Graphics Cards - Page 146

NVIDIA's new Ada Lovelace GPU architecture has been teased again, with the very latest rumor on the AD102 GPU is that it will have clock speeds that will hit 2.2GHz, and above.

The new rumors have come from @greymon55 on Twitter, who has said that AD102 will have its GPU frequency above 2.2GHz and that it will arrive on the 5nm node, and have its GDDR6X memory on a 384-bit memory bus. This should give the AD102-powered GeForce RTX 4090 a heck of a lot of GPU horsepower, and memory bandwidth.

We're looking at 18432 CUDA cores on the latest specs rumors for AD102, with 80 TFLOPs+ of compute performance that would blow away the 36 TFLOPs of compute performance that the GA102-based GeForce RTX 3090 is capable of. It won't be cheap, with rumors teasing the AD102-powered GeForce RTX 4090 could cost $2999... but you'll probably be able to Your next TV should be this huge new 85-inch 4K 240Hz... yeah, 240FPS.

XFX has launched its Radeon RX 6900 XT ZERO WB graphics card, which rocks an ultra-premium pre-installed water block from the good folks over at EK.

The new XFX Radeon RX 6900 XT ZERO WB graphics card is one of the very fastest custom Radeon RX 6900 XTs ever made, with 350W+ power available and promised 3GHz+ OC GPU clocks. XFX explains: "The base of the block is CNC machined out of nickel-plated electrolytic copper while the top is CNC machined out of glass-like Acrylic. Sealing is ensured by high-quality EPDM O-Rings. The brass standoffs are already pre-installed and allow for a safe and easy installation procedure".

XFX requires 3 x 8-pin PCIe power connectors on the custom Radeon RX 6900 XT ZERO WB graphics card, we don't know how much power the card can use at its maximum -- but it should be 350W+ while staying cooler than air-cooled RX 6900 XT cards using the same Navi 21 XTXH GPU.

Apple has been using dual Radeon GPU graphics cards in its highest-end Mac Pro systems for a couple of years now, with the latest AMD Radeon PRO W6800X Duo graphics card now benchmarked and man, is it an utter compute beast.

First off, the AMD Radeon PRO W6800W Duo graphics card features 2 x Navi 21 GPUs each with 60 compute units and 3840 stream processors. Each GPU has 32GB of GDDR6 memory for a total of 64GB of GDDR6 memory with 512GB/sec of memory bandwidth.

There's 30.2 TFLOPs of single-precision compute performance, and a whopping 60.4 TFLOPs of half-precision compute performance. There's an onboard Infinity Fabric Link connection that connects the 2 x W6800X GPUs together, pumping 84GB/sec in each direction. Then there's an external Infinity Fabric Link connection that enables 2 x W6800X Duo modules to connect 4 x W6800X GPUs.

I'm a huge fan of NVIDIA's kick-ass DLSS (Deep Learning Super Sampling) technology, especially in its DLSS 2.0+ form, but now we're going to be seeing more of DLAA (not a typo, that is Deep Learning Anti-Aliasing) which is going to first appear in The Elder Scrolls Online.

Zenimax unveiled the news during their recent ESO live stream, where we found more information out about the upcoming Deadlands zone DLC with base Update 32. The new patch will add DLSS to The Elder Scrolls Online, which will help out lower-end GPUs and those chasing 120FPS+ but what about high-end GPUs that want visual quality?

Where DLSS upscales the resolution, DLAA will offer "incredible anti-aliasing". ESO Creative Director Rich Lambert explains: "While we were working on adding NVIDIA DLSS, we also worked with them on some new tech that we're going to be the first game that's ever done this before. This is debuting their new tech, which is called NVIDIA DLAA. It's the same kind of concept, you won't get a performance boost out of this but what you get is absolutely incredible anti-aliasing. It's unbelievable, it's crazy how good it is".

AMD has caught up considerably against NVIDIA with its new RDNA 2 graphics architecture, with the Radeon RX 6800 XT and Radeon RX 6900 XT both kick-ass graphics cards -- but man-oh-man, the RDNA 3 architecture is going to be delicious.

