Slashdot Mirror
Nvidia Claims Its New Chip Is the 'World's Fastest GPU' for Game and VR Design (venturebeat.com)
An anonymous reader shares a VentureBeat report: Nvidia announced today the Quadro P6000 graphics card for workstations, using the "world's fastest GPU." The graphics card is targeted at designers who have to create complex simulations for everything from engineering models to virtual reality games. The Quadro P6000 is based on Nvidia's new Pascal graphics architecture, and it uses a GPU with 3,840 processing cores. It can reach 12 teraflops of computing performance, or twice as fast as the previous generation. Nvidia unveiled the new platform for artists, designers, and animators at the Siggraph graphics technology conference in Anaheim, Calif. AnandTech has more details on this. From their article:As NVIDIA's impending flagship Quadro card, this is based on the just-announced GP102 GPU. The direct successor to the GM200 used in the Quadro M6000, the GP102 mixes a larger number of SMs/CUDA cores and higher clockspeeds to significantly boost performance. Paired with P6000 is 24GB of GDDR5X memory, running at a conservative 9Gbps, for a total memory bandwidth of 432GB/sec. This is the same amount of memory as in the 24GB M6000 refresh launched this spring, so there's no capacity boost at the top of NVIDIA's lineup. But for customers who didn't jump on the 24GB -- which is likely a lot of them, including most 12GB M6000 owners -- then this is a doubling (or more) of memory capacity compared to past Quadro cards. At this time the largest capacity GDDR5X memory chips we know of (8Gb), so this is as large of a capacity that P6000 can be built with at this time. Meanwhile this is so far the first and only Pascal card with GDDR5X to support ECC, with NVIDIA implementing an optional soft-ECC method for the DRAM only, just as was the case on M6000.
but sorry cant leave contact info here, have to make a first post on slashdot.
And, like literally all new GPUs this year, it will be totally unbuyable.
My brother's wanted a new GPU for months, so I told him I'd get him one. No Rx480 stock as it was sold out day one and has pretty much remained so since it launched. So I looked at the 1060, no stock available whatsoever there either. Same goes with Nvidia's 1080 and 1070, which have apparently had vastly limited quantities, even moreso than the smaller cheaper GPUs.
And of course their less available, that's how chip manufacturing works. The larger the chip's die, the more transistors that can go wrong during manufacturing, the more likely the chip is to come out as a dud. Don't expect this "world's most powerful GPU for gaming!" to actually be buyable by most anyone. Probably have to wait till next year for anything beyond the smallest new GPUs to be available in any kind of decent quantity.
According to Anandtech, the GPU is the GP102 which is the same as in the recently announced top-end consumer card "Titan X" (note: not "GTX Titan X".. confusing? yes)
The Titan X has 3584 shader processors while the Quadro P6000 has 3840 and twice the memory. I assume that this means that the Titan X has a lower-binned chip.
Previous generation of Nvidia GPUs ("Maxwell" architecture) has a GPU called GTX 980 Ti, which was a lower-binned GTX Titan X with half the memory. Now when the 10-series Titan X is already the lower-binned GPU, I suppose this means that there will not be any "GTX 1080 Ti".
"We mustn't be caught by surprise by our own advancing technology" -- Aldous Huxley
Currently on linux, modern AMD cards have the "best-of-both-world" driver support.
Nvidia currently only produce closed-source drivers.
(Nouveau is exclusively the work of reverse engineering. Recieving nearly no support from Nvidia, except for the occasional patch to enable modesetting)
AMD provides a hybrid stack:
- they develop an kernel module (amdgpu) which is available up-stream. (i.e.: new versions of the kernel feature it out of the box).
above this, you have two choices:
- AMDGPU-Pro, the closed source drivers (which are the modern day equivalent of the user-space portion of Catalyst).
Nowadays, they seem pretty stable, run games without bugs, and because they require a module which is already in mainstream kernel, they do work even with the latest kernel update. (unlike nvidia's driver which need the nvidia.ko some adaptation in case of variation of the kernel API).
- RadeonSI, the opensource back-end to the Mesa driver.
These are devloped by people of whom some are on AMD's payroll (i.e.: AMD doesn't only provide information, but even salaries for opensource development)
With the Polaris, the driver was available at release day, and has a decent performance compared to the closed source one, and runs lots of games.
That's quite some achievement compared with the early "fglrx" that was buggy as hell, and that's quite some engagement for the opensource community.
As a Linux user, I actually like more the ADM driver situation.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
You're right, but the irony is the Crysis's relatively poor optimisation means that it still doesn't really run at the levels you might expect even on new hardware. And Crysis 3, of course, is only just now starting to be edged out as a go-to game for benchmarking (the current favoured games for that role seem to be The Witcher 3, Ashes of the Singularity and Assassin's Creed: Unity/Syndicate).
The Radeon Pro SSG just killed that off. Most likely, anyway. 8K video rendering at 90+fps is pretty impressive to me. The other machine barely hit 17fps.
Uh, the Radeon RX 480 is $220 or so, and besides the GTX 1060 is only about 15% faster.