Graphics Cards - Page 138

MSI has made some tweaks to its custom Radeon RX 6900 XT and Radeon RX 6800 XT GAMING Z TRIO and GAMING PLUS TRIO graphics cards, which include some boosted GPU clock speeds.

The new MSI Radeon RX 6900 XT GAMING Z TRIO will be the fastest custom RDNA 2-based graphics card on the market when it drops, with the Navi 21 GPU boosting up to 2425MHz. This is 60MHz higher than the 2365MHz GPU clock on MSI's custom Radeon RX 6900 XT GAMING X TRIO.

MSI's other Radeon RX 6900 XT cards include the GAMING TRIO, GAMING X TRIO, and then the new GAMING TRIO PLUS and GAMING Z TRIO cards. As for GPU clocks, we're looking at 2250MHz for the GAMING TRIO, 2340MHz for the GAMING X TRIO, while the new GAMING TRIO PLUS hits 2365MHz and the new GAMING Z TRIO at the top with 2425MHz.

EVGA has just teased its monster new GeForce RTX 3090 Ti KINGPIN graphics card, with the company's in-house overclocked Vince "KINGPIN" Lucido teasing some photos of the card.

The new custom EVGA GeForce RTX 3090 Ti KINGPIN graphics card features a dual-tone black/silver color scheme and a beautiful OLED panel (that flips up, by the way). EVGA is using a hybrid cooling design, with a huge fan that cools the pure copper heat sink that's joined by a 360mm AIO radiator.

There's a beautiful PCB shot that shows off the updated design that will power the EVGA GeForce RTX 3090 Ti KINGPIN graphics card, with the beautiful gold traces and brand new power management system. EVGA uses a triple BIOS that users will get "Normal", "OC", and "LN2" profiles while PROBEBIT support is also included. This lets users analyze multiple GPU power metrics, all at the same time.

NVIDIA's next-gen Hopper GPU architecture has been teased once again, with the flagship GH100 GPU reportedly going to be an absolute silicon monster that measures in at up to a huge 1000mm² die.

The news is coming from leaker "kopite7kimi" on Twitter, where kopite7kimi tweeted: "GH100 has a huge single die of slightly less than 1000mm²" with a follow-up tweet adding "DO NOT overreact. We know the actual die size will be significantly smaller than the package size. But GH100>GA100 is confirmed".

Up until now, it was thought that NVIDIA's new GH100 Hopper GPU would be a multi-chip module (MCM) design, like AMD's upcoming RDNA 3-powered Radeon RX 7000 series GPUs -- but no, it seems Hopper GH100 will be a monster single-die design -- while GH102 will reportedly be an MCM design.

Intel's new Arc Alchemist discrete GPUs have been spotted again, with the new DG2-384 GPU spotted with 6 memory modules labeled as "Samsung K4ZAF325BM-HC16".

We should expect each of these GDDR6 memory modules to be 2GB a pop, clocked at 16Gbps -- which means with the 384 EU variant of the Arc Alchemist packing 12GB of GDDR6 @ 16Gbps on a 192-bit memory bus -- resulting in around 384GB/sec of memory bandwidth.

This is the second Arc Alchemist GPU, aka "SKU2" or the 384 EU variant -- packing 12GB GDDR6 @ 16Gbps -- while the flagship "SKU1" or the 512 EU variant, packs 16GB of GDDR6 (8 x 2GB modules) at 16Gbps on a 256-bit memory bus for up to 512GB/sec of memory bandwidth.

GPU shortages have been around for so long now that it feels like they'll never end, but it looks like there might be some silicon light at the end of the tunnel according to new reports out of DigiTimes.

The site reports that industry leaders are expecting a big shift in the GPU market in the summer of 2022, after some supply issues are worked out in between. Ajinomoto Fine-Techno Company is one of the key companies in the middle of the shortage chain, where they make Ajinomoto build-up films substrates -- or AFB substrates -- which is used on most GPUs, CPUs, chips, automotive processors, and so much more.

Once the company was completely flooded with orders, they couldn't make enough and it became a gigantic thorn in the side of many of the largest tech companies on the planet. Now we have AMD and Intel both promising that we can expect the companies to move into other ways of having this not happen again, and expand what they have on the market for consumers.

Intel will reportedly have 5 different SKUs of its Arc Alchemist GPU in mobile form, with 4 different VRAM configurations that top out at 16GB GDDR6 memory.

The flagship Intel Arc Alchemist mobile GPU offering will have 512 EUs and is known as "SKU1" with 8 x GDDR6 memory modules for a total of 16GB GDDR6 memory. The 16GB of GDDR6 memory will be clocked at 16Gbps, and placed on a 256-bit memory interface, offering up to 512GB/sec of memory bandwidth.