We have already heard rumors of the RDNA 3 GPU and its flagship Radeon RX 7900 XT powered by the Navi 31 GPU and shaping up to be a beast. Now we have Tom from Moore's Law is Dead teasing that RDNA 3 will "improve performance by 60% at a minimum" with multiple sources telling him that the Radeon RX 7900 XT will be "roughly double" the RX 6900 XT.

Tom does note that people need remember that "theoretical top performance, and real-world performance are not the same thing". Tom continues, adding that "just because the top model performs a certain way in a simulation, does not mean it will in practice once its actually fabricated in silicon, packaged, and requires cooling".

NVIDIA founder and CEO Jensen Huang has made the list on TIME's 100 most influential people of 2021.

Huang explains that it all started with GPUs back in 2003 with the introduction of the GeForce 256, which resulted in me changing my nickname -- and it still is to that day, to anthony256 -- and moved into the world of AI and deep learning where they are the undisputed champion.

This worked for NVIDIA, where they shifted from being known for only making gaming GPUs to the world of supercomputers, AI, deep learning, and more. These advancements are driven by Huang through NVIDIA, and TIME is recognizing that by donning him one of the most influential people of 2021.

China is stepping up in the GPU department with Jingjia Micro announcing two new GPUs made in China, the first domestically-produced GPU out of the country and it's as fast as NVIDIA's GeForce GTX 1080.

Not only that, but the new Jingjia Micro JM9 Serries GPU is as fast as the Pascal-based NVIDIA GeForce GTX 1080 and the Vega-based AMD Radeon RX Vega 64 graphics cards. Jingjia Micro has two chips in its series, the entry-level "JM9231" which has GTX 1050 levels of performance, while the "JM9271" has performance closer to the GTX 1080.

Jingjia Micro has now taped out its first JM9 Series GPU -- after what should've been a launch in 2020 becomes a launch in Q3 2021 -- but hey, it's nearly here. There's still a bunch of things the GPU needs to go through -- with no test work, mass production, or external sales just yet -- that could take another year. China can say whatever it wants, I want to see it in action -- not just some PR.

We've already heard whispers that Intel's upcoming Xe-based ARC Alchemist gaming GPU will fight NVIDIA's GeForce RTX 3070 and AMD's Radeon RX 6700 XT -- now we have some purported leaked slides with similar claims of performance.

The new slides tease that the Intel ARC Alchemist will be in the upper mid-range GPU market, with SOC1 having 175-225W of power at its disposal and gets compared to NVIDIA's GeForce RTX 3070 and RTX 3060 Ti as well as AMD's Radeon RX 6700 XT and RX 6600 XT graphics cards.

Intel will have a lower-end SOC2 chip that would have up to 75W TDP making it compatible to the GeForce GTX 1650 SUPER, which will be priced in the $100 to $200 market. The higher-end Intel "SOC1" ARC Alchemist should be priced somewhere in the $300 to $499 market.

If you thought it was painful paying the inflated prices of graphics cards today, then this new rumor on NVIDIA's next-gen GeForce RTX 40 series pricing isn't going to make you feel any better.

YouTuber "Graphically Challenged" is predicting the MSRP prices of NVIDIA's next-gen Ada Lovelace-based GeForce RTX 40 series graphics cards, with the flagship GeForce RTX 4090 costing $2999 and the GeForce RTX 4080 Ti under that costing $1999.

The pricing situation doesn't stop there, with the RTX 4080 predicted to cost $1199 and ends at the RTX 4050 with a predicted price of $279. In between we have the RTX 4050 Ti at $329, RTX 4060 at $399, RTX 4060 Ti at $499, and RTX 4070 at $799.

In the middle of the whirlwind rumors of a beefed-up GeForce RTX 3090 SUPER graphics card, and even plenty of rumors on the next-gen Ada Lovelace GPU architecture -- but prepare for a Turing-based GeForce RTX 2060 with 12GB of memory. Yeah. That's right.

