Graphics Cards - Page 126

NVIDIA has just pushed out a new GeForce 516.79 hotfix driver, which fixes some issues that were present in the not-so-old previous GeForce driver.

The new NVIDIA GeForce 516.79 hotfix drivers "improves gameplay stability" in Apex Legends, and some of the games freezing in Overwatch. Red Dead Redemption 2 receives a performance improvement when using DLSS, where in the previous driver gamers were getting less performance.

NVIDIA staffer Manuel explains on the NVIDIA website: "A GeForce driver is an incredibly complex piece of software, We have an army of software engineers constantly adding features and fixing bugs. These changes are checked into the main driver branches, which are eventually run through a massive QA process and released".

Intel seems to have kinda kicked off its marketing campaign, without changing much: giving their new cards to YouTubers (this isn't free) to save on huge multi-million-dollar marketing campaigns... and now, we have a tease of the smaller (and what should be called the) Intel Arc A310 graphics card.

In a new photo, Alexey "Unwinder" Nikolaychuk aka the creator of RivaTuner/MSI Afterburner overclocking tools for graphics cards, posted a picture of the new Intel Arc A310 underneath a picture of MSI's absolutely mammoth GeForce RTX 3090 Ti SUPRIM X graphics card.

We don't know what pre-production card this actually is, but it should be the Arc A380... with a tiny card and dual-fan cooler and single-slot (?) design. We shouldn't have too much longer to wait before we see the full details on Intel's new tiny, and cute Arc A310 graphics card. In this area of the GPU market, AMD has its new Radeon RX 6400 and NVIDIA with its GeForce GTX 1630 graphics cards (also available in low-profile, and single-slot designs).

NVIDIA is waiting and watching to see what AMD can accomplish with its upcoming RDNA 3-based GPU architecture and its flagship Navi 31-based Radeon RX 7000 series graphics card.

So much so that it will be holding back the power on its GeForce RTX 4090, preparing a 600W beast, if RDNA 3 can fight for the next-gen GPU performance crown, according to the latest from Tom @ Moore's Law, who says that NVIDIA will launch AD102 in 450W first in October, and wait to see how Navi 31 performs for AMD once RDNA 3 is released.

If the flagship Navi 31-based Radeon RX 7000 series GPU can actually beat NVIDIA and its new GeForce RTX 4090 -- which should be twice as fast as the GeForce RTX 3090, just to refresh you -- then NVIDIA is ready with a more powerful AD102 GPU with up to 600W that would launch as the GeForce RTX 4090 Ti.

NVIDIA is in a pickle... they have got their AIB partners with warehouses filled with older-gen Turing, and even Pascal-based GeForce RTX series graphics cards and they have forced NVIDIA's hand into helping them get rid of the older-gen cards before the launch of the next-gen cards in a few months.

In a new video, Tom @ Moore's Law is Dead says that his AIB sources are pissed over the situation, and are threatening their next-gen Ada Lovelace GPU allocation -- so, effectively they're willing to give up how many Ada Lovelace GPU dies to them over another AIB if NVIDIA don't help them get rid of old stock -- which is a big, big move.

The AIB sources explained: "This month we told NVIDIA to go pound sound. We don't want to buy anymore Ampere or Turing stock until we liquidate our existing warehouses. At this point we're even willing to risk Lovelace allocation if they threaten us". They in that situation, being NVIDIA.

Intel's upcoming Arc A750 graphics card has been taken -- I'm sure in a sealed suitcase -- to the offices of Gamers Nexus. Check it out:

In the video, we get an explanation from Tom "Free Tom" Peterson (ex-NVIDIA, now at Intel) and Ryan Shrout (ex-PC Perspective, now at Intel) arrived to hang out with Steve "Tech Jesus" Burke at the Gamers Nexus offices, which I still haven't visited. Anyway, it shows off the lower-end Intel Arc A380 through to the higher-end Intel Arc A750 graphics card.

Intel's upcoming Arc A750 Limited Edition graphics card might have been teased, but no specifications have been confirmed by Intel just yet. Intel has said that the launch of these new Arc desktop GPUs is "sooner than we think", but I don't trust anything they say just yet. What we do know about the Intel Arc A750 is that it features 24 Xe-Cores and 12GB of GDDR6 memory on a 192-bit memory bus.

We are all waiting with massive amounts of anticipation for NVIDIA's next-gen Ada Lovelace GPU architecture and forthcoming GeForce RTX 40 series graphics cards... but before that, we have something interesting.