The next one on the list is "SKU2" with 384 EUs and will have 6 x GDDR6 memory modules for a total of 12GB of GDDR6 at the same 16Gbps, but on a smaller 192-bit memory bus that results in 384GB/sec of memory bandwidth. "SKU3" packs 256 EUs and 4 x GDDR6 memory modules at 16Gbps on a 128-bit memory bus that has up to 256GB/sec of memory bandwidth, while "SKU4" and "SKU5" will have 128 EUs and 96 EUs respectively, but will share 2 x GDDR6 memory modules at up to 14Gbps, but only 4GB GDDR6.

NVIDIA's new flagship GeForce RTX 3090 Ti graphics card is reportedly right around the corner, with the suped-up card packing the GA102 GPU -- the flagship GPU and the fastest gaming GPU that NVIDIA has ever created -- alongside 24GB of ultra-fast GDDR6X memory.

The new GeForce RTX 3090 Ti continues on the GPU dominance that NVIDIA built with its current flagship RTX 3090, offering the same but also changed 24GB of GDDR6X memory. We still have the same 24GB of GDDR6X but it is both faster -- clocked at 21Gbps -- and has a much wider 384-bit memory bus that for the first time ever, has a GeForce GPU pushing over 1TB/sec of memory bandwidth.

NVIDIA is using the GA102 GPU with 84 SM units that provide a huge 10,752 CUDA cores -- up from the 82 SM units and 10496 CUDA cores on the RTX 3090 -- but there's also next-gen RT (Ray-Tracing) cores, Tensor cores, and brand new SM (streaming multi-processor units). The GA102 GPU inside runs at a base GPU clock of 1560MHz, while boost GPU clocks hit 1860MHz. The TDP has been massively increased at 100W

AMD has launched its new Navi 24 GPU with barely a whimper, powering the new Radeon RX 6500 XT -- of which I'll have my review up soon, my card only arrived a few hours ago -- and now an AMD staffer has cleared up how the company will be deploying its Navi 24 GPU.

AMD Linux NPI SW Architect John Bridgman explains: "The primary use of Navi 24 will be in laptops paired with a Rembrandt APU, which has full video functionality and Gen4 PCIe. My impression was that it was just encode that was limited in Navi 24, not decode - still not sure if that limitation is real or just a typo on the product page. Trying to find out a definitive answer".

The reason Bridgman addressed the video functionality on Navi 24 is that the GPU itself doesn't support H264/265/AV1 encoding -- which is a big deal, something all Navi 24 GPUs -- the workstation-focused GPUs in the PRO W6400, W6500M, and W6300M, and the new W6400.

AMD has officially launched its new low-end workstation GPU with the introduction of the Radeon PRO W6400, which is based on the Navi 24 GPU and packs 4GB of GDDR6 memory.

The new AMD Radeon PRO W6400 workstation GPU is based on the Navi 24 GPU which is made on the 6nm TSMC process node, the first in workstation GPU on 6nm, with 4GB of GDDR6 on a 64-bit memory bus. The Navi 24 GPU has 768 Stream Processors, so you won't be taking this into your work at Pixar and booting up your dailies, but the Radeon PRO W6400 is here.

AMD is only providing 4 x PCIe 4.0 lanes wired to the GPU, even though the Radeon PRO W6400 workstation card arrives with a full-sized PCIe 4.0 slot. The new Navi 24-based Radeon PRO W6400 workstation GPU slots under the higher-end Apple Max exclusive Radeon PRO W6000X series -- as well as the Navi 21-powered W6800, Navi 23-based W6600, and Navi 24-based W6400 -- the new Navi 24 GPU skips over the Navi 22, which we might eventually see later in the year.

AMD has been enjoying filling computers, consoles, servers, and supercomputers of the world with its various CPU and GPU designs on 7nm thanks to TSMC, while NVIDIA hobbles along with Samsung 8nm... but the floodgates of 5nm are opening later this year.

But it looks like we can expect Chinese GPU manufacturers to have both 5nm and 7nm GPU designs on the market sometime in 2022, according to sources of DigiTimes. Their sources said that Innosilicon -- which works with Imagination Technologies IP, Changsha Jingjia Microelectronics, and Biren Technology -- all make GPUs in China.