The new GeForce RTX 2060 would be released with 12GB of GDDR6 memory, with the Turing-based PG161 GPU -- the regular RTX 2060 had 6GB of GDDR6 memory and used the PG160 GPU. Rumor has it that AIB partners have been told to expect the new RTX 2060 12GB by the end of the year, possibly January depending on manufacturing.

As VideoCardz depressingly points out, this is a "grim outlook for mid-range GPU availability" given that NVIDIA has to re-release a previous-gen GPU with double the VRAM for the mid-range market. This new card would sit somewhere above the RTX 2060 6GB and below the RTX 3060, meanwhile Intel will be striking that market -- and above -- with its Xe-based ARC Alchemist graphics card in 2022 and AMD with the Radeon RX 6600 XT and Radeon RX 6700.

You kinda need thermal pads on the thermal pads of the GDDR6X memory of your new GeForce RTX 3080 Ti, or GeForce RTX 3080 graphics card -- but two Redditors have discovered the lack of thermal pads on their new cards.

One of them was the ASUS GeForce RTX 3080 Ti TUF Gaming OC graphics card which Reddit user 'kamaloo922' saw his 12GB of GDDR6X memory running at up to 110C. He put some thermal pads down on the GDDR6X memory modules and the temps dropped down to 80C.

Another Redditor found that the GDDR6X memory modules on his NVIDIA GeForce RTX 3080 Founders Edition were misplaced, as well as the thermal pad on the GPU itself -- the GPU temps dropped from 76C to just 61C.

NVIDIA is gearing up for a big GPU refresh release in the coming months, with rumors that the company will be launching their new GeForce RTX 30 SUPER series in January 2022.

The company could launch both mobile and desktop variants of its new GeForce RTX 30 SUPER graphics cards, and with multiple teases now of the GeForce RTX 3080 Ti with 20GB of GDDR6X floating around -- it would be interesting to see if the RTX 3080 Ti 20GB gets repurposed (refreshed) into the RTX 3080 SUPER.

NVIDIA should also release a new flagship GeForce RTX 3090 SUPER which will be very interesting too, as we've actually been hearing more and more about the GeForce RTX 3090 SUPER. The latest on the RTX 3090 SUPER is that it would use the GA102-350-A1 GPU, the same 21Gbps GDDR6X memory, and have a large 450W TGP.

AMD is shifting into the world of GPU chiplets with its future Instinct MI300 GPU, which is reportedly set to have 4 GCD (Graphics Complex Die) according to a new leak.

The use of MCM or MCD (multi-chip module or multi-chip die) on AMD's next-gen Instinct accelerator is going to be a big deal, where we should see the upcoming Instinct MI200 accelerator packing either 110 CUs or 220 CUs (2 x 110 or 2 x 55 CUs).

But when it comes to the next-gen MI300 accelerator, the new leaks tease is that AMD will use twice as many GPU dies inside of the MI300 over the MI200 -- so we're looking at a huge 440 CUs in total from the 4 graphics tiles/chiplets.

Intel is coming back into the graphics card business with so much hype that if they failed to deliver at this point, it would be embarrassing -- given how badly it has been struggling to combat AMD with its Zen moment and the launch of the Ryzen CPU family.

We know that Intel's new Xe GPU architecture will power the ARC Alchemist Xe-HPG graphics card when it launches in 2022, but now Intel CEO Pat Gelsinger said that the company would be fighting against NVIDIA in a big way, with Gelsinger providing the juicy statement during an interview with Pierre Ferragu.

Ferragu asked if there would be a "Zen moment for Intel" to which Gelsinger replied:

NVIDIA has officially killed support for its Kepler GPU architecture with its 500 series drivers, after confirming the news a few months ago -- it's now official with the new NVIDIA 510.06 WSL driver.

The new NVIDIA 510.06 WSL (Windows Subsystem for Linux) driver has been released to Windows 11 users through the Microsoft Insiders Program, with the 894MB download having no changes to both the installer, or the control center.

Maxwell and Pascal GPUs are still supported with the GeForce GTX 900 series and GeForce GTX 10 series GPUs, but the Kepler GPUs under the GeForce GTX 700 series are gone. VideoCardz notes that there are no changes to the control panel, which is really disappointing -- but I'm hoping things change between now, and the final release of the 500 series driver.