NVIDIA's second-fastest Ampere flagship GeForce RTX 3090 graphics card has had a custom BIOS installed, to see how it would perform with less memory, and on a slower memory bus. NVIDIA's GeForce RTX 3090 has 24GB of GDDR6X which was reduced to 20GB and 12GB in different tests, as well as its memory bus being dropped.

For the 20GB test, PRO Hi-Tech dropped down to a 320-bit memory bus, while for the 12GB test the memory bus was dropped to 192-bit. This reduced the memory bandwidth from NVIDIA's usual 936GB/sec on the GeForce RTX 3090 down to 780GB/sec with 20GB + 320-bit memory bus, and just 468GB/sec with 12GB + 192-bit memory bus.

Samsung has just announced the industry's first GDDR6 memory at 24Gbps, ushering in a new era of crazy high memory bandwidth but not the higher-end, and more power-hungry GDDR6X memory standard.

The new Samsung GDDR6 DRAM at 24Gbps is being built on Samsung's new 10nm EUV technology, with the company saying that the new memory is "designed to significantly advance the graphics performance for next-generation graphics cards (Video Graphics Arrays), laptops and game consoles, as well as artificial intelligence-based applications and high-performance computing (HPC) systems".

Samsung's new 24Gbps GDDR6 is 30% faster than the previous 18Gbps GDDR6 standard, where at full speed the new 24Gbps of GDDR6 will see memory bandwidth hitting 1.1TB/sec (1100GB/sec). At this speed, Samsung points out that it's fast enough to transfer around 275 x 1080p movies in a single second.

EVGA unleashed its monster new GeForce RTX 3090 Ti KINGPIN graphics card not too long ago, as well as its PowerLink 41s power adapter... and now... the upgraded EVGA PowerLink 52U has been unveiled.

The new EVGA PowerLink 52U features 5 x 8-pin PCIe power connectors, that plug into the 2 x 16-pin PCIe power connectors on EVGA's flagship GeForce RTX 3090 Ti KINGPIN graphics card. If you purchased the EVGA GeForce RTX 3090 Ti KINGPIN -- which costs $2500 and includes a free 1600W PSU, with 9 x PCIe power connectors -- there's no native 16-pin PCIe power connectors.

That's where the PowerLink 52U steps in, making your system look much nicer to look at, with the whole system looking much cleaner: especially, because there's so many damn PCIe power connectors required with all of the adapters (5 x 8-pin PCIe power connectors is insane for a single GPU).

I guess we're seeing the marketing launch of Intel's upcoming Arc desktop GPUs, with Intel preparing a new Arc desktop GPU bus that is hitting LANFest in September, and now Gamers Nexus have the new Intel Arc A750 Limited Edition graphics card.

Gamers Nexus have been sent over the new Intel Arc A750 Limited Edition graphics card, but haven't done much with it at all. VideoCardz incorrectly reported that GN was reviewing the card, but Gamers Nexus explained in a tweet that a "review isn't "already in the works." We said we ran some benchmarks on it (when an engineer visited), but we aren't reviewing it yet".

It seems Intel marketing is running GPUs into the hands of who they think matters... with PC Perspective's previous EIC and owner -- now working at Intel -- Ryan Shrout with the first GUNNIR Arc A380 Photon 6G OC graphics card in the US. Now, we have Gamers Nexus with an unreleased Intel Arc A750 Limited Edition graphics card. This must be the start of the Odyssey.

AMD is still cooking its next-gen RDNA 3 graphics architecture in the oven, but the flagship Navi 31-based GPU is getting some rumor mill goodness today: kinda, accidentally confirmed MCD counts.

The MCD (multi-chip die) is something we've been hearing about for a while, with more details arriving through leaks in April 2022, but now we're hearing that Navi 31 will have up to 6 x MCDs on a 384-bit memory bus. The news is coming from information posted about the RDNA 3 architecture in new Linux code, but was immediately retracted... not surprised.

AMD should use a single GCD (Graphics Complex Die) with six MCDs that should see a 384-bit memory bus, while we should see 192MB of Infinity Cache (or up to 384MB if AMD moves with its revolutionary 3D V-Cache technology on each of the MCDs).

Intel is ramping up the marketing for its Arc desktop GPU by... sending an Intel Arc gaming bus to LANFest Colorado in September.

The news was announced by Intel staffers on the Intel Insiders Discord channel, where the Intel Arc gaming bus is completely wrapped in "Arc High Performance Graphics". But what about inside? Inside isn't ready to game on yet, as Intel will be installing as many as 30 gaming stations that will be powered by their own Intel NUC 12 Extreme "Dragon Canyon" systems.

