Graphics Cards - Page 136

Palit has just unleashed its beautiful and very unique new ColorPOP graphics card series, with the introduction of the new Palit GeForce RTX 3060 Ti ColorPOP graphics card.

The new Palit GeForce RTX 3060 Ti ColorPOP is a full-custom design with a 2.7-slot cooling solution, with Palit changing things up by rebranding their GamingPro series to the new ColorPOP series. The changes are extensive: new colors, new stickers, and of course; new packaging.

What is special here with the Palit GeForce RTX 3060 Ti ColorPOP is that it features beautiful color-shifting paint dabbed on its shroud and backplate, where the card will flex between different colors including green, violet, and blue when you look at it from different angles and lighting conditions.

NVIDIA is currently preparing another flagship graphics card for a big unveiling in January 2022, with the purported GeForce RTX 3090 Ti graphics card that will rock Micron's even-more-ultra-fast 21Gbps GDDR6X memory.

The current flagship NVIDIA GeForce RTX 3090 arrives with 24GB of GDDR6X memory clocked at 19.5Gbps, but it easily overclocks to 21Gbps and barely offers any perceivable performance difference. But, the new GeForce RTX 3090 Ti will not only have upgraded GDDR6X memory (21Gbps vs 19.5Gbps) but with it, it'll have slightly more memory bandwidth at 1008GB/sec (1TB/sec) compared to 936GB/sec on the RTX 3090.

Inside, NVIDIA is expected to use the GA102-350 GPU which bumps up everything across the board: 84 GPU clusters (versus 82 on the GA102-300 inside of the RTX 3090) while there's 10752 CUDA cores (up from 10496 CUDA cores), 84 RT Cores (up from 82 RT Cores) and finally: 336 Tensors/TMUs (up from 328 Tensors/TMUs).

We've been hearing more and more about NVIDIA's new GeForce RTX 2060 12GB graphics card, a refreshed Turing-based RTX 2060 with 12GB of GDDR6 memory (up from the 6GB GDDR6 on the regular RTX 2060).

But it looks like it'll be more than just double the VRAM of a regular GeForce RTX 2060, as the new GeForce RTX 2060 12GB will reportedly have a higher TDP of 184W (with the RTX 2060 SUPER having a 175W TDP). This means that the RTX 2060 12GB will actually be around the same performance as the RTX 2060 SUPER.

The 12GB of GDDR6 memory on the new GeForce RTX 2060 12GB will be on a 192-bit memory bus, while the RTX 2060 SUPER has its 12GB of GDDR6 on a wider 256-bit memory bus. Not only that, but the overclocked AIB variants of the new GeForce RTX 2060 12GB could see GPU boost clocks at default 1650MHz and up to 1690MHz and beyond with OC.

NVIDIA is reportedly hearing up for the launch of the lower-end GeForce RTX 3050 graphics card in Q2 2022 according to the same rumors.

The rumors are coming from the same source that first broke that NVIDIA is working on a new GeForce RTX 2060 with 12GB of GDDR6 memory, while in between we've been hearing plenty of rumors about the new mobile-focused GeForce RTX 3050 Ti.

VideoCardz is reporting that the non-Ti desktop GeForce RTX 3050 should come with 8GB of memory, but this could change -- we could see a 4GB variant as well. As for performance, NVIDIA's purported GeForce RTX 3050 would be faster than the GeForce GTX 1660 SUPER but slower than the GeForce RTX 2060 12GB (which isn't here yet).

Intel has laid out its GPU roadmap right into 2025 and beyond, with what will start with the Intel 1st Gen Arc "Alchemist" GPU launching in Q2 2022 for the desktop and Q1 2022 for mobile.

But in a few years, we're going to see the Intel 4th Gen Arc "Druid" GPU arrive and that should be sometime in 2025. In a new tweet, Intel Arc Community Advocate and Driver Guru, @theBryceIsRt, replied to @Freeman2352 who joked and tweeted "2025: RTX 6090 Super | 800W".

Bryce replied, tweeting: "If all goes well, that could be when Intel ARC 'Druid' cards comes out" and ends it with adding "game on" with a controller emoticon and nice little wink tease. Interestingly, he used "ARC" in all caps which was recently noted to be "Arc" without the all-caps -- OCD to the side, the 4th Gen Arc "Druid" GPU should drop in 2025 which isn't tooooo far away.

