Polaris 10 and 11 confirmed to be GDDR5 based

[3DVagabond]

no if you think the next release is about perf/watt

How does that justify this card now?

 

[kagui]

How does that justify this card now?

you will get the same performance, but whit better upkeep cost, and you will run bigger simulations

 

[Samwell]

You know they just released a 24GB GM200 Quadro?
http://www.techpowerup.com/221113/nvidia-unveils-the-quadro-m6000-24gb-graphics-card.html

Seems strange to release a new flagship Quadro if 16nm is close to ready.

Quadro Market wants Vram. HBM2 from Samsung will just start with 4GB per stack. 8GB at end of year. That's means 16GB for a possible GP100 Quadro, which is not enough. There will be no Quadro until 32Gb cards can launch.

 

[3DVagabond]

you will get the same performance, but whit better upkeep cost, and you will run bigger simulations

so why would you buy this card now though if a more efficient Pascal is about to launch? If something better is soon to launch you don't release this.

 

[Silverforce11]

You know they just released a 24GB GM200 Quadro?
http://www.techpowerup.com/221113/nvidia-unveils-the-quadro-m6000-24gb-graphics-card.html

Seems strange to release a new flagship Quadro if 16nm is close to ready.

GP100 Quadro wouldn't be close. I doubt there's good yields on huge Pascal for them this year.

Tesla GP100 is a possibility later this year. But only after they fulfill some major contracts first.

 

[kagui]

so why would you buy this card now though if a more efficient Pascal is about to launch? If something better is soon to launch you don't release this.

yeap you are right, the new card wont give you better performance making my first point moot (thinking about time, well this point is more complex) , but you dont know when the next quadro will launch?, so if you need and upgrade now you buy this. of course this card will not take all the market, but will make you think twice about buying an amd card, if you expect a new quadro on 2017.

so yea i build that comment on several assumptions

 

[AtenRa]

This time I believe both will use smaller chips, the density difference between 28nm and 14nm is more than 2x, power consumption and leakage is way lower and both AMD and NV will have new architectures.
All those combined will bring ~250mm2 14nm chips very close to last year 28nm 500-600mm2 performance at extremely lower power.

 

[kagui]

This time I believe both will use smaller chips, the density difference between 28nm and 14nm is more than 2x, power consumption and leakage is way lower and both AMD and NV will have new architectures.
All those combined will bring ~250mm2 14nm chips very close to last year 28nm 500-600mm2 performance at extremely lower power.

yes that means 390 performance for the masses but no higher perf for the top end, but at least AMD will bring some arch refinements, NVIDiA PR only have pushed fp16 and VR DIRECT,

 

[airfathaaaaa]

This time I believe both will use smaller chips, the density difference between 28nm and 14nm is more than 2x, power consumption and leakage is way lower and both AMD and NV will have new architectures.
All those combined will bring ~250mm2 14nm chips very close to last year 28nm 500-600mm2 performance at extremely lower power.

the only official thing we have from nvidia about the maxwell v3.0 is that it will have bigger sp and dp perfomance anything else so far has been only rumor quite literally

 

[William Gaatjes]

Processing unit count between architectures is not meaningful. Maxwell had less than Kepler yet crushed it. There are lots of low hanging IPC fruit in GPUs because many times it makes more sense to build out more units than to make individual units faster. But nVidia took the other route with Maxwell and perhaps AMD is too. What matters is end performance, not paper specs.

That's not even accounting for the fact that you can count processor numbers differently based on what you're counting

Yep, i remember reading that Nvidia Kepler is more capable on the gpgpu compute side of gpu's than Maxwell and that Nvidia Maxwell was optimized for gaming graphics.

AMD might have gone the same way.

 

[ShintaiDK]

GP100 will be 50-100% faster than 980TI. The question is when a retail gaming product will hit. And if its going to be 2017 before its under 1500$.

The already shipping Knights Landing is making sure GP100 will come fast, no matter the cost. Gaming card however, up in the air. But it will be a 500-600mm2 chip with an absolute minimum of 16GB. Else Intel won the HPC segment for good. Its clear AMD is out of that race at this point.

