• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

Someone in chat on Stardock's Ashes launch stream was spouting nonsense about the Surface 5 Pro (due in Q4) having Pascal and 4GB of HBM, then asking if it would be able to run Ashes.

The dev's reply was interesting.

He basically said Pascal wouldn't be power efficient enough to be in a Surface product. He said Polaris could be efficient enough, but doubted that it'd be included in a Surface product and thought it would have integrated graphics.

Evidently he has some knowledge of Pascal's architecture that hasn't been made public. He doesn't seem to think it compares well to Polaris in efficiency.
 
It's more likely about the time frame and what will be out. AMD are bringing a low end Polaris this year, Nvidia might only have the midrange chip out this year. Historically on a new process Nvidia take a hell of a long time to bring out the low end chip. Meaning a 1080 gtx wouldn't be remotely appropriate in a low power Surface type device but a 120mm^2 Polaris 11 at 25-35W would be.
 
If MS had tendered a contract that large with the possibility of them winning it, regardless of desktop plans, they'd have made absolutely sure they had a small chip ready to go.

I don't think there was much chance of either AMD or NVIDIA getting anything. 120mm2 and 25W would still be too much for surface. It'd be a custom chip. They'll use integrated graphics. Then next year a Zen APU might be possible.

But he said efficiency and nothing about size, which refers to architecture rather than SKUs, and effectively stated that Pascal was inferior in this regard.
 
Last edited:
Pascal not being as efficient as expected will be due to the readded dedicated DP hardware. Only reason for Maxwell's efficiency was its removal.
 
pmc25 said:
Someone in chat on Stardock's Ashes launch stream was spouting nonsense about the Surface 5 Pro (due in Q4) having Pascal and 4GB of HBM, then asking if it would be able to run Ashes.

The dev's reply was interesting.

He basically said Pascal wouldn't be power efficient enough to be in a Surface product. He said Polaris could be efficient enough, but doubted that it'd be included in a Surface product and thought it would have integrated graphics.

Evidently he has some knowledge of Pascal's architecture that hasn't been made public. He doesn't seem to think it compares well to Polaris in efficiency.
Click to expand...

Still in fantasy-land I see.

Occam's Razor: he misunderstood, conflated or otherwise misspoke.
 
Orangey said:
Still in fantasy-land I see.

Occam's Razor: he misunderstood, conflated or otherwise misspoke.
Click to expand...

Occam's razor means making as few assumptions as possible, yet you're assuming he's wrong based on nothing. He said what he said and presuming he meant something else isn't the easiest explanation. The simplest explanation(with no other information available) is that the guy who can reasonably be expected to have information about upcoming hardware believes Pascal to be less viable for the next Surface tablet than Polaris.

Assuming he meant to say something else is a more complex explanation and supported by nothing at all. The interpretation of what the developer said is up for debate, but jumping to assumptions that he meant to say something different is illogical.
 
Well it is a custom Maxwell 940m in the surface book, so I see no reason at all that a Pascal chip on the smaller node which should will be more power efficient, wouldn't be able to go into a future surface product.
 
Reading that again, I can assume that they meant that Pascal with HBM wouldn't be power efficient enough and being that only the High end Pascal is likely to have HBM and those chips certainly wouldn't be efficient enough, he is right.
 
bru said:
Reading that again, I can assume that they meant that Pascal with HBM wouldn't be power efficient enough and being that only the High end Pascal is likely to have HBM and those chips certainly wouldn't be efficient enough, he is right.
Click to expand...

No. He blathered about hbm during the dev's short reply. I.e. dev was responding to pascal in it generally and if he read the bit about hbm would have only seen after.

He said what he said.

His accuracy re: what may be in the new Surface may be entirely off, but he clearly thinks Pascal is less efficient than Polaris.
 
Mauller said:
Pascal not being as efficient as expected will be due to the readded dedicated DP hardware. Only reason for Maxwell's efficiency was its removal.
Click to expand...

Not true, as GK104 also had no DP and anyway Fiji was stripped of DP too. DP is not a reason für Maxwells or Fijis better efficiency. It's a small advantage but not more.
Both companies will need to pay a small fee for reintroducing DP into their offerings, but as it's only available in the highest end (Big Vega and GP100) it will play no role for polaris or the other pascal offerings.
 
Last edited:
Samwell said:
Not true, as GK104 also had no DP and anyway Fiji was stripped of DP too. DP is not a reason für Maxwells or Fijis better efficiency.
Click to expand...

Figi has not had it's dp stripped out. It's just disabled to 1/16 and could be enabled to 1/2 if AMD chose to do so which can't be done with maxwell as it's not there (Stripped out). Due to only having 4gb Hbm it would be no use in the professional market. The Gpu itself would be a compute monster.

http://vrworld.com/2015/09/07/amd-r9-fury-x-potential-supercomputing-monster/
 
Last edited:
Samwell said:
Not true, as GK104 also had no DP and anyway Fiji was stripped of DP too. DP is not a reason für Maxwells or Fijis better efficiency. It's a small advantage but not more.
Both companies will need to pay a small fee for reintroducing DP into their offerings, but as it's only available in the highest end (Big Vega and GP100) it will play no role for polaris or the other pascal offerings.
Click to expand...

The old style Titan did have DP, this is one of the reasons why it was no match for a GTX 690 which did not.

And the above was despite the fact the Titan had a higher transistor count and all on one core, not split like a GTX 690.
 
TheRealDeal said:
Figi has not had it's dp stripped out. It's just disabled to 1/16 and could be enabled to 1/2 if AMD chose to do so which can't be done with maxwell as it's not there (Stripped out). Due to only having 4gb Hbm it would be no use in the professional market. The Gpu itself would be a compute monster.

http://vrworld.com/2015/09/07/amd-r9-fury-x-potential-supercomputing-monster/
Click to expand...

Old News and false. GCN 1.2 can't do more than 1:16 in hardware. They were talking at HPC events that they can't do more. It's the reason you see S9300 with 1:16 DP.
http://anandtech.com/show/10209/amd-announces-firepro-s9300-x2
 
Kaapstad said:
The old style Titan did have DP, this is one of the reasons why it was no match for a GTX 690 which did not.

And the above was despite the fact the Titan had a higher transistor count and all on one core, not split like a GTX 690.
Click to expand...

The Titan and Titan Black also sold in droves to semi professionals, and professionals that wanted top Quadro performance for significantly cheaper; while also having great gameplay performance.

I'm really looking forward to Polaris/Vega as myself and some family members are always looking for the best of both worlds for GPUs, I'm really hoping I can replace my two 980Tis with a single card "soon".
 
Samwell said:
Old News and false. GCN 1.2 can't do more than 1:16 in hardware. They were talking at HPC events that they can't do more. It's the reason you see S9300 with 1:16 DP.
http://anandtech.com/show/10209/amd-announces-firepro-s9300-x2
Click to expand...

I'm glad someone else on the forum isn't afraid to post the truth. I've been sy wing this reputedly and getting shot down by the usual suspects. AMD and nvidia had to strip FP64 compute because they needed the transistor budget with it a node shrink.
 
Last edited:
You must log in or register to reply here.
Back
Top Bottom