Inside of the Arc GPU gaming bus there's just renderings of what to expect with a suss-looking guy who doesn't like his picture being taken, or is looking at another monitor inside of the render. Intel has partnered up with SteelSeries which will provide peripherals, and AOC to provide gaming monitors. Jokes aside, I think Wccftech's Hassan Mujtaba puts it beautifully: "that truck better have a good driver".

The very latest version of GPU-Z adds improved support for Intel's finally-kinda-launching Arc GPU series, as well as NVIDIA's new GeForce GTX 1630 and AMD's new Radeon RX 6700 (non-XT) graphics cards.

GPU-Z 2.47.0 has "many improvements" for Intel Arc discrete GPU reporting, as well as added or improved fake detection for graphics cards based on the NVIDIA G98, GT200, GK104. There's a fix for the GeForce RTX 3080 12GB being reported as non-LHR, and so much more.

NVIDIA's new -- and slowest GPU of 2022 so far -- GeForce GTX 1630 is supported under GPU-Z 2.47.0 as well as the new non-XT model AMD Radeon RX 6700 graphics card. The GPU-Z changelog actually mentions the SAPPHIRE Radeon RX 6700 non-XT by name, but there are other Radeon RX 6700 non-XT graphics cards out there floating around in the wild.

NVIDIA is getting rid of its high-end GeForce RTX 30 series GPU family with some free games: Ghostwire: Tokyo and DOOM Eternal.

NVIDIA actually says they're giving away 4 free games with select GeForce RTX 30 series GPUs, but there are only 2 games because the other 2 are actually DLC. The free games are given away with NVIDIA's higher-end Ampere GPUs including the GeForce RTX 3080, GeForce RTX 3080 Ti, GeForce RTX 3090, and flagship GeForce RTX 3090 Ti graphics cards. The deal includes desktop gaming PCs with the eligible GeForce RTX 30 series GPUs inside, as well as the GeForce RTX 3080 Laptop GPU inside of gaming laptops.

You'll get Ghostwire: Tokyo and DOOM Eternal, as well as The Ancient Gods - Part One and Part Two DLC for DOOM Eternal. The promotion runs from July 12, 2022, through to August 1, 2022, or while stocks last. You can read all about it on NVIDIA's post about the "PC Game Bundle" on their official website.

A new version of the HYDRA overclocking tool is out with HYDRA 1.2 supporting AMD's upcoming Zen 4-based Ryzen 7000 series, auto GPU tuning, and so much more.

The new HYDRA 1.2B PRO software has been teased by developer Yuri "1usmus" Bubliy, where it features a complete design of all features, new Curve Optimizer diagnostics (and for Zen 4) and more flexibility for AMD PBO (Precision Boost Overclocking) and HYBRID OC.

HYDRA 1.2 will also feature an automatic problem solving assistant, auto GPU tuning, unlimited GPU profiling, powerful new monitoring, and more safety. Personally (and professionally), I can't wait to try out the GPU side of things with HYDRA 1.2, but we'll be waiting a couple more months before testing the Zen 4-based Ryzen 7000 series CPUs.

NVIDIA's flagship GeForce RTX 3090 Ti Founders Edition graphics card has just dropped down to $1599 on Best Buy, which is a decent $400 cheaper than the MSRP of $1999 when it launched back in March 2022.

In a world where there were listings just months ago for the ASUS ROG Strix GeForce RTX 3090 Ti, listed for close to $5000, as well as the ASUS ROG Strix LC GeForce RTX 3090 Ti which was online for over $4000. Even the GALAX GeForce RTX 3090 Ti EX Gamer appeared online for $3840.

Best Buy is an official NVIDIA partner, as they're the only place that you can buy NVIDIA's in-house GeForce RTX 3090 Ti Founders Edition in the US. NVIDIA hasn't dropped the price of its GeForce RTX 3090 Ti FE, but rather Best Buy has slapped $400 of discount on it -- but do you think Best Buy is doing that without NVIDIA's permission -- for their best GeForce GPU, and a Best Buy exclusive in the US at that. Nah, they know. But the bigger question is why?

It's not something big at all, but it's something: a new leak on Intel's upcoming Arc A770M mobile GPU, which has been benchmarked in some 3DMark runs alongside the slower A550M GPU.