We've already been reporting on Innosilicon Fantasy One GPU, which the company offers in an impressive dual-GPU design with up to 32GB of GDDR6X memory (16GB per GPU). Performance is only hitting the levels of NVIDIA's previous-gen mid-range GeForce RTX 2070 graphics card. But Intel's upcoming Arc Alchemist is around the corner, and that RTX 2070 level performance is actually opening the door to China competing with Intel this year.

You know how hard it is to buy not just graphics cards these days, but DDR5 RAM, high-speed SSDs, high-end PSUs, next-gen consoles, and so much more are next to impossible to buy in some countries... and in roll GPU loot boxes.

Yeah, you read that right -- a Japanese Amazon reseller is offering customers "mystery boxes" that will include a new, or used graphics card of some description. You will pay $100 for the mystery box, and not know what you're going to buy. You might spend $100 and end up with an older-gen NVIDIA Maxwell-based GPU, or an AMD Radeon RX 300-series GPU.

But... you might end up with a flagship NVIDIA GeForce RTX 3090, RTX 3080 Ti, RTX 3080 Ti, RTX 3070 Ti, RTX 3060 Ti -- or in AMD's Team Red camp you could score yourself -- for $100 mind you, a Radeon RX 6900 XT, RX 6800, RX 6700 XT, RX 6600 XT, and many others. The Ampere and RDNA 2-based GeForce RTX 30 and Radeon RX 6000 series GPU, respectively, will be brand new -- the others, will be used cards. Not a bad gamble for $100, isn't it?

Update: SK hynix reached out to tech media to clear the air about its HBM3 detailing at ISSCC 2022, and that its HBM3 @ 896GB/sec is the same memory as the one the company unveiled back in October 2021 but with a different "standard of performance indication".

SK hynix will be detailing its latest under-development HBM3 memory spec at the ISSCC 2022 (IEEE International Solid-State Circuits Conference) with up to 24GB through 12-layer HBM3 DRAM. With capacities of up to 196Gb (24GB) the company is using TSV (through silicon via) auto-calibration and machine learning optimizations.

We know what to expect from the title of the session SK hynix is involved in is dubbed "A 192-Gb 12-High 896GB/s HBM3 DRAM with a TSV Auto-Calibration Scheme and Machine-Learning-Based Layout Optimization". FYI: "TSV" is through silicon via, but we don't know if this is prototype memory or something that will enter mass production and see SK hynix printing it onto devices en masse.

AMD looks to be preparing a new Radeon RX 6000 series GPU refresh with faster 18Gbps GDDR6 memory, with a purported Radeon RX 6950 XT, Radeon RX 6850 XT, and Radeon RX 6750 XT graphics cards.

New information is coming from leaker "Wjm47196" that posts on Chiphell, where it is teased that the Radeon RX 6x50 XT model graphics cards will be powered by the Navi 21 and Navi 22 GPUs and might "replace" the existing Radeon RX 6000 series GPUs.

The new Radeon RX 6000 series GPU refresh would have a larger upgrade in the form of faster 18Gbps GDDR6 memory, which is already deployed on AMD's flagship Radeon RX 6900 XT Liquid Cooled Edition -- oh, and AMD's upcoming mid-range Radeon RX 6500 XT. Samsung recently unveiled its spiffy new 20Gbps and 24Gbps GDDR6 memory modules, where AMD could be using this faster GDDR6 and downclocking it to 18Gbps.

NVIDIA has just enabled GPU Select Processor (GSP) on certain GPUs with its new 510.39 driver, with select Tesla and Data Center GPUs based on the Ampere and Turing GPU architectures.

GSP will offload tasks that are otherwise handled by the CPU, over to the GPU -- these tasks include GPU utilization and management tasks -- but you can still force-disable GSP if you want. You will lose some display-related and power management features, but those features aren't built into GSP at this time.

NVIDIA explains: "Some GPUs include a GPU System Processor (GSP) which can be used to offload GPU initialization and management tasks. This processor is driven by the firmware file /lib/firmware/nvidia/510.39.01/gsp.bin. A few select products currently use GSP by default, and more products will take advantage of GSP in future driver releases".

NVIDIA is preparing a new GeForce RTX 3060 Ti graphics card, with a new GA103-200-A1 GPU that will reportedly feature the same specs as the higher-end GA104 GPU.

The new GA103-200 GPU is a big deal here as it is different to the GA104 GPU that powers a large range of NVIDIA's current-gen Ampere GPUs: RTX 3070 Ti, RTX 3070, RTX 3060 Ti, RTX 3060, RTX 3070 Lite Hash Rate, RTX 3060 Lite Hash Rate. This is now the third variant of the GA103 GPU, and we might actually have a huge supply from NVIDIA which would be great at this price point of the (insane) market.

