[ Various] NVIDIA could launch Pascal-based mobile GPUs by end-2016

[AtenRa]

GM206 is much faster than Pitcarn: http://www.hardware.fr/articles/945-6/consommation-efficacite-energetique.html

And GM206 is much more advanced. Pitcairn is missing a lot of features which would make the die much bigger.

According to latest TPU gaming suite, R7 270X (Full Pitcairn) is equal to GTX960 at 1080p.

http://www.techpowerup.com/reviews/Sapphire/R9_390_Nitro/23.html

 

[PPB]

GM206 is much faster than Pitcarn: http://www.hardware.fr/articles/945-6/consommation-efficacite-energetique.html

And GM206 is much more advanced. Pitcairn is missing a lot of features which would make the die much bigger.

Yeah, perf/watt at 1440p is really relevant to the discussion of 220mm2 28nm chips.

Ninja'ed chart by AtenRa

Like Async Compute? Yep, Pascal has a lot of stuff to add in too considering how "advanced" Maxwell was :awe:

 

[sontin]

Ha, wrong link: http://www.hardware.fr/articles/945-23/recapitulatif-performances.html

Yeah, "Async Compute". Pitcarn doesnt even support Feature Level 12_0. But i guess you dont care about DX12 at all, right?

And techpowerup doesnt show the performance numbers in the various game benchmarks. It just there in the average one. And no, Pitcairn is not as fast as GM206.

It doesnt even beat the reference GTX960 in Dirt:Rally or Battlefront.

 

[AtenRa]

Just to point out that Sontin's hardware.fr link only has the R7 370 which is not the Full Pitcairn.

 

[sontin]

R7 370 uses the full Pitcarn chip: http://www.amd.com/en-us/products/graphics/desktop/r7#

 

[PPB]

Ha, wrong link: http://www.hardware.fr/articles/945-23/recapitulatif-performances.html

Yeah, "Async Compute". Pitcarn doesnt even support Feature Level 12_0. But i guess you dont care about DX12 at all, right?

And techpowerup doesnt show the performance numbers in the various game benchmarks. It just there in the average one. And no, Pitcairn is not as fast as GM206.

It doesnt even beat the reference GTX960 in Dirt:Rally or Battlefront.

I care about DX12, and FL11.1 doesnt mean bad DX12 performance. No Async Compute on games that make use of it on the other hand...

 

[sontin]

In Ashes the GTX960 is 26-30% faster than the r7 370: http://www.computerbase.de/2015-08/...idia/2/#diagramm-normale-anzahl-an-draw-calls

In Fable Legends the GTX960 is 16% faster: http://techreport.com/review/29090/fable-legends-directx-12-performance-revealed/2

So, where exactly is Pitcarn as fast as GM206?

 

[maddie]

For the second time in a few minutes, I say again im not comparing Polaris to whatever NVIDIA will bring at 16nm. Im comparing a 125mm2 Polaris at 14nm vs Tonga.

They are also not accounting for the fab density changes.

Nvidia = TSMC 28nm > TSMC 16nm 1>2.0

AMD = TSMC 28nm > Samsung 14nm 1>2.3

https://www.semiwiki.com/forum/content/3884-who-will-lead-10nm.html

 

[Samwell]

Were you at Supercomputing 2015, or do you have a link to where nVidia engineers were discussing it?

Friend of mine was there, i'm not in this business.
You can hear it in this talk mentioned, that they have it in lab:
http://images.nvidia.com/events/sc15/SC5102-path-exascale-computing.html

Also discussing two new techniques which should be implemented in pascal.

 

[USER8000]

R7 370 uses the full Pitcarn chip: http://www.amd.com/en-us/products/graphics/desktop/r7#

No it doesn't - look at the spec sheet. It says 1024 shaders and Pitcairn has 1280 shaders.

The R9 270X has 25% more shaders.

Edit to post.

The R9 270X and GTX960 are compared here in Fable:

http://www.pcgameshardware.de/DirectX-12-Software-255525/Specials/Spiele-Benchmark-1172196/

They are about the same at 1080P.

 

[PPB]

No it doesn't - look at the spec sheet. It says 1024 shaders and Pitcairn has 1280 shaders.

The R9 270X has 25% more shaders.

Edit to post.

The R9 270X and GTX960 are compared here in Fable:

http://www.pcgameshardware.de/DirectX-12-Software-255525/Specials/Spiele-Benchmark-1172196/

They are about the same at 1080P.

Some people just cant take the 960, being 3 years newer that Pitcarin XT, can only match its performance while consuming 16% less.

They even argue AMD cant make on a new node a SKU with similar shader count, a little more perfomance and be in the 50w range like Polaris 10 will be. Its the LEAST they should achieve with a new uarch aimed at efficiency and a new node.

