REGISTER | LOGIN
Breaking News
Comments
Newest First | Oldest First | Threaded View
rick merritt
User Rank
Author
Re: ...
rick merritt   1/5/2017 5:59:50 PM
NO RATINGS
Yes, Alex also started his thinking process with what was known about the Vega chip on the MI25 board!

realjjj
User Rank
CEO
Re: ...
realjjj   1/5/2017 3:08:20 PM
NO RATINGS
A distinction must be made between Vega the architecture and Vega 10.

About Vega 10 we know a few details. The Instinct MI25 uses it and AMD claimed almost 25TFLOP FP16 so the assumptions are that it is 4096 cores at 1500+MHz. After a quick look at the pictures available, the die size appears to be at about 500mm2 but unclear if on 16FF or 14nm. If so, they might be taking a step back in perf density(in FLOPS) but gaining a lot in efficiency and likely utilization in gaming. In gaming it could be anywhere from 10% bellow Titan X(1200$) to beating it by 15% or more. TDP was stated at bellow 300W,remains to be seen how much bellow that.

 

Vega the architecture is bound to reach GPUs in all price bands and APUs.The APU bit is interesting, if Vega turns out to be a rather large step forward. It could enable certain key design wins by offering what Intel can't and maybe your next notebook would be AMD based - just exploring the possibilities Vega could offer, not aimed at starting any rumors.

rick merritt
User Rank
Author
Re: ...
rick merritt   1/5/2017 1:35:30 PM
NO RATINGS
@realjjj  Alex of JPR was clear that his performance esitmates were educated guesses given AMD has not revealed many key details such as frequency, power consumption, die size, etc. That all comes with benchmarks at the product roll out I suppose.

realjjj
User Rank
CEO
...
realjjj   1/5/2017 12:47:11 PM
NO RATINGS
"Vega could deliver an average 50% boost on Viewperf performance and 20% on application benchmarks compared to its current Polaris chip,"


How much of that is on facts and how much an assumption?

Even if it was that, it tells nothing about power and area and it would still be plenty on the same node.

8-bit is a big deal for inference and don't think Nvida can pack 8 bit so the AMD Instinct MI25 would reach 50TFLOPs 8-bit vs 21-ish for the Nvidia P100.

The new rasterizer seems to catch up with Nvidia and Nvidia likely saved quite a bit of power there.

The biggest step forward would be if they have a variable length SIMD but it's unclear if they do.

In DX11 gaming, Nvidia had some 30% advantage in gaming perf vs TFLOPs. The question is how much better can Vega do?

Anyway, this is the biggest update for their GPU architecture since 2011 so maybe analysts expect too little.

 

""I don't expect Nvidia to move to 10 nm or 7 nm until Volta," said Krewell."

This might be interpreted as Volta not arriving before 7nm but Nvidia has teased Xavier with Volta on 16ff.



Like Us on Facebook
EE Life
Frankenstein's Fix, Teardowns, Sideshows, Design Contests, Reader Content & More
Martin Rowe

Test Tool Finds Ethernet Wiring Errors
Martin Rowe
Post a comment
When my house was renovated several years ago, I had the electrician install network outlets in numerous places, then run the LAN cables to a wiring closet. But he didn't document the ends ...

Martin Rowe

Local Electronics Store Supplies Engineers and Hobbyists
Martin Rowe
5 comments
Rochester, N.Y. — Tucked away in this western New York city known for its optics is Goldcrest Electronics, a local store that's supplied businesses and individuals with electronic ...

Martin Rowe

How to Transform a Technology University (Book Review)
Martin Rowe
1 Comment
The Presiding Genius of the Place by Alison Chisolm. WPI, Worcester, Mass., 234 pp., 2016. Engineers love to discuss, and often criticize, engineering education. They often claim ...

Max Maxfield

Aloha from EEWeb
Max Maxfield
Post a comment
Just a few minutes ago as I pen these words, I posted this blog about this month's Cartoon Punchline Competition over on EEWeb.com.