Graphics Cards - Page 135

It looks like there might be some quick price relief on NVIDIA GeForce RTX graphics cards, with NVIDIA reportedly lowering the cost of its GPUs by up to 12% to AICs, which will then filter down to you: the consumer.

NVIDIA has reportedly informed AICs of a "decrease in its cost basis which it will be passing on to AICs and who will be passing it on to SIs (system integrators) in return", according to Wccftech. We might not see the price drops instantly, as there are still supply issues, cryptocurrency mining is still very popular, and more.

Wccftech continues: "we have received word that NVIDIA has informed AICs that it saw a lowering in its cost of manufacturing by around 8 to 12% which it will pass on to AICs. AICs in return will pass this saving on to SIs where it should eventually end up saving a few dollars in the gamer's pocket".

Apple unleashed its new M1 Ultra SoC last week, powering the new Mac Studio -- which can cost over $10,000 when configured high enough -- but the GPU inside of the new SoC was meant to beat the GeForce RTX 3090.

Inside of Apple's new Mac Studio is the M1 Ultra SoC with up to 21 TFLOPs of compute power, 128GB of unified memory, and a GPU that Apple marketed as offering performance of flagship GPUs like the RTX 3090 in "industry-standard" benchmarks... spoiler alert: it doesn't.

Diving right into some benchmarks, if we look at something like Shadow of the Tomb Raider... at 1080p the Apple M1 Ultra is doing alright, with 108FPS average. 96FPS average at 1440p is impressive, and 60FPS at 4K isn't bad at all for a tiny little SoC. The RTX 3090 rips it apart though... going against everything Apple said in its marketing. But that shouldn't surprise people at this point.

AMD is about to launch its new Radeon Adrenalin GPU drivers, and with it comes a new teaser video that shows off what the new drivers will include: including Radeon Super Resolution (RSR) technology.

AMD Radeon Super Resolution is the latest spatial upscaling technique based on FidelityFX Super Resolution technology, which was meant to drop in Q1 2022 -- which ends March 31, 2022 -- and will drop inside of the new Radeon Adrenalin GPU drivers tomorrow.

RSR will benefit in a huge way in the fact that game developers don't need to integrate anything in order to have the benefits of RSR, versus FSR needing compliance, meaning that Radeon Super Resolution should work in thousands of games. The loss here? RSR has lower image quality than FSR, as RSR upscale the whole frames (including the UI and menus) and should provide more mixed results over FSR.

We all know that GDDR6X memory temperatures can really get crazy, where the 24GB of ultra-fast GDDR6X memory inside of a GeForce RTX 3090 can even hit 110C... well, what about with some mods to the VRAM chips?

But now those insane 110C memory temperatures have been reduced down to nearly half, with YouTuber 'DandyWorks' modding his ASUS TUF Gaming GeForce RTX 3070 Ti graphics card. He ripped the card apart and replaced the thermal pads with copper pads -- which isn't easy -- and the GDDR6X memory temps dropped from 110C to 64C.

You'll need a few things before you can reduce your GDDR6X memory temperatures inside of your expensive, and probably very hard to get, GeForce RTX 30 series graphics card. You'll need copper shims, non-conductive Kapton tape, and thermal paste with an applicator to get it onto your GDDR6X memory chips. You might also need 400 grit sandpaper, so that you can smooth over those beautiful copper shims.

AMD first announced its DLSS competitor in the form of FSR back in late May 2021, and is now gearing up into debuting its next-gen FSR 2.0 technology.

The company will be detailing its next-gen FSR 2.0 upscaling tech at GDC 2022 (the Game Developers Conference) later this month... but ahead of that, the developer of CapFrameX benchmarking and monitoring software, posting on Twitter saying that he'd seen an FSR 2.0 demo and that there was "impressive performance and image quality".

CapFrameX also said that "AMD even claims that it can be better than native", with "temporal upscaling + optimized AA", it "does not need AI" (so no dedicated Tensor Cores like GeForce RTX GPUs), and it "runs on GPUs from all vendors".

