Monday, December 12th 2016
AMD "Vega" Demoed in Sonoma, California
AMD's next-generation high-end graphics card, based on the "Vega" architecture, was showcased at an event in Sonoma CA, earlier this week. While the architecture is being debuted with the Radeon Instinct MI25 deep-learning accelerator, a prototype graphics card based on the silicon was exhibited by the company, showing Vulkan API gaming.
AMD was pretty tight-lipped about the specifications of this prototype, but two details appear to have slipped out. Apparently, the chip has a floating point performance of 25 TFLOP/s (FP16), and 12.5 TFLOP/s (FP32, single-precision). On paper, this is higher than the 11 TFLOP/s (FP32) of NVIDIA TITAN X Pascal. The other important specification that emerged is that the card features 8 GB of HBM2 memory, with a memory bandwidth of 512 GB/s. This, too, is higher than the 480 GB/s of the TITAN X Pascal. It remains to be seen which market-segment AMD targets with this card.
This article was updated on Dec 15 to accommodate AMD's request to remove all info regarding the demo system, the shown game and its performance, which has been put under NDA retroactively.
Source:
Golem.de
AMD was pretty tight-lipped about the specifications of this prototype, but two details appear to have slipped out. Apparently, the chip has a floating point performance of 25 TFLOP/s (FP16), and 12.5 TFLOP/s (FP32, single-precision). On paper, this is higher than the 11 TFLOP/s (FP32) of NVIDIA TITAN X Pascal. The other important specification that emerged is that the card features 8 GB of HBM2 memory, with a memory bandwidth of 512 GB/s. This, too, is higher than the 480 GB/s of the TITAN X Pascal. It remains to be seen which market-segment AMD targets with this card.
This article was updated on Dec 15 to accommodate AMD's request to remove all info regarding the demo system, the shown game and its performance, which has been put under NDA retroactively.
120 Comments on AMD "Vega" Demoed in Sonoma, California
Apparently people think that if something is more costly, it should also be better. It has worked for NV till now....
Ofcourse, AMD won´t do that and will put it 100 bucks cheaper, between the cost of the 1070 and 1080, having that score in DX11.
AMD train is arriving a 2nd time. First one was shrugged of by NV fanboys (the 1060 being 12% faster...which is no longer true!); and now this.
Even if the Zen is only somewhat decent; a lot of people will be going full red.
Basically because I like a cool and quiet man cave.
The forum member just throws a title and a link. Pretty easy and fast.
Me thinks this card has two stacks of 4 GiB each which leaves the door open to a 1 TiB/s card with 16 GiB in the future (a workstation card)...
Edit - I read wrong. Dual vega is not yet shown, if there is such a thing.
Edit - Only 1 vega shown, others cards shown are polaris and nano(fiji).
Post 4 had some compares, IIRC R9 290X was against GTX 780, launch prices$549 vs $649 . Then Fury X was against 980/Ti, $649 vs $549/$649. It took a fair time before Fiji cards were cheaper than say a 980 Ti in the UK @ etailers. I was toying with the idea of getting a 980 Ti, basically the £50 cheaper price of Fury X and then say future purchase of FreeSync monitor vs G-Sync swayed it for me. So I think it will be closely priced to nVidia 1070/1080 and the lack of premium for a FreeSync monitor vs G-Sync will be the sweetener for a buyer.
The "instinct" slide shows Vega has HBM, regardless of if it's HBM 1 or 2 I would think it's more costly compared with GDDR5X, plus the interposer for GPU/HBM must add cost to card production price?
In fact, Anandtech had an article a while back saying 4096 cores is about as much as the architecture can take, Fiji or otherwise. So, if this Vega 10 is touted at 12.5 T/flops at a certain speed (1525Mhz), it works out at 4096 shaders.videocardz.com/64706/amd-vega-doom-4kultra-gaming-performance-demo-possible-specs
I don't see Vega 11 being bigger than 4096 shaders.
Lets hope they can speed things up this time...
And I beg to differ about the launch date of Vega 10. Most possibly in January 2017.
Raja Koduri answered that on twitter