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

Is there any actual reason to believe Volta has suddenly jumped 1 year early from the roadmap?
Yes, there was one rumor saying that, but that's just one rumor - there's also a rumor it would be 7nm which rules out 2017 release, and then the actual fact that it has always been 2018 on the roadmap (since it returned to it)

"Few quarters" would imply the worst-case scenario for Vega 10 releasing in June and Volta in Q1 2018.
But yeah, most likely they're ~1 year apart, and Volta will release closer to Vega 20 than Vega 10.

Vega 10 will probably have its hands full with GP102/4, anyway.
 
I wonder how large vega 11 is. Should be between polaris 10 and vega 10 but by how much? Vega 10 looks to be 2x the flops of polaris 10 so that leaves a lot of room for another chip.
 
Roughly 350-370mm^2, AMD've released gpus in and around that size for a while (cypress, cayman, tahiti, tonga) so it's a good guess for the smaller Vega.
 
I wonder how large vega 11 is. Should be between polaris 10 and vega 10 but by how much? Vega 10 looks to be 2x the flops of polaris 10 so that leaves a lot of room for another chip.
Should it? AMD stated clearly with Polaris that the number doesn't reflect chips positioning, for all we know Vega 11 could be bigger than 10 (there were tons of rumors about that earlier, too)
 
Then what chip will replace Hawaii in the high-DP compute market?

- Hawaii is 3 years old.
- It seems to me AMD is sticking with 2 graphics chips/year.
- 2016 has had its fill.
- 2017 is Vega 10 and Vega 11
- By late 2017 Hawaii will be 4 years old

If Vega 10 can't replace Hawaii in all its markets, then you're suggesting AMD will just stop competing in that market for several years in a row?
People are shocked enough that AMD isn't competing in the high-end consumer market for 3 or 4 quarters..

I didn't mean to say that Vega 10 couldn't have fast DP (though I think that's shaky for unrelated reasons). I only meant that if your argument for Vega 10 having fast DP uses any instance of "But that's what AMD did with Hawaii!", then it's flawed. Hawaii wasn't a bad GPU by any means, but as I stated earlier, so much of its storied history reads like a book on how-not-to-make/release-a-GPU that AMD surely wouldn't be taking pointers from Hawaii's release/lifespan (but the sick bastard in me almost hopes that it happens again only because Hawaii was a fantastic deal, much to AMD's detriment).

Should it? AMD stated clearly with Polaris that the number doesn't reflect chips positioning, for all we know Vega 11 could be bigger than 10 (there were tons of rumors about that earlier, too)

I think he only meant that there would be a massive gap between a ~64 CU chip and a 23 CU chip. It would be smart if Vega 11 helped plug that gap.
 
Just guessing here, but if all Vegas use an interposer and they're gluing chips together. I'd imagine the designs are probably ~1.4x the performance of each other to hit various tiers. Say 225mm2 and 300mm2 so they double into 450mm2 and 600mm2 while being reasonably sized to fit onto HPC Zen alone. Any larger two big ones wouldn't fit on an interposer with HBM. Smaller, a good old fashioned APU would likely be better with costs and yields. Differentiated from each other enough to hit relatively even performance tiers: 1x, 1.4x, 2x (1.8x cut), 2.8x (2.5x cut). While leaving enough room for 10% cut down parts to be sold to differentiate the high end and enthusiast parts. Only concern is 6 is a lot of high to enthusiast SKUs, but 390, 390x, Nano, Fury, and FuryX is already 5? Excluding the Pro Duo as that could still be possible with this design or how it's actually implemented.

P10 would still exist for discrete GDDR parts that are price sensitive.

As for the fast FP64, the guys wanting that will be paying top dollar for the biggest GPU with all the ram they can get. I'm just not sure the transistors are worth spending on it when power efficiency and size likely matter. Nvidia purpose built P100 and an entire rack for the task.
 
Hawaii wasn't a bad GPU by any means, but as I stated earlier, so much of its storied history reads like a book on how-not-to-make/release-a-GPU that AMD surely wouldn't be taking pointers from Hawaii's release/lifespan

Hawaii wasn't a bad example of how to make a GPU. It was poorly released, however, a tradition that AMD continued with the flop that was Fury X. It didn't have enough supply and its reference coolers were atrocious.

Hawaii was a hell of a workhorse and absolutely invaluable to AMD in 2014 & 2015

If it was, then AMD wouldn't have had dropped market share during this period.
 
Vega11 could be bigger but the dual Vega10 rumor makes it unlikely.

On other note, I doubt that AMD would have HBM in xbox and that they wouldn't use the design for other products unlike Polaris10. So a 384-bit chip with gddr5/x with more shader engines might be on cards?
 
Vega11 could be bigger but the dual Vega10 rumor makes it unlikely.

On other note, I doubt that AMD would have HBM in xbox and that they wouldn't use the design for other products unlike Polaris10. So a 384-bit chip with gddr5/x with more shader engines might be on cards?
Xbox will have an APU, even if the iGPU would have exactly identical specs to a discrete chip, they wouldn't be the same design - it would be nothing but a coincidence.
 
Then what chip will replace Hawaii in the high-DP compute market?

- Hawaii is 3 years old.
- It seems to me AMD is sticking with 2 graphics chips/year.
- 2016 has had its fill.
- 2017 is Vega 10 and Vega 11
- By late 2017 Hawaii will be 4 years old

If Vega 10 can't replace Hawaii in all its markets, then you're suggesting AMD will just stop competing in that market for several years in a row?
People are shocked enough that AMD isn't competing in the high-end consumer market for 3 or 4 quarters..
What's Nvidias available High-End-DP chip atm? Isn't this GK210 which basically is GK110+more registers? It's pretty old now as well.
 
They shouldn't be that different either. And I was going more on the rumors of AMD using data fabric for CPU+GPU MCMs,
If Scorpio was using a MCM I'd imagine the case design would look significantly different. The MCM design would be far more compact and likely allow a different, more efficient, cooling setup with HBM. A refresh for a slim version I could definitely see as costs come down a bit.

It's world of a difference to make GPU part of APU compared to discrete, and I doubt Xbox would have the fabled MCM
Generally yeah, but in the case of a MCM the APU and discrete cards could actually use the same chip. Run PCIE over the interposer on the APU and would the GPU know any different? Beyond the likely lower voltages and latency? The biggest difference between APU and discrete has been the memory pool. With HBM it's effectively on the GPU.
 
Haven't seen a live one yet and also none outside of Nvidia. That's why.

They are only available for specific contract, they exist, but if you want one you need contracts as supercomputer centers or entreprise who are now starting to upgrade their installations. And thoses contract are not really public, so...impossible to know how much pieces have been done and running in test system or system who are build at this moment. basically they could order 2500 pieces today and have only 20 right now, 100 in 2 months.. 500 in 6 etc etc .

Availlability for "consumers", through HP, Cray etc are still set at Q1 2017. ( i insist, delivery pass through HPC vendor system, but its a bit obvious with this type of system anyway )
 
Last edited:
What I mean is, I have not seen a live one anywhere installed and reported upon. I am quite aware that I cannot walk into mom-and-dad's electronics paradise and buy one right off the shelf.
 
Well all servers from Dell, HP, and others, that have supposedly been getting supply, won't come out till end of Q4 this year, and Q1 next year, so yeah probably won't see one till then.
 
Back
Top