AMD has unleashed its highest-end Radeon RX 6900 XT Liquid Cooled Edition graphics card for the DIY market in Europe.

The Radeon RX 6900 XT Liquid Cooled Edition graphics card sees AMD using a pre-binned Navi 21 XTXH GPU, with some of the very highest RDNA 2 GPU clocks in the wild. AMD explains: "AMD will be making a limited number of Radeon RX 6900 XT liquid-cooled reference design graphics cards available to select system integrators".

"System integrators are a critical part of the gaming ecosystem and offering this model through their channels helps ensure that as many of these products as possible get into gamers hands".

GPU shipments are a bit funny right now and we all know it, with analyst firm JPR (Jon Peddie Research) posting a new report that details the global GPU market has hit a milestone of 101 million units in Q3 2021 alone.

JPR also notes that GPU shipments increased by 9% year-over-year while discrete GPU numbers inside of PCs is expected to hit 31% over the next 5 years leading up to 3.24 billion GPUs by the end of 2025. AMD had its GPU market share increase by 1.4% in the three-month period, while NVIDIA GPU market share increased by 4.86% in the same quarter.

AMD GPU shipments dropped by a decent chunk, by 11.4% in Q3 2021 while NVIDIA had its GPU shipments increased by 8% for the quarter. The latest numbers have AMD with 17% of the discrete GPU shipment market share, and NVIDIA with 83% of the dGPU shipment market share.

The price of AMD Radeon RX 6000 series GPUs is about to be a little worse, with AMD reportedly ramping up prices of its RDNA 2 cards by 10%.

AMD's increased pricing on Radeon RX 6000 series GPUs is reportedly starting with the next batch of GPU shipments from AMD to its AIB (add-in board) partners. We should expect the new pricing to be reflected online and in-store in the coming weeks, right as we lead into the holidays.

There was a 9% increase in pricing of AMD's new Radeon RX 6000 series GPUs in November 2021 over October 2021, and now we have another 10% increase moving into December. TSMC wafer costs increasing are being blamed, so AMD has got no other move to make other than take the hit -- which isn't nice -- or boost pricing (also not nice).

NVIDIA has just announced its new RTX A4500 workstation GPU, which uses the GA102 GPU with 7168 CUDA cores and 20GB of GDDR6 memory.

The introduction of the new NVIDIA RTX A4500 workstation GPU offers its 7168 CUDA cores alongside 224 Tensor Cores, and 20GB of GDDR6 memory clocked at 16Gbps on a 320-bit memory bus which will deliver up to 640GB/sec of memory bandwidth.

NVIDIA says the new RTX A4500 workstation GPU will have 23.7 TFLOPs of FP32 compute performance, while it has 46.2 TFLOPs of RT performance. NVIDIA's new RTX A4500 workstation GPU has a 200W TDP, which is 15% lower than the RTX A5000 and 17% slower.

NVIDIA is reportedly preparing a beefed-up GeForce RTX 2060 release in January 2022, which will see the card be released with 12GB of VRAM -- an upgrade from the 6GB of RAM by default.

But before its announcement or release, Russian YouTuber and modder VIK-on has upgraded his GeForce RTX 2060 6GB to a GeForce RTX 2060 12GB. The original GeForce RTX 2060 launched with the cut-down Turing-based TU106 GPU, rocking 1920 CUDA cores, and 6GB of GDDR5 memory on a 192-bit memory bus, clocked at 14Gbps.

NVIDIA is expected to make the GeForce RTX 2060 with 12GB official on December 7, with VIK-on testing the modded RTX 2060 with 12GB of VRAM in crypto mining and Unigine Superposition at first with roughly the same performance as the 6GB model -- duh. We should expect more official news and benchmarks in the coming weeks.

GALAX has just introduced the new GeForce RTX 3060 Metaltop Mini [FG] graphics card, a new GeForce RTX 3060-based graphics card on the super-small Mini-ITX form factor.

The new GALAX GeForce RTX 3060 Metaltop Mini [FG] graphics card rocks a single-fan design, while the "FG" in the name stands for "For Gamers". GALAX thinks by adding "FG" and it being "For Gamers" that the big bad crypto miners won't buy these... hah.