We're hearing about the Intel Arc A770M and A550M GPUs from leaker "@_rogame" with some 3DMark Fire Strike Extreme (1440p) benchmarks. The lower-end Intel Arc A550M has 16 Xe-Cores, and scores 6017 points in 3DMark Time Spy using the older 1726 driver, but this benchmark was using Intel APO (Advanced Performance Optimizations) enabled... which as VC points out "basically renders it invalid as per 3DMark rules".

This means that the Intel Arc A550M is much slower than the A730M, but so it should be: it has half of the ACM-G10 GPU cores.

NVIDIA's next-gen Hopper GPU architecture is one of the most monsterous pieces of technology the human race has ever created... but it wasn't just humans... artificial intelligence (AI) helped in a huge uway, too.

In a recent post on NVIDIA's own Developer website, the company underlines how it used AI to design the greatest GPU it has created so far: the new NVIDIA Hopper H100 GPU. NVIDIA normally designs most of its next-gen GPUs using Electronic Design Automation (EDA) tools, but AI helped out with Hopper H100 through the PrefixRL methodology, which is an optimization of Parallel Prefix Circuits using Deep Reinforcement Learning. This allows NVIDIA to design smaller, faster, and more power-efficient GPUs... all while having more performance.

Rajarshi Roy, Applied Deep Learning Research at NVIDIA, tweeted: "Arithmetic circuits were once the craft of human experts, and are now designed by AI in NVIDIA GPUs. H100 chips have nearly 13,000 AI designed circuits! How is this possible?" and then posted a link to the blog that he co-wrote on NVIDIA's website.

AMD will debut its next-gen RDNA 3 architecture later this year, where rumor has it that the big boy flagship Navi 31-based offering will feature a GPU chiplet-based design... and NVIDIA won't have anything with GPU chiplets, and AMD knows that.

In a recent interview with VentureBeat, AMD senior vice president, corporate fellow and product technology architect, Sam Naffziger talked about GPUs using up to 600W of power by 2025, and so much more. But then VentureBeat's last question was a doozy "Compared to NVIDIA and Intel, do you feel like we're in a state of divergence when it comes to designs, or some kind of convergence?"

Naffziger replied, explaining: "It's hard to speculate. NVIDIA certainly hasn't jumped on the chiplet bandwagon yet. We have a big lead there and we see big opportunities with that. They'll be forced to do so. We'll see when they deploy it. Intel certainly has jumped on that. Ponte Vecchio is the poster child for chiplet extremes. I would say that there's more convergence than divergence.

We all know that the next-gen GPUs from both AMD and NVIDIA will be consuming much more power, especially for NVIDIA's highest-end Ada Lovelace GPUs, with AMD expecting to see GPUs using up to 600W of power by 2025.

AMD set an aggressive goal of 25x20 (hitting 25x better energy efficiency for its CPUs and GPUs by 2020) and has smashed it, so much so that the company has set a brand new 30x25 goal (hitting 30x better energy efficiency) by 2025 in the machine learning, and high-performance computing (HPC) markets in data centers.

The news is coming from VentureBeat, which spoke with AMD senior vice president, corporate fellow and product technology architect, Sam Naffziger. Naffziger said: "We've driven the frequency up, and that is something unique to AMD. Our GPU frequencies are 2.5 GHz plus now, which is hitting levels not before achieved. It's not that the process technology is that much faster, but we've systematically gone through the design, re-architected the critical paths at a low level, the things that get in the way of high frequency, and done that in a power-efficient way".

You can't buy Intel's new Arc Alchemist desktop GPU in the USA right now, but that is for us... the mere mortals... but the China-exclusive GUNNIR Arc A380 Photon 6G OC graphics card has made it into the Intel labs, and into the hands of Ryan Shrout.

The new GUNNIR Arc A380 Photon 6G OC graphics card packs the full Intel ACM-G11 GPU with 8 Xe-Cores, and 6GB of GDDR6 memory. Ryan Shrout, who does "graphics marketing" for the Intel GPU department, used to own and operate PC Perspective, has some hands-on with the GUNNIR Arc A380 Photon 6G OC in a video posted to his personal Twitter above.

He is of course talking it up, in its brief showing -- even though Intel's new Arc A380 GPU is disappointing compared to its competitors in their low-end SKUs -- with a very short demo of PlayerUnknown's Battlegrounds (PUBG) at a strange 2560 x 1080 (why... why...) on medium settings. The results seem alright, but we don't know any FPS details at all. Sad, given how Ryan had to do that when he ran PCPer and knows the tech press would pull this apart. #TeamIntel, I guess.