AMD GDC2016 Thread

[Kenmitch]

Then I don't believe it.

Polaris 10 was in the demo. -_-

There's no way a 115mm2 GPU is 390X+ class. No way!

Did they ever state that Polaris 10 is 1 single gpu? Maybe depending on yields it's multiple skus?

 

[IllogicalGlory]

It looks promising but they didn't provide us any actual FPS measures or which settings they were using.

Still, it seems very impressive, considering that Polaris 10 is 100 mm^2 or something.

 

[Silverforce11]

OMG did he just say "games ruined with GameWorks integration!"

Salvod.

 

[Glo.]

http://imgur.com/68y9WUg

This has been posted on another forum, that Im sure most of users are familiar with. IF this is correct, Tahiti with 128 Bit memory bus would be around 115 mm2 on GloFo 14 nm LPP.

 

[Mahigan]

It looks promising but they didn't provide us any actual FPS measures or which settings they were using.

Still, it seems very impressive, considering that Polaris 10 is 100 mm^2 or something.

New geometry processor, memory controller, L2 cache, command processor and re-worked CUs. Yep, entirely possible.

They likely boosted utilization and reduced the amount of times the data over flows out of the cache banks and into the frame buffer.

That will invariably lead to impressive performance.

 

[PPB]

Then I don't believe it.

Polaris 10 was in the demo. -_-

There's no way a 115mm2 GPU is 390X+ class. No way!

Dont forget settings. We could argue what settings were used as the earlier BFSW demo wasnt on ultra. Even at mid/high settings, its nothing to sneeze at considering the estimated TDP/power consumption of the card.

Whatever brings me the most performance in BF5 at either Vulkan (assumed) or DX12 at 960 power consumption is getting my money this round.

 

[dacostafilipe]

It looks promising but they didn't provide us any actual FPS measures or which settings they were using.

Still, it seems very impressive, considering that Polaris 10 is 100 mm^2 or something.

Let's hope the press will be able to "confirm" performance on the demo PCs.

 

[DownTheSky]

Then I don't believe it.

Polaris 10 was in the demo. -_-

There's no way a 115mm2 GPU is 390X+ class. No way!

TOTAL WAR: WARHAMMER!! DX12 hype! lol

I find it too very hard to believe. Maybe Polaris is AMD's Maxwell, super optimized for games. Leaving Vega the DP/compute monster.

 

[jpiniero]

Perhaps Polaris 10 is the bigger one (~232 mm2) and Polaris 11 the smaller one? Still that would be better than expected if it's closer to Fiji performance and something like 125-140 W.

Expect $$$ though.

 

[el etro]

Polaris 10 is the bigger GPU(232mm2), confirmed via WCCF moderator comments. Polaris 11(12x mm probably) was the GPU demoed in january.<br />
<br />
This Hitman performance puts Polaris 10 around R9 fury performance, something closer to three times the performance density versus GCN1.2. I would expect Pascal giving results around this, too.

 

[DownTheSky]

We don't know the clocks either. Maybe they're clocked at 1500mhz or so.

 

[USER8000]

Perhaps Polaris 10 is the bigger one (~232 mm2) and Polaris 11 the smaller one? Still that would be better than expected if it's closer to Fiji performance and something like 125-140 W.

Expect $$$ though.

After they mentioned they were going to demo Polaris 10,they had a slide which mentioned price/performance,so I expect competitive pricing.

 

[Silverforce11]

Polaris 10 is the bigger GPU(232mm2), confirmed via WCCF moderator comments. Polaris 11(12x mm probably) was the GPU demoed in january.<br />
<br />
This Hitman performance puts Polaris 10 around R9 fury performance, something closer to three times the performance density versus GCN1.2. I would expect Pascal giving results around this, too.

This is much more believable.

232mm2 with the density of 14FF, it's equivalent to a very big GPU on 28nm.

 

[dacostafilipe]

https://twitter.com/thracks/status/709528808416673793

Clarification: We showed Polaris 11 at CES. This was the first showing of Polaris 10.

 

[Mahigan]

I had estimated that BaffinXT would be around the same performance as Fury-X. I had thought this to be Polaris 11. I had thought that the bigger of the two chips would be Polaris 11.

I doubted that the smaller chip would give 390x performance but I guess anything is possible right now.

 

[boozzer]

http://imgur.com/68y9WUg

This has been posted on another forum, that Im sure most of users are familiar with. IF this is correct, Tahiti with 128 Bit memory bus would be around 115 mm2 on GloFo 14 nm LPP.

this could mean all your predictions to be true right? I am super excited.

 

[DownTheSky]

This Hitman performance puts Polaris 10 around R9 fury performance, something closer to three times the performance density versus GCN1.2. I would expect Pascal giving results around this, too.

Keep in mind he said LOCKED 60 fps. So minimum 60. What we got from benchmarks are avg fps.

 

[Glo.]

Sorry, my bad. I was under impression that the smaller GPU was Polaris 10.

Still, it is still very good. So, the smaller one was Polaris 11, this was 10. So the P10 is bigger GPU with supposedly 232 mm2 of die size.

 

[Mahigan]

We don't know the clocks either. Maybe they're clocked at 1500mhz or so.

Not likely. They're probably clocked closer to 1,050MHz. Higher clocks would defeat the purpose of perf/watt. That being said, they'll likely over clock Very well.

 

[sandorski]

HP dude looks wired.

 

[Timmah!]

Did people really expect a massive GPU on a brand new node with yield issues that early?

AFAIK its not brand new and there are no confirmed yield issues, thats all assumptions based on previous generation history. Not to mention Nvidia released massive GPUs on brand new node before, so its not like such thing never happened.

 

[el etro]

Keep in mind he said LOCKED 60 fps. So minimum 60. What we got from benchmarks are avg fps.

Oh yep. 980Ti does less than 60FPS minimum on Hitman Dx12.

 

[MajinCry]

Concerning the power usage discussed on the first page...Isn't it pretty damn plausible that AMD is not giving a cock and bull story?

The thing just has to sit between the 7790 (XB1) and the 7870 (PS4). Consider that the 7850 fits snuggly between them, and uses around 130 watts. From the node-shrink alone, the GPU would be using around 55 watts or something.

Then factor in architectural improvements, conservative clocks, chip integration, etc., 35 watts is about right in line.

 

[Maverick177]

It is generally expected for the next gen flagship product to do better than the previous gen flagship. If the big Polaris somehow doesn't pull ahead of Fiji in nearly every categories, it would be a huge disappointment.

I really do not want to see another Bulldozer.

 

[Silverforce11]

If it's a small chip, it's not really next-gen flagship. It's more a mid-range chip and if it's sold at flagship prices, it's only gamers who suffer.