AMD is getting ready to drop its refreshed Radeon RX 6950 XT graphics card, with some 'Super-style' upgrades including faster 18Gbps GDDR6 memory.

But in some fresh rumors, industry insider "Greymon55" has teased that AMD's in-house reference Radeon RX 6950 XT graphics card will feature an all-black cooler. This would be a shift away from the silver + black + red theme of the first wave of Radeon RX 6000 series GPUs -- including the flagship Radeon RX 6900 XT reference graphics card -- so an all-black cooler must mean business.

An all-black cooler reminds me of the earlier TITAN series GPUs from NVIDIA, where the TITAN X dropped almost to this day in 2015. NVIDIA used the GeForce GTX TITAN X as a much more premium-focused GeForce GPU with much more VRAM (12GB at the time) and a serious all-black cooling design.

NVIDIA's next-gen GeForce RTX 4090 is expected to be quite the beast: not just in brute performance, where we've got rumors of double the RTX 3090 performance, and huge 600W power claims.

Now there's some more confirmation -- from the rumor mill, but still this source is pretty solid -- from "kopite7kimi" on Twitter. In a new tweet, he says "someone told me the xx90 with 600W TGP has been confirmed. I think it's too early to talk about it".

600W is a huge amount of power for a single GPU, but with rumors of the Ada Lovelace-powered GeForce RTX 4090 (with the AD102 GPU in particular) we're looking at double the performance of the GA102-powered GeForce RTX 3090. That's an incredible amount of horsepower, and less than double the 300-350W+ that the RTX 3090 consumes.

NVIDIA has a secret stock of flagship GeForce RTX 3090 graphics cards, with the company giving away 8 of them at GTC 2022 (its upcoming GPU Technology Conference) and they'll be signed by CEO Jensen Huang.

Jensen Huang will be signing the NVIDIA GeForce RTX 3090 Founders Edition graphics cards, a pity because the new GeForce RTX 3090 Ti would have launched by then -- RTX 3090 Ti launches on March 29, custom AIB versions as well -- but a signed RTX 3090 is still freaking awesome.

GTC 2022 is all virtual this year, which is unfortunate because it was one of my favorite trade shows to attend pre-pandemic. NVIDIA hasn't been clear on how it will be giving away the GeForce RTX 3090 Founders Edition graphics cards signed by CEO Jensen Huang, but the only requirement for the giveaway is to be a participant of the hybrid (virtual/in-person workshops) GTC conference.

It's pretty hard to find some silicon on this planet right now, next-gen consoles, any generation GPU, some generations of CPUs... but now a man has been busting illegally importing 160 CPUs into China.

Chinese Customs Officials discovered that the "Walking CPU" or "CPU-Man" was trying to illegally import the processors into China, noticing that his walking posture was abnormal, custom officials pulled him to the side. Not only was he busted with 160 processors -- mostly Intel 11th Gen and 12th Gen Core CPUs -- he had 16 foldable smartphones on his person, too.

Chinese Customs Office explains: "[Walking CPU] On March 9, the gate customs seized an entry case of a passenger hiding a central processing unit (CPU). At about 1 a.m. that day, a man named Zeng entered the country through the customs "no declaration channel" at the travel inspection site of Gongbei Port. Customs officers found that his walking posture was abnormal and stopped him for inspection".

It looks like NVIDIA will not only be unveiling its new GeForce RTX 3090 Ti graphics card, but review NDAs, custom GeForce RTX 3090 Ti cards from AIB partners, and everything in between will be unleashed on March 29, 2022.

The new information is coming from VideoCardz, which reports that "product reviews, including custom models, will also be published" on March 29. NVIDIA has reportedly shared the embargo information for the GeForce RTX 3090 Ti with its board partners, as well as final dates for submission of the press release (so that NVIDIA can green light it) and the review seeding list (to people like me).

NVIDIA's new flagship GeForce RTX 3090 Ti graphics card will be unveiled on March 29 at 6AM PST.

NVIDIA's new GeForce RTX 40 series GPUs have been detailed in great length over the last couple of months, but so too have AMD's next-gen Radeon RX 7000 series GPUs.

