Home General Various News AMD Vega 10 – Vega 11 and Vega 20 GPUs Mentioned

AMD Vega 10 – Vega 11 and Vega 20 GPUs Mentioned

225

Earlier on at an investors event AMDs CTO mentioned AMD Vega 10 and Vega 11, but now also Vega 20 GPUs can be added to the list as it got mentioned in a roadmap. Not just that, some new specs can be added as well.

More and more rumors are being spread and can be read on multiple websites, most of them saying other things and others the same. Let’s make a quick overview:

The latest information suggests VEGA as the company most high-end GPU to be released in two years time. Interesting was that he mentions it’ll be released in Q1 2017, on par with earlier rumors, but contradicting rumours of a Q4 October 2016 VEGA release. “Significant performance/Watt refinements have been made with Vega” is stated. Now a new part, Vega 20, information was spotted and posted by VideoCardz. More interesting, three segments on VEGA architecture have now surfaced.

Vega 10

The start of the “Vega” will happen with will be called the Vega 10 (codename) GPU, a series to compete with Nvidias more high-end gear. It’s mentioned that this GPU has 64 Compute Units, multiply that with 64 shader processors per cluster and you’ll get to 4096 shader processors. It will offer up to 24 TFLOP/s 16-bit (half-precision) floating point performance. You read it right, half-precision. The GPU will be paired with either 8 or 16 GB HBM2 memory at an up-to 512 GB/s memory bandwidth, logic would dictate 8GB though.
 

 
Vega 10 would thus be the GPU that replaces the Hawaii/Grenada series. It is mentioned that total board power (not TDP) sits at 225 Watt. Vega 10 should hold HBM2 memory and that’s the culprit, it is just not available in volume. Hence even I myself expected a VEGA10 product released in October armed with GDDR5(X) graphics memory). That last bit thus is the uncertain part.

Vega 11

Vega11 – lets call it “Big Vega’ which is to replace the FIJI / Fury (X) parts. So as suggested, where Vega10 would get 4096 stream processors and could compete with the GeForce GTX 1080 and GTX 1070 Vega11 on its end would feature 6144 stream processors (=rumor) and would be lined against the GP100/GP102 aka Big Pascal GPU (Titan X). This all is obviously known and discussed many times already. Then there now is some new and intersting info.

Vega 20

Then there is a new one, Vega 20. This bit comes from Videocardz who apparantly spotted it in a server roadmap. It is interesting as that is a codename we have not seen/heard from before. Apparently this will be released at a much MUCH later timeframe as Vega 20 will be a die-shrink of Vega 10 towards, and here it comes 7 nm GF9 process being developed by GlobalFoundries. 7nm (UV) production rumors (as I know right now) will see the light of day in 2019. But that VEGA 20 GPU should get the very same 4096 shader processors with maybe more memory, higher clocks who knows really. So a die-shrunk GPU, likely by that time will be a mainstream part.

The Near future

After Polaris (460 / 470 / 480) in 2016, in 2017 the first Vega based graphics card will make an appearance. Vega is the brightest star in the constellation Lyra, the fifth brightest star in the night sky and the second brightest star in the northern celestial hemisphere, after Arcturus. Next in line in the 2018 timeframe we see Navi with the two keywords being scalability and Nextgen memory.
  

 
The slide is plotted in years/time against perf per watt, so you will notice that the perf/watt ratio almost makes a 2x step every new release. It will be interesting to see when AMD will make that move to 7nm as it is next step in fabbing, but again that technology is due for release in 2019/2020.

When will the first VEGA card be released ?

Since all these rumors and fud raise more and more questions, the honest answer is .. I don’t know. Earlier rumors suggested an October released for the more ‘simple’ VEGA model with GDDR5(X) and a HBM2 part to follow later. If you look at the rumors from this and last week, it’s likely going to be early 2017 with HBM2 as suggested. Time will tell.

Source link

LEAVE A REPLY

Please enter your comment!
Please enter your name here