I can totally see a 1280 shader Polaris 10 with 380 performance aimed at the notebook and "I have a crappy psu, cant handle more than motherboard PCI-E power" crowd just like the 750/ti was.

But what is important this time around is that AMD mantains it's perf/watt ratio more or less in their whole lineup. Tahiti perf/watt was horrible compared to Pitcarin and that showed when fanboys needed an excuse to wait 3 months for 680. Not going above the sweet spot clock/voltage curve and better binning will be important this time around.

 

[sontin]

No it doesn't - look at the spec sheet. It says 1024 shaders and Pitcairn has 1280 shaders.

The R9 270X has 25% more shaders.

Edit to post.

The R9 270X and GTX960 are compared here in Fable:

http://www.pcgameshardware.de/DirectX-12-Software-255525/Specials/Spiele-Benchmark-1172196/

They are about the same at 1080P.

Ah, thank you. I thought it would be only 1024 shaders. :thumbsup:

 

[USER8000]

Ah, thank you. I thought it would be only 1024 shaders. :thumbsup:

TBH,AMD and Nvidia naming schemes for the last few years have been rather crap at times,so its easy to make a mistake here.

Some people just cant take the 960, being 3 years newer that Pitcarin XT, can only match its performance while consuming 16% less.

They even argue AMD cant make on a new node a SKU with similar shader count, a little more perfomance and be in the 50w range like Polaris 10 will be. Its the LEAST they should achieve with a new uarch aimed at efficiency and a new node.

I can totally see a 1280 shader Polaris 10 with 380 performance aimed at the notebook and "I have a crappy psu, cant handle more than motherboard PCI-E power" crowd just like the 750/ti was.

But what is important this time around is that AMD mantains it's perf/watt ratio more or less in their whole lineup. Tahiti perf/watt was horrible compared to Pitcarin and that showed when fanboys needed an excuse to wait 3 months for 680. Not going above the sweet spot clock/voltage curve and better binning will be important this time around.

The thing is even if it was 1024 shaders or 1280 shaders,we don't know how much the design of the shaders has changed,as we saw Maxwell SKUs with less shaders beat out Kepler.

So,will be interesting if AMD has made changes in a similar way. I still think we will get around GTX960/R9 380 level performance if it is under 150MM2.

I think AMD needs to be careful pricing these and not go with stupid pricing.

At worst they should keep pricing at GTX960/R9 380 levels or if they were clever a bit below,ie,more like GTX950 level pricing.

I just hope they do not screw up the launch.

 

[PPB]

TBH,AMD and Nvidia naming schemes for the last few years have been rather crap at times,so its easy to make a mistake here.



The thing is even if it was 1024 shaders or 1280 shaders,we don't know how much the design of the shaders has changed,as we saw Maxwell SKUs with less shaders beat out Kepler.

So,will be interesting if AMD has made changes in a similar way. I still think we will get around GTX960/R9 380 level performance if it is under 150MM2.

I think AMD needs to be careful pricing these and not go with stupid pricing.

At worst they should keep pricing at GTX960/R9 380 levels or if they were clever a bit below,ie,more like GTX950 level pricing.

I just hope they do not screw up the launch.

Thats why I said 1280 shaders with 380 (1792) performance. Also the IPC gains from Maxwell were exagerated a la SB vs Nehalem: remember clocks went up a lot (specially in GM204 vs GK110 comparison).

I would rather see a wide but efficient architecture, and leave the high clocks as OC margin, something AMD had with launch Tahiti for example but had to start binning their skus higher and higher.

 

[tential]

Thats why I said 1280 shaders with 380 (1792) performance. Also the IPC gains from Maxwell were exagerated a la SB vs Nehalem: remember clocks went up a lot (specially in GM204 vs GK110 comparison).

I would rather see a wide but efficient architecture, and leave the high clocks as OC margin, something AMD had with launch Tahiti for example but had to start binning their skus higher and higher.

who cares what the clocks are?

Maxwell has high clocks, and still lower power consumption than the lower clocked AMD GPUs.

All that matters to me is the end result.... and the end result is that Maxwell clocked high, and OCed even higher.

So no, I'd rather Nvidia do whatever works best for them. Clock low, or high, just do what they do to make performance happen.

 

[maddie]

who cares what the clocks are?

Maxwell has high clocks, and still lower power consumption than the lower clocked AMD GPUs.

All that matters to me is the end result.... and the end result is that Maxwell clocked high, and OCed even higher.

So no, I'd rather Nvidia do whatever works best for them. Clock low, or high, just do what they do to make performance happen.

I believe he was referring to AMD's approach to GPU design, not Nvidia's.

 

[Glo.]

I saw a post that showed 3 new GPU's in AMD's drivers. I can't find it now, but will continue trying. There are rumors, despite what Raja is saying that there are 3 "Polaris" GPU's in house now.