We now have some more information from Tom @ Moore's Law is Dead with his sources providing some new details, and pieces of the Navi 31, Navi 32, and Navi 33-based Radeon RX 7000 series graphics cards will form into. The flagship Radeon RX 7970 XT graphics card is teased here, with the Navi 31 GPU and up to 32GB of GDDR6 memory.

AMD's next-gen Radeon RX 7970 XT graphics card would tote its 32GB of GDDR6 memory on a 256-bit memory bus, and have a TDP of between 375-450W. As for the Navi 31, it is an MCM (multi-chip module) GPU and packs up to a ridiculous 15360 FP32 shading units (Navi 21 GPU inside of the RX 6900 XT has 5120 shading units).

Basemark has just unveiled its all-new GPU performance benchmarking suite, which is fully cross-platform and works on all modern graphics APIs and operating systems.

The new GPUScore works on graphics APIs including Vulkan, Metal, and DirectX while operating systems like Windows, Linus, macOS, Android and iOS are supported. There are three different testing suites available, with the first of GPUScore's new benchmarks -- Relic of Life -- released immediately, while the others will be released over the "following months".

GPUScore Relic of Life will become a big part of high-end gaming PCs and their super-fast graphics cards, with support for ray tracing benchmarking of both Vulkan and DirectX, on Windows and Linux, respectively. Tero Sarkkinen, Founder & CEO of Basemark explains: "As veterans of the graphics performance benchmarking industry, we know that it takes the most modern tools to reliably benchmark the latest GPU technologies. I am proud of the new GPUScore Relic of Life benchmark".

NVIDIA is going the whole hog with its upcoming Ada Lovelace GPU architecture, with the new GeForce RTX 40 series cards -- even in mid-range form -- expected to offer huge performance increases over the current-gen GPUs.

The new flagship AD102-based GeForce RTX 4090 and GeForce RTX 4080 graphics cards will be absolute behemoths, offering double the performance of the GeForce RTX 3090 and GeForce RTX 3080. NVIDIA will reportedly be outfitting its highest-end Ada Lovelace GPUs with next-gen GDDR7 memory, which begins its speed-run at 32Gbps (GDDR6X taps out at 21Gbps).

But it's the new mid-range AD103 GPU that will be really interesting... AD103 will power the GeForce RTX 4070 and GeForce RTX 4060 Ti graphics cards, and according to sources of Tom @ Moore's Law is Dead, the GeForce RTX 4070 should offer another 10-30% performance over the current flagship RTX 3090, and offer "2-3 tiers better" ray-tracing performance than the best Ampere GPU.

NVIDIA's next-gen Ada Lovelace GPU will be powering the GeForce RTX 40 series graphics cards later this year, with rumors already circling that power numbers will be Fermi levels of high -- fresh rumors are here -- so prepare your power bills.

In his latest video, Tom @ Moore's Law is Dead has said that his sources are saying "we were told 400W+ is going to be the new norm for enthusiast products. However, we have been given no indication of anything close to 800W". Another source said: "NVIDIA has told us power consumption is going way up. I expect at least 450-550W".

Another two sources added: "500-600W is the current expectation" and "AD102 will use at least 450W, and I think 600W is doable on air".

Intel is preparing to launch its Arc Alchemist GPUs in the next few months, with a new report suggesting a (really wide) launch window of May 8 to June 5, 2022.

In this time period we'll have Computex 2022 between May 24-27, which normally takes place in Taipei, Taiwan. Intel usually has a gigantic Computex 2022 keynote, and now we're hearing from Igor's Lab that Intel is gearing up to launch its desktop Arc Alchemist GPUs sometime before, during, or after Computex.

Igor writes: "Several different sources have told me of another Intel card postponement, at least as far as SKU 1 to 3, the three performance models, are concerned. There are currently consistent rumors that the first models will be released between May 2, 2022 and June 1, 2022".

AMD will have its refreshed RDNA 2 graphics cards out in the coming months, with a reveal expected next month in April -- with the new flagship Radeon RX 6950 XT leading the way -- packing faster GDDR6 memory at 18Gbps.

