCD (Graphics Chiplet Die) and 6 x MCDs (Memory Chiplet Dies). Each of them will contain 16MB of Infinity Cache, joined by GDDR6 memory and 64-bit wide PHYs according to the report. Navi 31 is expected to rock out with 48 WGPs (Work Group Processors) and a 384-bit memory bus using GDDR6 memory.

According to this new information, Navi 31 will have a total of 96MB of Infinity Cache, which is actually lower than the current-gen RDNA 2-based Navi 21 GPU which has 128MB of Infinity Cache in total. Angstronomics adds that there will be a cut-down SKU that will be made available, packing 42 WGP and 5 x MCD (80MB of Infinity Cache in total, on a 320-bit memory bus using GDDR6 memory).

It was barely two weeks ago that I reported from leaks that Intel's Arc desktop GPU is so bad, it could be CANCELLED altogether... But now a major analyst has said that Intel could shut its GPU business because it's bleeding too much money: $3.5 billion so far.

In a new market research report from Jon Peddie Research, the firm estimates that Intel's Accelerated Computing Systems and Graphics (AXG) group has lost $3.5 billion SO FAR. Intel bleeding out $3.5 billion in a division that should be around for decades to come and pull hundreds of billions in that time, is not a good sign, and neither is one of the biggest analysts in the world saying Intel should shut the GPU department down because of it.

JPR founder Jon Peddie said: "So, the rumor mill has been hinting that the party is over and that AXG would be the next group to be jettisoned".

Exclusive: Intel has confirmed with TweakTown that its upcoming Intel Arc graphics cards will NOT support multi-GPU for gaming, which is not a surprise, but still disappointing to hear.

In the last 36 hours, it's been a whirlwind about Intel supporting multi-GPU for its upcoming Arc GPUs and their oneAPI software. Our own Rob Squires wrote: "software support for an Intel multi-GPU solution is here and almost made its debut at SIGGRAPH 2022".

But then this morning I woke up to emails from an Intel representative clearing it all up, with the Intel representative explaining: "Intel showed a Blender Cycles rendering demo at SIGGRAPH with Intel Arc graphics. Multi-GPU rendering support for Intel Arc and Intel Arc Pro graphics cards through oneAPI is supported starting in Blender 3.3. Intel Arc graphics does not support multi-GPU for gaming".