AMD's next GPU uarch is called "Polaris"

[3DVagabond]

Who says the "vendor specific code" isn't just disabling async compute and going with serial compute for Nvidia cards?

Why would they say "Vendor specific code" if there wasn't any? And do you think if they added a different render path for nVidia they wouldn't make it optimized for the hardware that's using? Just read what they say and stop trying to make stuff up that you don't know simply because it fits your position that nVidia is better.

 

[96Firebird]

If hardware = Nvidia, use serial compute.
If hardware = other, use asynchronous compute.

BAM, vendor specific code.

 

[SPBHM]

This is the closest thing I could find. A 950 should be running somewhere around 80-85% of its max power consumption when capped at 60fps. Nvidia has "Adaptive VSync", so this should limit the power consumption on a 950 to around 80% of full power. The numbers AMD provided do make sense. I think its a fair test. The 950 appears to be drawing roughly 75 watts, and the Polaris card only about 25 watts. If they release that chip onto a 80W TDP card, it is going to be a frickin monster midrange 970 killer, at probably $199. And yet Nvidia will still sell more 970's....

not sure what AMD used for FOV and AF
but another test with probably a different map/gameplay gives lower results at 1080p medium profile
on the graph you posted it gives 66FPS for a 750 Ti, on this test 48FPS

on this test scenario the 950 would probably be really close to 60FPS even if uncapped

 

[dogen1]

If hardware = Nvidia, use serial compute.
If hardware = other, use asynchronous compute.

BAM, vendor specific code.

I don't think serial is the right term..

 

[monstercameron]

If hardware = Nvidia, use serial compute.
If hardware = other, use asynchronous compute.

BAM, vendor specific code.

So you are proposing that they are so incompetent, they need a vendor to specifically disable a feature they know runs poorly...I can't even...

 

[MrTeal]

So you are proposing that they are so incompetent, they need a vendor to specifically disable a feature they know runs poorly...I can't even...

Isn't that exactly what Kollock from Oxide said? The nVidia DX12 driver reported that Async Compute was fully supported, but using it caused a host of issues so they disabled it on nVidia hardware.

 

[Paul98]

anyone tried a similar test for themselves to see what sort of power they pull? That power consumption number seems very low

 

[lilltesaito]

Isn't that exactly what Kollock from Oxide said? The nVidia DX12 driver reported that Async Compute was fully supported, but using it caused a host of issues so they disabled it on nVidia hardware.

He is also the person who said.

Personally, I think one could just as easily make the claim that we were biased toward Nvidia as the only 'vendor' specific code is for Nvidia

Edit:
Quote isn't really for you MrTeal, was just going off yours. It is for 96Firebird

 

[freeskier93]

anyone tried a similar test for themselves to see what sort of power they pull? That power consumption number seems very low

I assume you're talking about the 950? Obviously nobody can confirm the Polaris GPU...

950 number seems fine to me. It's a 90W TDP GPU and i7 4690k is 88W but that i7 would barely break a sweat in that situation. People on the internet like to grossly exaggerate power requirements, so if you're going by typical PSU recommendations just realize that.

 

[Silverforce11]

https://www.youtube.com/watch?v=hvD37UUcdIo

I'm not much of a screen quality aficionado, but are HDR displays that good? Thoughts?

 

[Arachnotronic]

https://www.youtube.com/watch?v=hvD37UUcdIo

I'm not much of a screen quality aficionado, but are HDR displays that good? Thoughts?

I know Intel was touting HDR as the "next big thing" in displays at IDF 2015, and now we have AMD touting them too, so maybe there's something to them?

 

[at80eighty]

https://www.youtube.com/watch?v=hvD37UUcdIo

I'm not much of a screen quality aficionado, but are HDR displays that good? Thoughts?

Most accounts state it to be a noticeable IQ improvement. Personally i tend to notice colours & gradation more easily, so it's exciting and a must have, if the price sits well with the other monitor tech i am looking forward to

 

[Madpacket]

HDR looks exciting and I love the fact it'll work on older GCN hardware through a driver update. I'm not sure what Nvidia plans on offering this year but AMD appears to be hitting it out of the park so far with Polaris. 14NM Finfet for smaller dies, 16NM Finfet for larger dies makes sense to me. Fury X should hold them over until Fall 2016.

 

[Tuna-Fish]

HDR is great. The biggest problem with it is that for good results, your content needs to be HDR too. Given how few people have HDR displays, this creates a chicken-and-egg problem -- who's going to spend time and money to make their game/video look better on HDR screens when no one has them, who's going to get one when no content looks better on it?

 

[3DVagabond]

If hardware = Nvidia, use serial compute.
If hardware = other, use asynchronous compute.

BAM, vendor specific code.

Yeah! I'm sure that those two lines are all it takes. I'll bet you anything that you could type those two sentences anywhere you want to and all that would happen is it would throw errors up the wazoo when you attempted to compile.

Besides they actually say "vendor specific code". Deny whatever you want to. It doesn't make you right.

 

[3DVagabond]

Isn't that exactly what Kollock from Oxide said? The nVidia DX12 driver reported that Async Compute was fully supported, but using it caused a host of issues so they disabled it on nVidia hardware.

You can't simply disable something. You do that and it all stops working. You have to code the alternative render path. When they did that they would have logically optimized it for the hardware it was written for. Not to would have been dumb.

They stated that all they had to do for AMD was to write the code as specified by MSFT and it worked on AMD. So there's the possibility that they can optimize it better for AMD. No point in doing that this early on though considering they very well might have to do multiple more rewrites before it's done.

