Video Cards & GPUs News - Page 259

We knew that from the last lot of leaks from an internal AMD slide last year that the company was planning PCIe 4.0 support for its upcoming Vega 20 GPU, and now we have some more support for these claims.

AMD engineers are hard at work on Linux drivers for its fresh new Vega 20, which is being made on the smaller 7nm node. The new PCIe 4.0 standard will usher in some huge bandwidth increases, as PCIe 3.0 has been with us for close to a decade now and feels stale with its 8.0GT/s of raw bit rate and 8Gbps of link bandwidth, where PCIe 4.0 doubles those numbers to 16GT/s and 16Gbps of bandwidth.

The previous rumor pegged Vega 20 for the second half of 2018, made on the 7nm node with 4-stack HBM2 giving us 1TB/sec with 16GB or 32GB HBM to choose from, PCIe 4.0 x16 support and between 150-300W of power. Now we can expect PCIe 4.0 motherboards around the corner, perfect timing for AMD to push into the datacenter.

It seems NVIDIA is ramping up towards something big with a new prototype turning up just days after the company celebrated Alan Turing's birthday, we have a picture of a prototype board of a next-gen GPU.

The board in question has 3 x 8-pin PCIe power connectors, and 12 modules of Micron's super-fast GDDR6 technology on-board, as well as what seems like NVLink connectivity at the top. There's no GPU on the board itself, but with the 12 modules of GDDR6 we could see a next-gen GeForce GTX 1180 with 12GB of GDDR6 RAM.

This might not be a new GeForce GPU, and could be a new Tesla (based on Turing). From what I've been told the new GeForce will be a heavily refined Pascal GPU architecture named Ampere, while the GPU to succeed Volta will be Turing. This could very well be a Turing-based card being tested with GDDR6, mostly for the strange connectivity at the top of the board, which could be NVLink.

NVIDIA truly dominates the AI supercomputer race with its Tensor Core GPUs, and now the company has announced today that the world's top AI supercomputers are ALL powered by NVIDIA Tensor Core GPUs.

The new list of the world's 500 fastest systems sees NVIDIA taking the top spots with Tensor Core GPUs, with the two latest AI supercomputers securing the #1 and #3 spots of the world's fastest supercomputers with Summit, and Sierra, respectively. The fastest supercomputer in Japan, ABCI, is also powered by NVIDIA Tensor Core GPUs.

GPUs now power 5 out of 7 of the world's fastest systems, and better yet NVIDIA Tensor Core GPUs power 17 of the 20 most energy efficient systems on the GREEN500 list. Most of the GREEN500 systems are NVIDIA GPUs, just the nail the point home that NVIDIA dominates the supercomputer market.

NVIDIA seems to be ramping up into its next-gen GeForce GTX 11 series graphics card launch, with teases from TSMC entering 7nm production with NVIDIA as a customer, and GDDR6 going into mass production over the next few months. During Computex 2018, we heard that NVIDIA would be announcing the GTX 1180 on July 30, while AIB partners told me they would have cards on the market in August or September.

But it was a tweet by the official NVIDIA Twitter account that celebrated Alan Turing's birthday, with a quote from Turing that's interesting: "We can only see a short distance ahead, but we can see plenty there that needs to be done". Rewinding the clock to earlier this month, where NVIDIA CEO and founder Jensen Huang said that the next-gen GeForce cards were a while away, where he said: "It's a long time from now".

TSMC enters 7nm production, GDDR6 is being spun up into mass production and AMD has nothing at all for Radeon gamers until this time next year. NVIDIA is already ahead of the best AMD offers with the Radeon RX Vega 64 being absolutely dominated by the GeForce GTX 1080 Ti let alone TITAN Xp, so the launch of a new GPU would put them an entire generation and a half ahead, minimum.

TSMC has announced that its exciting 7nm node has entered production, with the company also announcing that it will be investing $25 billion into 5nm node technology.

The company has been talking about and developing 5nm node technology since early 2016, with its eyes on a launch in 2020, and now it is putting its money where its mouth is. Reuters reports: "Taiwan Semiconductor Manufacturing Co Ltd, a supplier to Apple Inc, said on Thursday it expects to invest $25 billion in 5-nanometer node technology".