We're finding out about the new GeForce RTX 3060 Ti with its GA103-200 GPU (with its 7680 CUDA cores) from the new 511 Studio driver NVIDIA pushed out, which had a new device ID of 2414 -- matching the PCI Device family of NVIDIA's GA103 GPU -- the new GPU making the GeForce RTX 3080 Ti Laptop GPU.

It looks like AMD Is about to reveal a new workstation GPU, with the upcoming Radeon PRO W6400 card teased for a January 19 announcement. Check it out:

AMD has teased some blue gemstones that could be Sapphire, Tanzanite, or Kynite -- with a new king -- a king of its class, that is. We're not talking about a new performance king that would hit the high-end, but rather a new Navi 24-based budget workstation GPU that will compete against the NVIDIA T600 Turing GPU.

It should be the last member in the stack of RDNA 2 cards, but offer AMD a battle partner to NVIDIA's T600 Turing GPU. We will know more details in a few days, but I would expect less VRAM than usual and a single PCIe power connector. We'll know all about AMD's new Navi 24-based Radeon PRO W6400 (and whatever else the company unveils) on January 19.

NVIDIA teased the world with one of the best custom GeForce RTX 3080 Ti -- hell, one of the best custom graphics cards ever -- with its GeForce RTX 3080 Ti Matrix Resurrections Edition graphics card.

The GeForce RTX 3080 Ti Matrix Resurrections Edition graphics card is being given away in China, based on the reference RTX 3080 Ti, with a custom front black and backplate that says "War Starts on 1/14" as a laser-etched LED logo. It looks pretty damn good, celebrating the release of The Matrix Resurrections in China on 1/14.

If you want to win the card, first off you'll need to be a resident of China, and then you'll need to write a sentence about how you're a huge fan of both The Matrix and GeForce GPUs, and post it on Weibo.

NVIDIA's new GeForce Game Ready 511.23 WHQL drivers have been released, with the new drivers including support for NVIDIA's just-released GeForce RTX 3080 12GB graphics card.

The new drivers also include support for NVIDIA's new DLDSR (Deep Learning Dynamic Super Resolution) technology, as well as Game Ready Driver support for Tom Clancy's Rainbow Six Extraction, Hitman III, The Anacrusis, GRIT, and Monster Hunter Rise. There's also support added for Windows 11 Dynamic Refresh Rate, and support for CUDA 11.6.

The biggest part here with the new GeForce Game Ready 511.23 WHQL drivers is the support for the new GeForce RTX 3080 12GB graphics card, which you could buy last week before even the press had drivers. The new drivers have baked-in support for the RTX 3080 12GB, DLDSR, a bunch of new games, and the usual fixed issues along the way.

MSI's new flagship GeForce RTX 3090 Ti SUPRIM X graphics card has been teased, with new specifications leaked that include a default 480W TDP.

The new MSI GeForce RTX 3090 Ti SUPRIM X not only has a 480W TDP, which is 30W higher than their custom MSI RTX 3090 SUPRIM X, but it has 24GB of even-more-ultra-fast GDDR6X memory clocked at 21Gbps. GPU boost clock wise we're looking at "EXTREME" mode pushing the GPU to 1900MHz, while "GAMING" and "SILENT" modes have the GPU clocked at 20MHz less, 1880MHz.

You will need 3 x 8-pin PCIe power connectors, where we should see the sleeker, single 16-pin PCIe 5.0 power connector on NVIDIA's own flagship GeForce RTX 3090 Ti Founders Edition. MSI most likely won't have this card for a while, as I exclusively reported a few days ago that an industry source of mine said NVIDIA has "halted production" on the GeForce RTX 3090 Ti... and we still don't know why.

Well, it looks like you'll need to deal with having just a single kidney for NVIDIA's new GeForce RTX 3090 Ti -- especially if you live where Hostel was based, in Europe -- and have to pay anywhere between $3571 and $4497. Yeah.

The price for two of MSI's upcoming custom GeForce RTX 3090 Ti graphics cards has been listed by some distributors and retailers in Europe, with two Swiss retailers and one German distributor listing prices for NVIDIA's new flagship GPU. MSI's new custom GeForce RTX 3090 Ti SUPRIM X and GeForce RTX 3090 Ti GAMING X TRIO have been priced, and oh boy.

We're looking at anywhere between $3654 and up to a huge $4497 for the MSI GeForce RTX 3090 Ti SUPRIM X (after conversion from its overseas pricing), while the MSI GeForce RTX 3090 Ti GAMING X TRIO is a little cheaper, coming it at only $3571.