AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

From AMD's documentation the xGMI PCS blocks are what can be muxed through the PCIe lanes.

https://support.amd.com/TechDocs/54945_PPR_Family_17h_Models_00h-0Fh.pdf

GMI would be the on-package links that are given a dedicated interface type.

 

except that it's supposed to combine both VS and GS stages into one

 

I think this is one part of the new pipline. I think with the Next Generation Geometry pipline, AMD try to simmulate a Frontend on normal Shader, which will be very flexible

Maybe in the Future you don't have a front end. You have only the shader and a Workloaddistributer. You will get than a realy flexible Frontend. As more Polygons come in as more shaders are blocked for polygon work.

 

Vega20 id shows up in 3dmark11, clockspeed doesnt seem correct, only 1000Mhz.

https://videocardz.com/76076/early-amd-vega-20-3dmark11-benchmark-result-emerges

 

Yeah and 70% faster than Vega 64. Hard to believe, unless it has more CUs.

 

70?

 

According to wccftech:

https://wccftech.com/amd-7nm-vega-20-32gb-gpu-3dmark-benchmark-leaked-up-to/

 

It's obvious that it's not reading the clock correctly.
Also,

 

For what it's worth, at least I couldn't dig out the claimed results from 3DMark result database

 

wccftech article is by Khalid Moammer who always hypes AMD rumors. Videocardz author posted this link on his twitter,

https://www.3dmark.com/compare/3dm11/12756151/3dm11/12754835#

He also thinks that the clocks are placeholders in the bios which are getting read by 3dmark.

Another long shot would be dual GPU which will have a Vega20 pci id.

 

So there's 3 possible explanations for the results then: dual-gpu which doesn't get recognized by 3DMark, clocks which are read wrong by 3DMark, All claimed leaks and rumors of Vega 20 configuration are just dead wrong and it's a giant chip with like 8192 streamprocessors?

 

Did anybody recognise that the reaults are the same? So i think it´s the same clock Speed and the same architecture but at lower Power

 

They're not the same, 2 very similar results though. And WCCFTech compared one of the results to the Vega 64 Liquid Edition, which gets around same score, too.

 

Forgot that it'd look at SLI/CF too, so quite likely that VC explanation is correct and it's picking up the default values for the clocks and not the boost clocks.

 

You mean 1.7ghz without causing local brownouts and green house effect?

 

Same slide set also presented Vega 11 as discrete card, while we now know it's the IGP in Raven Ridge

 

http://images2.freshpatents.com/imageviewer/20180115496-p20180115496
http://www.freshpatents.com/-dt20180426ptan20180114290.php


Vega 20 GPU patents?

 

It's not. Confusingly AMD GPU 'codenames' (based on some sort of internal ordering) and product names (based on CU count) overlap.

 

I think his post is in context of this slide at Videocardz Jan 2017:




But then maybe they changed their strategy from that Q4 2016 presentation or there is a bit of misinformation around *shrug*.