Apple should be interested in TSMC's focus on 5nm as they've been using TSMC for chip production for years now. TSMC is also forward-looking to 2022, when we should expect 3nm to pop its head out of the super-small chip production.

Taiwan Semiconductor Manufacturing Company, otherwise known as TSMC, has announced that commercial production of 7nm chip has started. TSMC also teased that the newer 5nm node will enter mass production by the end of 2019, or early-2020.

The news comes directly from TSMC boss CC Wei, with DigiTimes reporting: "The ramp-up of 7nm chip output will boost TSMC's overall production capacity to 12 million 12-inch equivalent wafers in 2018, up 9% from 10.5 million units in 2017, Wei indicated, without elaborating on specific orders and customers".

DigiTimes continued: "TSMC will tape out more than 50 chip designs with its 7nm process technology by the end of 2018, Wei said. AI, GPU and cryptocurrency applications take up the majority of the tape-outs, followed by 5G and application processors. TSMC will also start taping out chips built using an enhanced 7nm node with EUV in the second half of 2018, Wei noted. The foundry expects to move the node to risk production in the third quarter".

It looks like NVIDIA might be delaying the launch of its next-gen GeForce GTX 1180, with sources saying a major Taiwanese OEM has returned 300,000 GPU units to NVIDIA... which could delay the new GTX 11 series cards.

SeekingAlpha is reporting that a major OEM that is considered in the "Taiwan Top 3" has returned 300,000 GPUs to NVIDIA. The site reports: "Reports out of Taiwan now suggest that NVIDIA has a gaming GPU inventory problem. 'Semiaccurate' reported on the issue yesterday, and cited excess inventory in the channel as the primary reason for new gaming GPU delay. The glut is so severe that one top Asian OEM partner reportedly returned 300k GPUs to NVIDIA".

Why would there be so many GPUs floating around and then returned? Well, the cryptocurrency mining epidemic would've caused this, and while it was good for a few months business-wise, now it could be biting NVIDIA in the (green) ass.

Acer's new Predator Helios 500 gaming notebook has been somewhat of a mystery with its interesting choice of AMD's larger-than-life Radeon RX Vega 56... but now we have the skinny on what Acer will be doing with it.

We know the Acer Predator Helios 500 can be configured with up to the Intel Core i9-8950HK processor and NVIDIA GeForce GTX 1070, but it can also be configured with all-AMD components including the new Ryzen 7 2700 processor and Radeon RX Vega 56 graphics.

But we know that the Radeon RX Vega 56 has a 210W TDP in desktop form with a second "power save" BIOS that drops it to 135W, so what the hell (haha, it's HOT down there) will Acer do with the Predator Helios 500 to keep it cool? Well, it looks like the RX Vega 56 inside of the Helios 500 will have its TDP locked to 120W.

We told you the week before Computex 2018 that AMD would show off its next-gen 7nm GPU ready for the new wave of Radeon Instinct accelerators, and now Vega 20 has reportedly been through some benchmarks of Ashes of the Singularity.

The card has turned up with a device ID of "66A0:00", where it will be offering twice the density, twice the power e fficiency, and 35% improved performance over the Vega 10 on the 14nm FinFET node. The benefits of 7nm seem to be huge for Vega so far, but we'll need to see Vega 20 in Radeon RX Vega form before we can pass judgement.

Don't expect that until next year, before Navi arrives as a mid-range part with the purported Radeon RX 680. As for Vega 20, it rocks 32GB of HBM2 and 4096 stream processors into an area smaller than the current Vega 10 GPU. If that's not impressive, I don't know what is.

AMD's latest and greatest graphics card looks to be the new Radeon PRO V340, which is a multi-user computing solution for virtualized environments... you know, so not for gaming.

What makes AMD's new Radeon PRO V340 so special is that it features dual Vega 10 GPUs on 7nm, with a huge 32GB of HBM2 on-board. This seems to be the card that was teased at Computex 2018, and now it is becoming a reality.

The new Radeon PRO V340 features shared hardware encoding for H.265 and H.264 codecs, with something that AMD teases as a "built-in security processor" that we have no idea about just yet. This card is made for design and manufacturing, oil and gas research as well as the media and entertainment industry.