As for the measurements, we're looking at the card measuring in at just 16.8cm in length, 11.6cm in width, and 4cm in height. The single-fan design means it is nice and small and supports 0dB technology which means the single fan will disable when the GPU load is low enough (not gaming).

Meta today announced that the Mercenaries DLC package for Resident Evil 4 is officially in development for the Meta Quest. Along with that announcement, the company revealed that RE4 is the fastest-selling game on the platform. It's almost as if people really want AAA games in VR.

Capcom and Oculus (now Meta) released the Quest version of Resident Evil 4 on October 21st, and in just over a month, the game has become the fastest-selling title in the Quest platform's history. Meta has never released sales numbers for any of its headsets or games, but we know several titles have made millions of dollars in profits. Resident Evil isn't likely in the top-selling ranks yet, but it will be if the high sales volume continues. Beat Saber and Population: One have big head starts, though.

To keep the excitement about RE4 VR going, Capcom is working on a DLC expansion for the Quest title. The Mercenaries expansion will launch next year and let you experience more of Resident Evil in terrifying first-person VR. The Mercenaries expansion adds a hoard-mode experience to the game, where you need to fight off a never-ending hoard of Ganados until your rescue helicopter arrives.

AMD has a couple of new graphics cards are coming: the Radeon RX 6500M and Radeon RX 6300M, which will both feature the Navi 24 GPU.

We have heard about the Navi 24-powered GPUs before from Tom @ Moore's Law is Dead, with the entry-level Navi 24 GPU taking care of at least 6 new graphics cards from AMD. We should see the Radeon PRO W6500M, Radeon PRO 6300M, and then gaming SKUs in the Radeon RX 6500M and Radeon RX 6300M.

These 4 graphics cards are expected to join the already-known Radeon RX 6500 XT and Radeon RX 6400 series desktop graphics cards -- also powered by the same Navi 24 GPU. What we do know about the Navi 24 GPU is that it has a PCI Device ID of 7420-743F which has been spotted in Linux kernel/driver leaks previously.

It's been a couple of hot weeks of rumors and news about Intel's new Arc Alchemist family of GPUs, with Tom @ Moore's Law is Dead has unveiled new details about the Intel DG2-128EU GPU.

We should see Intel's new DG2-128EU GPU in desktop form with the Arc Alchemist, so expect 6GB of GDDR6 memory and a 96-bit memory bus. This beats out previous rumors of a 64-bit memory bus, but Tom also notes that we could see a 4GB GDDR6 variant, as well as a 64-bit mobile Arc Alchemist SKU in the future.

Tom expects the DG2-128EU GPU will have GPU clocks of between 2.2GHz and 2.5GHz, with a 75W TDP while Tom "can't promise they won't overclock it for desktop like Navi 24, but so far everything is low power". With the notes on power consumption, Tom continues: "with such low power, and 6GB of VRAM instead of 8GB, this thing can cost $179 or less with decent margins".

Intel's new flagship Arc Alchemist GPU has been spotted flexing some benchmark muscle on Geekbench, but with low GPU clocks and nowhere near a final driver, this isn't going to beat AMD or NVIDIA right now.

The GPU was spotted with detailed JSON data with 4 benchmark runs, with the GPU itself described as "OpenCL HD Graphics" or "Intel Xe Graphics". This GPU has 512 Execution Units and 12.7GB of VRAM, while the GPU was clocked at 1329MHz which from its max GPU clock of 2.1GHz is 63% lower.

Intel looks like it was testing the Arc Alchemist on an Intel Coffee Lake CPU, with an Intel Core i5-9600K processor, and not their latest 12th Gen Core "Alder Lake" CPU. But rather multiple generations old CPU and motherboard hardware are being used to test the very latest GPU.

Innosilicon Technology has just announced its new "high-performance GPU for servers" with the introduction of the Fenghua No.1 graphics card.

There's no detailed specifications of the Innosilicon Technology Fenghua No.1 GPU, but it will pack high-end GDDR6X memory, PCIe 4.0 support, and HDMI 2.1 connectivity with e-DisplayPort 1.4 thrown in. Fenghua No.1 having GDDR6X memory is a big deal as NVIDIA is the only one using it on their highest-end GPUs, while AMD uses regular GDDR6 memory.