The new rumors are coming from Enthusiast Citizen, with a post on Chiphell where he explained (through machine translation): "4/20 or ‍4/21 Several new cards, RX 6950XT, RX 6750XT, RX 6650XT, will be a little more expensive than the current ones without 50. All three cards have been upgraded to 18Gbps, and the performance has also been slightly improved. There is a low-end bright machine card RX 6400".

We're to expect a refreshed Radeon RX 6950 XT, Radeon RX 6750 XT, and lower-end Radeon RX 6650 XT. AMD will be utilizing faster 18Gbps GDDR6 memory, up from the 16Gbps GDDR6 memory on the current Radeon RX 6000 series GPUs. There'll be additional performance on top, which will be interesting to see if it's enough to match NVIDIA's new flagship GeForce RTX 3090 Ti which is now expected on March 29.

NVIDIA's next-gen Ada Lovelace GPU architecture is going to buck heads with AMD's next-gen RDNA 3 GPU architecture later this year, but now the rumors on the Ada Lovelace GPU side of things really heated up.

In his latest video, Tom @ Moore's Law is Dead said that the next-gen AD102 GPU -- and other GeForce RTX 40 series GPUs -- could support next-gen GDDR7 memory. Tom said "the memory controllers in Lovelace dies are vastly improved over Ampere, with AD102 being built to support GDDR7 memory speeds, and possibly GDDR7 itself".

The source continued, adding that the "AD102 memory controller was designed to handle GDDR7 memory speeds (32Gbps) and possibly built around an early 0.x GDDR7 specification. It is unlikely GDDR7 will be ready for Lovelace's initial launch, but either way NVIDIA will be able to use the fastest GDDR chips on the market in Q3. They will also have a pathway to a full refresh in 2023 with vastly higher memory bandwidth if they need to rush out an RTX 5000 series after defeat... just like Fermi...".

It looks like we'll be enjoying 3nm products sooner than we thought, with Taiwan Semiconductor Manufacturing Company (TSMC) seeing yield improvements with its N3e production.

The news is coming from Retired Engineer, who tweeted that with TSMC's new 3nm node seeing yield improvements that volume production on N3e may start in Q2 2023 (April-June). Morgan Stanley has reportedly checked with equipment vendors that have said TSMC could "freeze" the N3e process flow, meaning the design parameters -- kinda like TSMC is saying its N3e has 'Gone Gold' and is ready for production -- yeah, that works.

Retired Engineer explained: "N3e production yield improves; schedule being pulled in: Our recent checks with equipment vendors suggest that TSMC may freeze the N3e process flow sooner - by the end of this March. This means that volume production of N3e may start in 2Q23, around a quarter ahead of the original schedule of 3Q23".

NVIDIA has now halted sales of its products to Russia over the ongoing conflict in Ukraine, joining the likes of its competitors AMD and Intel.

An NVIDIA spokesperson confirmed the news to PCMag, but they didn't have any reasoning behind it. It's more of a "yeah, we're not selling our gear to them anymore" kind of thing. I'm not going to get into the politics of it all, as that's not our place to report on these things at TweakTown, but NVIDIA is a big deal -- so too is AMD and Intel -- and now, no more GeForce GPUs for Russian gamers.

The NVIDIA spokesperson said: "We are not selling into Russia".

Man, what an absolute mess the launch of NVIDIA's GeForce RTX 3090 Ti has been so far... the card was meant to be unleashed in late January, and now we're in the first week of March and it's still not here.

The latest on NVIDIA's new flagship GeForce RTX 3090 Ti graphics card is that the company will release it on March 29, just over three weeks from now. The reasons for the delay? Reportedly they've been memory-related... due to the 24GB of single-sided GDDR6X memory at 21Gbps.

There could be more hardware-related or even BIOS-level issues with the card... but with the amount of power driving through the card, Samsung's 8nm process not being the best (up against TSMC and its 7nm and lower nodes), and many other factors (the pandemic, supply chains in South Korea, etc).