If I understand what has been already said: Polaris 10 and 11 are two new Polaris GPUs that Raja talked about.

The one other GPU is Vega 10, and according to scheme of naming it may be different architecture GPU.

 

[Arachnotronic]

If I understand what has been already said: Polaris 10 and 11 are two new Polaris GPUs that Raja talked about.

The one other GPU is Vega 10, and according to scheme of naming it may be different architecture GPU.

That may very well be the case. I really hope it's not too long before we see Vega 10, this looks like the enthusiast GPU we all want...

 

[Techhog]

That may very well be the case. I really hope it's not too long before we see Vega 10, this looks like the enthusiast GPU we all want...

It'll be 2017.

 

[Arachnotronic]

It'll be 2017.

I figured that from the "we have 2 Polaris chips coming" but I'm hoping we're talking early 2017 availability and not like, mid-2017.

 

[maddie]

The quotation from the Venturebeat article.

VB: Is that with a generation coming in 2016?
Koduri: Yes. We have two versions of these FinFET GPUs. Both are extremely power efficient. This is Polaris 10 and that’s Polaris 11. In terms of what we’ve done at the high level, it’s our most revolutionary jump in performance so far

Two versions of these FinFet GPUs, not two GPUs. Polaris 10 and Polaris 11. How to interpret that is up to you.

Maybe one version uses GDDR5 and the other uses HBM.

Our most revolutionary jump in performance so far, implies a high end offering, at least to me.

 

[Techhog]

The quotation from the Venturebeat article.




Two versions of these FinFet GPUs, not two GPUs. Polaris 10 and Polaris 11. How to interpret that is up to you.

Maybe one version uses GDDR5 and the other uses HBM.

Our most revolutionary jump in performance so far, implies a high end offering, at least to me.

I'm pretty sure that they said that they'll have two FinFET GPUs in 2016 before even CES. And no, they definitely don't mean two versions of each GPU.

 

[maddie]

I'm pretty sure that they said that they'll have two FinFET GPUs in 2016 before even CES. And no, they definitely don't mean two versions of each GPU.

If the transcript of the interview is correct, the quote by Koduri, in my previous post, is an exact clip from the article.

I did not claim he said two versions of each GPU.

He said two versions of finFET GPUs. This is Polaris 10 and that's Polaris 11.

I make no claim as to if they have several versions of each. Please don't change what I wrote.

 

[Silverforce11]

Another article suggest NV's Pascal is due in the 2nd half of 2016, with validation of Pascal to be done in this first half. Samsung will be supplying them with HBM2 chips.

Looks like SK Hynix is supplying AMD, hence the exclusive rumor awhile back.

ST Micro is doing HMC (for Intel?).

http://www.kitguru.net/components/g...-gpus-rumoured-to-launch-in-2nd-half-of-2016/

The source is a Korean tech news, they have discussions with folks in Samsung to get the article with quoted sources from those in the industry.

This suggests NV is at least 1 quarter behind AMD, if AMD's plans to launch in Q2 isn't delayed.

Here's the interesting part, since the mid-range chip is unlikely to use HBM2, it means NV will be having a big Pascal w/ HBM2 sometime in 2H 2016. Likely Tesla, but maybe consumer Titan class!

IF AMD only has 2 Polaris in 2016, small and mid-range, they got nothing to compete with big Pascal. > This is potentially a major blow to AMD if NV release consumer Pascal Titan in late 2016, it's the perception, NV is supreme leader in graphics that will be very damaging. <

IMO, if AMD isn't working on a big Polaris for 2016 release, they will be in a world of hurt.

 

[3DVagabond]

Another article suggest NV's Pascal is due in the 2nd half of 2016, with validation of Pascal to be done in this first half. Samsung will be supplying them with HBM2 chips.

Looks like SK Hynix is supplying AMD, hence the exclusive rumor awhile back.

ST Micro is doing HMC (for Intel?).

http://www.kitguru.net/components/g...-gpus-rumoured-to-launch-in-2nd-half-of-2016/

The source is a Korean tech news, they have discussions with folks in Samsung to get the article with quoted sources from those in the industry.

This suggests NV is at least 1 quarter behind AMD, if AMD's plans to launch in Q2 isn't delayed.

Here's the interesting part, since the mid-range chip is unlikely to use HBM2, it means NV will be having a big Pascal w/ HBM2 sometime in 2H 2016. Likely Tesla, but maybe consumer Titan class!

IF AMD only has 2 Polaris in 2016, small and mid-range, they got nothing to compete with big Pascal. > This is potentially a major blow to AMD if NV release consumer Pascal Titan in late 2016, it's the perception, NV is supreme leader in graphics that will be very damaging. <

IMO, if AMD isn't working on a big Polaris for 2016 release, they will be in a world of hurt.

Yeah. nVidia will own the $1000+ consumer GPU market.