We don't know how much GDDR6X memory the Fenghua No.1 GPU will have, nor what bandwidth it'll be cranking along with but we should expect it to be at 21Gbps. Innosilison Technology explains: "It is reported that "Fenghua No. 1" GPU is the industry's first rendering GPU that adopts the world's top GDDR6X high-bandwidth video memory technology and cutting-edge technologies such as Innolink chiplet".

NVIDIA has exclusivity over Micron's ultra-fast GDDR6X memory which debuted at 19.5Gbps and can crank up to 21Gbps and (a little) beyond, but now Samsung has teased its next-gen GDDR6X+ and totally next-gen GDDR7 memory.

Samsung's new GDDR7 memory was teased at the Samsung Tech Day 2021 event, but no slides or photos from the event have surfaced because it was stated to not do so in Samsung's NDA. ComputerBase covered the event since Samsung executives and engineering teams had some details to share.

The new GDDR7 memory announced only has two details so far: 32Gbps of bandwidth, and a new "real-time error protection feature". That's all we know, so with what we know let's decode some things. Micron has GDDR6X memory which comes in two flavors: 19.5Gbps and 21Gbps for NVIDIA's very fastest GeForce RTX 3080, GeForce RTX 3080 Ti, and flagship GeForce RTX 4090.

NVIDIA is about to have more competition than it has ever had in the GPU space in the history of computing, with AMD launching the industry's first MCM-based (multi-chip module) GPU that will offer an astounding amount of performance.

But never fear... NVIDIA founder and CEO Jensen Huang isn't scared, he's prepared. Huang was interviewed by The Next Platform recently, with Timothy Prickett Morgan talking to Huang about a few different things but one of the main question was that AMD is "much more competitive" with its MCM-based "Aldebaran" GPU.

Morgan asked: "Obviously, AMD is much more competitive with its "Aldebaran" Instinct MI200 series GPU accelerators than it has ever been. It is really two GPUs, not one, and I reminded everyone that AMD had "pulled a K80" by putting two GPUs on one device, but nonetheless, this GPU has won two exascale-class systems and many more smaller systems. I realize that there will not be new GPU announcements from NVIDIA until next year, based on the cadence, but what is your response to this competition from AMD, and soon, to a lesser extent, from Intel in the GPU compute arena?"

NVIDIA is making some big shifts in the image upscaling world this week with the introduction of both DLSS 2.3 and ICAT, as well as the announcement of making its Image Scaling technology open-source.

NVIDIA has had Image Scaling inside of its drivers for many years now, but the move into open source means it's easier for developers to use and will act separate to DLSS (Deep Learning Super Sampling). Where DLSS requires a specific NVIDIA GPU, NVIDIA Image Scaling will work with all games and virtually on GPUs.

With the new GeForce Game Ready 496.76 WHQL drivers you'll have access to the NVIDIA Image Scaling feature through the NVIDIA Control Panel, or through GeForce Experience. You can tweak per-game sharpening from within NVIDIA's in-game overlay, too.

NVIDIA's new Ada Lovelace GPU architecture and next-gen GeForce RTX 40 series graphics cards are going to be monsters, with new rumors on the flagship AD102 GPU which would power the new flagship GeForce RTX 4090.

The new leaks are from Greymon55 on Twitter, who said the new AD102 GPU would be made on 5nm TSMC with 18,432 CUDA cores with GPU clocks somewhere in the 2.3GHz to 2.5GHz range. FP32 compute performance would be gigantic, topping somewhere around 85 TFLOPs to 92 TFLOPs -- which when compared to the RTX 3090 and the Ampere GA102 inside -- has 35.5 TFLOPs, then we're talking 2.5x the performance of the RTX 3090.

NVIDIA would use the same 24GB of GDDR6X but clocked at 21Gbps on a 384-bit memory bus, the newer PCIe 5.0 x16 connector and a TDP range of a yeah-you're-going-to-need-to-buy-a-new-PSU-probably-maybe 450W to 650W of power. Previous rumors have pegged power consumption in the 500W+ range, but if we have a card that is over twice the power of the RTX 3090... then bring if the hell on.