 

[RussianSensation]

"PCPer’s Ryan Shrout issued a clarification stating that the Polaris GPU AMD demoed was up and running after just two months from when initial production kicked off, rather than being brought up to market after two months.

We’ve reached out to AMD and they have confirmed that the Polaris GPU demoed was in fact built using Globalfoundries’ 14nm FinFET process. Some publications have reported that Polaris will be mix of both TSMC 16nm and Globalfoundries 14nm GPUs, which is where some of the confusion could potentially have stemmed from. However, according to AMD Polaris is only 14nm. Furthermore, they’ve asked us to refer back to the company’s press release , which you can find quoted below in the original article, which makes clear mention of the Polaris Architecture being 14nm based."

Read more: http://wccftech.com/amd-polaris/#ixzz3wd4lblVU

 

[Arachnotronic]

"PCPer’s Ryan Shrout issued a clarification stating that the Polaris GPU AMD demoed was up and running after just two months from when initial production kicked off, rather than being brought up to market after two months.

We’ve reached out to AMD and they have confirmed that the Polaris GPU demoed was in fact built using Globalfoundries’ 14nm FinFET process. Some publications have reported that Polaris will be mix of both TSMC 16nm and Globalfoundries 14nm GPUs, which is where some of the confusion could potentially have stemmed from. However, according to AMD Polaris is only 14nm. Furthermore, they’ve asked us to refer back to the company’s press release , which you can find quoted below in the original article, which makes clear mention of the Polaris Architecture being 14nm based."

Read more: http://wccftech.com/amd-polaris/#ixzz3wd4lblVU

Right, the small Polaris chip they demo'd is Samsung 14nm.

 

[Silverforce11]

Right, the small Polaris chip they demo'd is Samsung 14nm.

English isn't my first language, but that wording strongly implies the entire Polaris uarch is designed for 14nm FF.

Interesting times ahead for them, as putting eggs into one basket with GloFo's track record...

I have a feeling if that is the case, we may indeed see some multi-smaller chip linked with an interposer approach for high end SKU... and it falls in line with what Raja said, they only have 2 Polaris chips (1 small one demo and maybe 1 other mid-range one).

Meeting their agreement with GloFo for wafer volume AND having good yields with smaller chips while be able to use interposer tech to make a seamless package out of multi-chips? Clever.

 

[Arachnotronic]

English isn't my first language

Really? Your English from your posts here is quite good. Props! :thumbsup:

but that correction from PCPER strongly implies the entire Polaris uarch is designed for 14nm FF. Interesting times ahead for them, as putting eggs into one basket with GloFo's track record...

I don't think so, but we'll see!

 

[Pantalaimon]

I know Intel was touting HDR as the "next big thing" in displays at IDF 2015, and now we have AMD touting them too, so maybe there's something to them?

I'm a bit confused but is this display HDR different from HDR on GPUs? I remember many years back there was a quite a stir when the HDR feature came to GPUs.

 

[NTMBK]

I'm a bit confused but is this display HDR different from HDR on GPUs? I remember many years back there was a quite a stir when the HDR feature came to GPUs.

What is often referred to as "HDR" is actually tone-mapping; calculating the image in a higher dynamic range space, and then dynamically remapping it to the standard 8-bit intensity range. This is increasing the actual range of intensities which can be displayed by the monitor.

 

[maddie]

English isn't my first language, but that wording strongly implies the entire Polaris uarch is designed for 14nm FF.

Interesting times ahead for them, as putting eggs into one basket with GloFo's track record...

I have a feeling if that is the case, we may indeed see some multi-smaller chip linked with an interposer approach for high end SKU... and it falls in line with what Raja said, they only have 2 Polaris chips (1 small one demo and maybe 1 other mid-range one).

Meeting their agreement with GloFo for wafer volume AND having good yields with smaller chips while be able to use interposer tech to make a seamless package out of multi-chips? Clever.

I agree with your assessment.

These two consecutive sentences in the quote seem to say it quite clearly.


1) Some publications have reported that Polaris will be mix of both TSMC 16nm and Globalfoundries 14nm GPUs, which is where some of the confusion could potentially have stemmed from.


2) However, according to AMD Polaris is only 14nm.
The only other possible option is that the name Polaris is for this specific die, and I don't think it is that narrow a code name.


As to the multi die on interposer approach, join the fun. Some of us have been speculating about this quite attractive possibility. This year will be great for graphics.

 

[MrTeal]

You can't simply disable something. You do that and it all stops working. You have to code the alternative render path. When they did that they would have logically optimized it for the hardware it was written for. Not to would have been dumb.

They stated that all they had to do for AMD was to write the code as specified by MSFT and it worked on AMD. So there's the possibility that they can optimize it better for AMD. No point in doing that this early on though considering they very well might have to do multiple more rewrites before it's done.

I didn't say otherwise. I just replied to monstercameron's post saying that looking for nVidia hardware and disabling Async Compute is exactly what Oxide said they did as they couldn't rely on the driver reporting to be accurate.

 

[PPB]

I assume you're talking about the 950? Obviously nobody can confirm the Polaris GPU...

950 number seems fine to me. It's a 90W TDP GPU and i7 4690k is 88W but that i7 would barely break a sweat in that situation. People on the internet like to grossly exaggerate power requirements, so if you're going by typical PSU recommendations just realize that.

To put things into perspective, my 4790K at 4.0/1v consumed roughly 50-55 watts on bf4 64p MP with a 80fps cap and both mesh and terrain quality on ultra (the only 2 taxing settings on the cpu).

Its totally plausible they got those numbers as the entire system power consumption. Remember they may very well use some ODM only titanium psu that nets you around 95% efficiency at lower power outputs, so it is totally possible to have those 2 numbers.