AMD: Speculation, Rumors, and Discussion (Archive)

[Deleted member 13524]

The way I see it, the fact that Vega is only 6 months away from Polaris means it's basically the same architecture and same process. It just uses HBM2 instead of GDDR5.
Apart from that, the rest of the higher performance/watt simply comes from the fact that Vega might be higher-end and bigger chips, which traditionally have better performance/watt.

 

[CSI PC]

I don't recall what was being claimed or what version of HBM it applied to, and I have not run across much either way.
Having some kind of additional benefit/arrangement as a result of working with Hynix seems plausible, but how that aligns with what was being claimed, or what kind of impact that could really have on Nvidia or anyone hasn't come up.

So far, it doesn't seem like getting early dibs on HBM has had major import if it is true.

Type in any search engine: +hbm2 +amd +supply +agreement
Or replace supply agreement with priority.
This will bring up those rumours that are meant to be from an AMD source, the story regarding HBM2 not being 2016 from the same AMD source would require different search.
But these stories may all link back to WCCF and their source in AMD, and if so not really ideal.
Cheers

 

[homerdog]

Some of their 28nm stuff undervolts massively as well, Hawaii for one.

This always suggested to me that AMD has simply pushed their cards well past the optimal point on the power/performance curve to compete with NVIDIA on absolute performance. Hawaii being the prime example.

 

[Razor1]

Yeah Vega sounds like the Polaris architecture with HBM2, makes sense too. This also correlates what Koduri stated only 2 new GPU's for this year.

 

[Adored]

This always suggested to me that AMD has simply pushed their cards well past the optimal point on the power/performance curve to compete with NVIDIA on absolute performance. Hawaii being the prime example.

Yes this is also undeniable, especially in the case of Hawaii/Grenada.

 

[gamervivek]

The PCB of AMD's Polaris 10 GPU is said to be very similar in size to AMD's Fury Nano, meaning that AMD is likely pushing for a compact design for their future reference high end GPU designs.

http://www.overclock3d.net/articles/gpu_displays/amd_polaris_10_engineering_sample_pictured/1

 

[Silent_Buddha]

http://www.overclock3d.net/articles/gpu_displays/amd_polaris_10_engineering_sample_pictured/1

Interesting that it hints at Polaris 10 being faster then R9 Fury X. Not surprising. How much faster is the question.

Regards,
SB

 

[3dilettante]

Which picture has Polaris 10 in it?

 

[Adored]

http://www.overclock3d.net/articles/gpu_displays/amd_polaris_10_engineering_sample_pictured/1

Biggest hint at HBM1 yet?

 

[McHuj]

Biggest hint at HBM1 yet?

Yes, interesting. A small form factor would necessitate a very small bus with few memory chips or HBM. But given how it may be faster than a Fury, I'd imagine that it would need similar bandwidth.

 

[SimBy]

There are some very short PCBs currently using GDDR5. 970 I think? Also Polaris 11 running passively is not really
that surprising, since some estimates put it under 30W.

 

[Grall]

Much PCB space in many of today's boards is wasted by putting power delivery at the rear of the card and leaving the area between GPU/DRAM and the mounting brackets almost completely barren. A lower wattage card might get away with putting its more compact voltage regs there, maybe moving the GPU slightly towards the rear, thus chopping off a significant piece of real-estate at the end of the board...

I'm surprised so few boards do this, and those who do are low-end cards from what I've seen.

 

[SimBy]

Much PCB space in many of today's boards is wasted by putting power delivery at the rear of the card and leaving the area between GPU/DRAM and the mounting brackets almost completely barren. A lower wattage card might get away with putting its more compact voltage regs there, maybe moving the GPU slightly towards the rear, thus chopping off a significant piece of real-estate at the end of the board...

I'm surprised so few boards do this, and those who do are low-end cards from what I've seen.

An example of that

And cmpared to Nano

 

[Adored]

Yeah it's definitely possible to do now even with GDDR5 but developer boards are normally a bit bigger aren't they, or at least not so well engineered? Unsure.

 

[fellix]

Any chance that would look like a GDDR5 chip?


p.s.:

Considering the size of the chassis used for the system, the board length is definitely constrained:

 

[Razor1]

Yeah it's definitely possible to do now even with GDDR5 but developer boards are normally a bit bigger aren't they, or at least not so well engineered? Unsure.

Usually ES samples that are given out to developers are the same as retail ones, there maybe clock speed differences with the core and vram and what not, but that's about it.

 

[Adored]

Cheers guys. Very interesting It seems we have multiple trusted sources making opposite claims about GDDR5 and HBM1 too

 

[Jawed]

Yes, interesting. A small form factor would necessitate a very small bus with few memory chips or HBM. But given how it may be faster than a Fury, I'd imagine that it would need similar bandwidth.

980Ti is proof that you don't need HBM-ish bandwidth.

 

[Jawed]

and bigger chips, which traditionally have better performance/watt.

Raja Koduri completely contradicts you on this point in his interview with Ryan Shrout.

 

[Frenetic Pony]

It's just the biggest version of Polaris, but with HBM2 instead of what will, probably, end up being GDDR5/X for the 2 smaller cards. Apparently the yields for GloFlo are as bad as those for TSMC, meaning both Nvidia and AMD have their big Finfet cards delayed. Looks like "Vega" is scheduled for "sometime" between the end of this year and beginning of the next, no doubt waiting on production ramp up and yields from the foundry.