AMD Polaris architecture – GCN 4.0

Mauller · 13 Apr 2016 at 18:52

Yeah, 1.152 is the apparent base clock in the open drivers.

Also, that little tidbit that GloFo are doing better than expected with 14nm parts is very good. Means they should be able to churn out Polaris and Zen parts with decent numbers per wafer.

D.P. · 13 Apr 2016 at 19:41

SimBy said:
Mobo manufacturers are irrelevant here AMD or Intel have to support it. And there is no reason they ever will.

Only to get the CPU interconnect, but for multiGPU performance then NVlink doesn't require anything form AMD or Intel CPUs.

Stanners · 13 Apr 2016 at 19:58

Orangey said:
You're wasting your breath mate it goes in one ear out the other. People just can't get it through their heads. It's like how they still want pooled VRAM in mGPU.

re: Pascal pretty obvious there will be some sort of GP110 in 2017 or whenever they can get more than 0.7 good dies per wafer.

Is that aimed at me?

Mtom · 14 Apr 2016 at 14:40

AMD Radeon R9 480 and Radeon R9 470 Series Detailed – Polaris 10 “Ellesmere” Has Around 100W TDP, Polaris 11 “Baffin” Under 50W/

kadath · 14 Apr 2016 at 14:43

Mtom said:
AMD Radeon R9 480 and Radeon R9 470 Series Detailed – Polaris 10 “Ellesmere” Has Around 100W TDP, Polaris 11 “Baffin” Under 50W/

http://wccftech.com/amd-radeon-r9-480-470-polaris-10-polaris-11/

flopper · 14 Apr 2016 at 14:50

source is here,
dont post wcc as they just repost the source

http://vrworld.com/2016/04/13/amd-polaris-10-gpu-beat-competing-pascal/

Mtom · 14 Apr 2016 at 14:52

flopper said:
source is here,
dont post wcc as they just repost the source

http://vrworld.com/2016/04/13/amd-polaris-10-gpu-beat-competing-pascal/

This is a different article.

AllBodies · 14 Apr 2016 at 15:10

Seems crazy the performance per watt improvement AMD has managed (if it turns out to be true).

If Polaris 10 is only ~110W (overclocking headroom anyone?

), then by their "2.5x improvement" it should perform the same as a 275W 28nm card. So Fiji presumably.

LoadsaMoney · 14 Apr 2016 at 15:15

No 490 then, or will that be Vega, or will there not be a one, and Vega will be a named like Fury ?

jupiter_man · 14 Apr 2016 at 15:18

TBH I dont really care about the voltage, performance is all I care...

kadath · 14 Apr 2016 at 15:32

AllBodies said:
Seems crazy the performance per watt improvement AMD has managed (if it turns out to be true).

If Polaris 10 is only ~110W (overclocking headroom anyone? ), then by their "2.5x improvement" it should perform the same as a 275W 28nm card. So Fiji presumably.

No way a 2.5x improvement.

muziqaz · 14 Apr 2016 at 15:37

Mtom said:
This is a different article.

yeah, URL says polaris beats pascal, but article doesn't even remotely mention anything about one GPU beating other

bru · 14 Apr 2016 at 15:45

Well if that rumour is true then 'oh dear'.
Replace the 380X with Polaris 10 at 100W that is a great improvement, similar performance for half the power.
Now of course they could've done twice the performance at the same power, which would bring it up to FuryX standards, but unfortunately they cannot do both at the same time. Even with the touted 2.5* Per/W at half the power 100W it would be nowhere near FuryX performance.

It will all depend on the truth of the rumour and of course what route NVidia take with their Pascal chips to how well this will work out for AMD.

TheRealDeal · 14 Apr 2016 at 15:53

bru said:
Well if that rumour is true then 'oh dear'.
Replace the 380X with Polaris 10 at 100W that is a great improvement, similar performance for half the power.
Now of course they could've done twice the performance at the same power, which would bring it up to FuryX standards, but unfortunately they cannot do both at the same time. Even with the touted 2.5* Per/W at half the power 100W it would be nowhere near FuryX performance.

It will all depend on the truth of the rumour and of course what route NVidia take with their Pascal chips to how well this will work out for AMD.

AllBodies said:
Seems crazy the performance per watt improvement AMD has managed (if it turns out to be true).

If Polaris 10 is only ~110W (overclocking headroom anyone? ), then by their "2.5x improvement" it should perform the same as a 275W 28nm card. So Fiji presumably.

As Allbodies said above it's most likely comparing to Figi (Fury X). Basically if the rumour is true a Polaris 10 at 110w should match a Fury X in performance. Something else to think about is this is not the fully fledged polaris 10 if you read the specs. It's slightly cut down like the 290 is from the 290x.

If it is replacing the 380x band of card this section is about to get a massive jump in performance. There is 2 different articles going on with similar information which is getting me a little mixed up as to what's going on.

bru · 14 Apr 2016 at 16:07

Why not the 275W 390, remember this supposedly has 2304 cores, yet you think it should be compared with 4096 core part, with twice the wattage and twice the performance. the article says replacing the 380X at half the power, not me.

Steampunk · 14 Apr 2016 at 16:10

bru said:
Well if that rumour is true then 'oh dear'.
Replace the 380X with Polaris 10 at 100W that is a great improvement, similar performance for half the power.
Now of course they could've done twice the performance at the same power, which would bring it up to FuryX standards, but unfortunately they cannot do both at the same time. Even with the touted 2.5* Per/W at half the power 100W it would be nowhere near FuryX performance.

It will all depend on the truth of the rumour and of course what route NVidia take with their Pascal chips to how well this will work out for AMD.

Let's wait and see what goes into the 490/490x spot. It's conspicuous by it's absence.

TheRealDeal · 14 Apr 2016 at 16:10

bru said:
Why not the 275W 390, remember this supposedly has 2304 cores, yet you think it should be compared with 4096 core part, with twice the wattage and twice the performance. the article says replacing the 380X at half the power, not me.

Well because the 390x is Gcn 1.1 so surely they would be comparing to there latest tech like they did with Figi to Hawaii.

Replacing it and giving it an **** kicking at the same time while using way less power.

The rumour is 2.5 time the performance per watt. It's not about what's being replaced here or there. If a Polaris card uses 100w then times it by 2.5 which is 250w. What in AMD's latest Gen used around this power. I would then think Fury pro and that's where i would guess performance should be give or take.

queamin · 14 Apr 2016 at 16:15

You can only compare on price as what slots in what price bracket I.E what slots in at 380x price point and so on.

bru · 14 Apr 2016 at 17:47

TheRealDeal said:
Well because the 390x is Gcn 1.1 so surely they would be comparing to there latest tech like they did with Figi to Hawaii.

Replacing it and giving it an **** kicking at the same time while using way less power.

The rumour is 2.5 time the performance per watt. It's not about what's being replaced here or there. If a Polaris card uses 100w then times it by 2.5 which is 250w. What in AMD's latest Gen used around this power. I would then think Fury pro and that's where i would guess performance should be give or take.

At 2*Per/W it sounds to me like your expecting twice the performance and half the power, it's one or the other you cannot have your cake and eat it too.

N19h7m4r3 · 14 Apr 2016 at 17:53

bru said:
At 2*Per/W it sounds to me like your expecting twice the performance and half the power, it's one or the other you cannot have your cake and eat it too.

That makes no sense. Yes you can. You can't eat any cake if you don't first have it.