AMD Greenland Vega 10 said to have 4096 stream processors

[HEXUS]

Information was discovered in an AMD R&D Manager's LinkedIn profile.

Read more.

[kalniel]

I'm getting a bit bored of the 'leak' via linkedin profile now. What possible point is there in advertising you worked on a product with that precise number of stream processors? It's either a product tease or it's doing a formula 1 and letting your competitors know something in the hope of getting a job with them.

But yes, I agree with the author that it looks likely this is a fiji replacement with HBM2. And it won't be for a good while yet

[chj]

What?! I thought Polaris would have HBM 2?

[kalniel]

What?! I thought Polaris would have HBM 2?

Nope, see the news a few weeks ago:

http://forums.hexus.net/hexus-news/3...vega-navi.html

[Corky34]

I thought that too, incorrectly it now seems, that could be a missed opportunity if the rumors of Pascal using HBM2 are true it would mean Nvidia getting the jump on them wouldn't it?
Haven't looked when Pascal is meant to hit market BTW.

[kalniel]

I thought that too, incorrectly it now seems, that could be a missed opportunity if the rumors of Pascal using HBM2 are true it would mean Nvidia getting the jump on them wouldn't it?
Haven't looked when Pascal is meant to hit market BTW.

Pascal isn't HBM2 either.

[Corky34]

Next thing your going to tell me is that the earth isn't round.
I was reading rumors just a few weeks ago that Pascal was going to have HBM2, or at least the top end card would.

[kalniel]

Next thing your going to tell me is that the earth isn't round.
I was reading rumors just a few weeks ago that Pascal was going to have HBM2, or at least the top end card would.

Normal pascal (this year) isn't HBM2 - the process isn't ready yet. There might be a top end (Titan type) pascal that comes out next year that could be HBM2, but AMD will be on Vega by then.

[Corky34]

Obviously rumors being rumors nothings certain but there are whispers that a GDDR5x Pascal will show its face around the time of Computex in May, with a HBM2 version to follow around the third or fourth quarter, it should be interesting to see how things pan out.

[badass]

It looks to me like the number of SP's on Polaris and later are not remotely comparable to current GCN SP's based on this. Here's hoping each SP has around double the real world performance.

[Corky34]

What's an SP, i know GPUs have ROPs, TMUs, and cores but whats a SP?

[DanceswithUnix]

Obviously rumors being rumors nothings certain but there are whispers that a GDDR5x Pascal will show its face around the time of Computex in May, with a HBM2 version to follow around the third or fourth quarter, it should be interesting to see how things pan out.

Given Nvidia were faking Pascal announcements in January (http://semiaccurate.com/2016/01/11/n...m-competition/) I expect that while they seem to have taped out at last anything they have in May will be a bit of an early prototype. Still, with Nvidia that might not stop them seeding a few boards to the tamest of press writers with a list of which games and benchmarks to run. I wouldn't expect to buy one for a while though.

[Jonj1611]

What's an SP, i know GPUs have ROPs, TMUs, and cores but whats a SP?

Stream Processor I believe.

[Corky34]

Given Nvidia were faking Pascal announcements in January (http://semiaccurate.com/2016/01/11/n...m-competition/) I expect that while they seem to have taped out at last anything they have in May will be a bit of an early prototype. Still, with Nvidia that might not stop them seeding a few boards to the tamest of press writers with a list of which games and benchmarks to run. I wouldn't expect to buy one for a while though.

Interesting take on it i guess, some may say it was just a mockup so they had something to wave at the crowd, seeing as we're dealing with rumors and speculation though I'm not going to damn or praise any company based on conjecture.

[kompukare]

Assuming Micron's announcement of GDDR5X just sampling now was accurate, surely in May(ish) any card which uses GDDR5X could only paper launch?
The other thing is that for now only Micron has announced GDDR5X (AFAIK) and Micron / Elpida were never the best for GDDR5 (poor overclocking and timings).
What is less known is whether a GDDR5 controller can easily be designed to take both, and if it can whether the GDDR5X announcements came early enough in the development cycle to allow this. If that is the case it might be possibel for a GDDR5X card to be paper launched, while a cut-down version of the die is shipped with normal GDDR5. That would be sensible risk-mitigation.

[scaryjim]

...some may say it was just a mockup so they had something to wave at the crowd ...

Some may say that, but the person who matters - NVidia's CEO - didn't: he said it was actual Pascal silicon. In the same way he said he was holding an actual Fermi card a few years earlier.

Let's be clear on what happened back in January - whilst AMD were showing retail games running on actual Polaris silicon, NVidia were waving around a previous generation card with the serial numbers filed off and claimed it was actually their next generation silicon. Again.

There are all sorts of reasons they might have done that, but none of them really excuse lying to the press - a number of the more moderate tech sites carried the same story with a less ... satirical edge than Charlie gave it. They point out that whatever NVidia's reasons were for showing Maxwell silicon but claiming it was Pascal, it doesn't encourage optimism or trust...