Back in July it was being reported that Amazon's new game "New World" was bricking EVGA GeForce RTX 3090 graphics cards, and once the company had multiple RTX 3090 cards returned, investigations began.

The cards were ramping up to huge frame rates in games, where EVGA RTX 3090s were being killed from Amazon's new game and no one knew why. But now we know: it was a "rare soldering issue" that EVGA found after it performed an X-Ray analysis of the killed cards.

EVGA says that it was only the cards made in 2020 that were affected, so any of the GeForce RTX 3090s made by EVGA in 2021 should be fine. PC World's Gordon Mah Ung explains:

Intel has just teased the first official artwork for its ARC Alchemist gaming graphics cards, teasing the artwork with our friends over at HotHardware. Check it out:

Intel Alchemist is the codename of Intel's first Xe-HPG gaming architecture, with the Alchemist being one of the characters from tabletop role-playing and RPG games. HotHardware explains the Alchemist: "Alchemists rely on their intuition and knowledge of potions and metallurgy to craft powerful enhancers from simple elements, herbs, and other crafting materials, and play a support role in just about every game in which they're found".

Better yet, in Final Fantasy XIV the Alchemist is pretty much me IRL... the alchemist are "typically reclusive loners who tinker away in their homes and laboratories endlessly". Sounds not just like me, but most of the tech review community who love playing and tinkering with GPUs... like the team that Raja Koduri has built inside of Intel. In Final Fantasy XIV, the alchemist creates potions you can consume -- and for your party in the game, as well as wands for the magic characters.

If you thought your GPU was safe from malware then you'd be mistaken, according to a new report from Bleeding Computer, malware that can execute code from your GPU is now a reality.

It seems that a proof-of-concept (PoC) was up for sale on a hacker forum and allows hackers to keep malicious code stored in your GPU memory buffer, which stops malware code from being scanned by security software when it scans the system RAM.

The seller of the GPU-focused malware says that it only works on Windows systems with support for version 2.0 and above of the OpenCL framework. This is required for executing the malware on the GPU, with the hacker testing it on Intel, AMD, and NVIDIA GPUs.

NVIDIA has just released its new NVIDIA GeForce Game Ready 471.96 WHQL drivers, which add the usual Game Ready performance tweaks to games but also include support for 6 new G-SYNC Compatible displays.

When it comes to GeForce Experience's one-click optimal settings, you can instantly configure game options for the specific hardware inside of your gaming PC. There are over 1000 games supported with GFE's one-click optimal settings feature, with 24 more games now added (I've got those listed below).

NVIDIA has also added support for 6 new G-SYNC Compatible displays, with the 48-inch Philips OLED806 TV and 77-inch Xiaomi 077M8-MAS TV being the latest BFGD (Big Format Gaming Displays) validated as G-SYNC Compatible. There's also support for EVE's new Spectrum ES07D03 which I just reviewed.

We've been hearing more and more about AMD's highly-tweaked Radeon RX 6900 XTX graphics card -- with the Radeon RX 6900 XTX teased in a water-cooled variant... and here it is again.

The purported AMD Radeon RX 6900 XTX graphics card would have a higher-binned Navi 21 XTXH GPU, higher-clocked 18-18.5Gbps GDDR6 memory, and it would be faster than NVIDIA's flagship Ampere-based GeForce RTX 3090.

The new rumor is coming from CyberPunkCat on Twitter, with 3DCenter.org quick to reply by saying that the Radeon RX 6900 XTX and its 24.93 TFLOPs of compute performance points perfectly to the boost clock of the Radeon RX 6900 XT Liquid Cooling GPU. 3DCenter.org added: "The 24.93 TFLops points (perfectly) to the boost clock of 6900XT "Liquid Cooling". This card is sometimes reported as "18 Gbps" instead of the real 18.5 Gbps. So, maybe just an old slide - back in the days, when AMD was unsure for how to release this LC card?"