 

[Adored]

If Nvidia is going with a smallish, polaris 10-sized die, it's the first bad mistake they've made in years. They must know how many times they got their high-end buyers to upgrade for 20%-30% last time.

Who is the market for a sub 980Ti performance at around 260mm2 and how does it pay Nvidia to release a part there? The only people interested in that performance range are AMD users and 970 owners, and the price needs to be in the 970 range for it to make sense. The new node may be expensive but not selling cards or selling them too cheaply is more so.

 

[ShintaiDK]

I for one would be a buyer for a 8GB 125W card with 980TI/Fury X performance

But unless GP104 delivers its hard to imagine. Polaris needs GDDR5X minimum at this point to even have a chance to reach it.

 

[airfathaaaaa]

I for one would be a buyer for a 8GB 125W card with 980TI/Fury X performance

But unless GP104 delivers its hard to imagine. Polaris needs GDDR5X minimum at this point to even have a chance to reach it.

i wonder when you will actually provide any link about all of those things you keep saying?
how did you manage to see a 100% increase from one flagship to another?
where did you see that polaris will need at minimum(LOL) a gddr5x?
what numbers did you saw in order to make those statements ?
does those numbers actually came from nvidia or from rumors?

 

[ShintaiDK]

Its quite clear that Polaris with 256bit GDDR5 6Ghz wont do. I already got 256bit with 7Ghz today. You are only fooling yourself. But I can understand why some people are upset. Because the alternative is they have to wait till 2017 or longer.

GP100 needs to deliver 50-100% to compete with KNL. If it doesn't KNL wins and NVidia can kiss most of their Tesla business goodbye. Its already bad enough that KNL can run as a standalone system with no memory limitations.

 

[airfathaaaaa]

Its quite clear that Polaris with 256bit GDDR5 6Ghz wont do. I already got 256bit with 7Ghz today. You are only fooling yourself. But I can understand why some people are upset. Because the alternative is they have to wait till 2017 or longer.

GP100 needs to deliver 50-100% to compete with KNL. If it doesn't KNL wins and NVidia can kiss most of their Tesla business goodbye. Its already bad enough that KNL can run as a standalone system with no memory limitations.

im sorry some posts ago you clearly stated "it will provide 50-100" not that "it needs"
then you kept saying that "it will be a 500-600mm2 with an absolute minimum of 16gb"
so i will ask again you have any link clue or something that can back that? if yes please provide us with it
also 6ghz on a gddr5x might actually be the same as a 7ghz gddr5 because the x being wider at any level doesnt really need the speed so much but a way to actually deliver the workload more effieciently
also no one has actually seen how the polaris is functioning on 14nm saying that a lower spec than 28nm is worse its like saying a laferrari is worse because it has smaller dimension than the 918

 

[ShintaiDK]

If Polaris gets GDDR5X we can talk about its performance. If its gets regular GDDR5 as everything currently points to you can forget it.

I dont think NVidia will abandon the Tesla line. And the road to success for that is clear. If they dont deliver that its gone for them. That's as simple as it gets.

 

[Leadbox]

If Polaris gets GDDR5X we can talk about its performance. If its gets regular GDDR5 as everything currently points to you can forget it.

I dont think NVidia will abandon the Tesla line. And the road to success for that is clear. If they dont deliver that its gone for them. That's as simple as it gets.

With the exception of Fiji, AMD architectures have never really gained much from memory Oc'ing. Bandwidth hasn't been the issue I don't think, using it effectively however.... A lot will obviously depend on how efficiently Polaris uses it's resources.

 

[ShintaiDK]

With the exception of Fiji, AMD architectures have never really gained much from memory Oc'ing. Bandwidth hasn't been the issue I don't think, using it effectively however.... A lot will obviously depend on how efficiently Polaris uses it's resources.

If that was so AMD wouldn't increase the memory speed on 390/390X. And Tahiti wouldn't beat Tonga.

And I am only talking 1080p. 850Mhz 280X(384bit) beating a 918Mhz 285(256bit+new memory compression). Memory bandwidth sure matters!

 

[Adored]

I for one would be a buyer for a 8GB 125W card with 980TI/Fury X performance

Given the 980 still sells north of $450, at what price does this card make sense to you and Nvidia both?

 

[Leadbox]

If that was so AMD wouldn't increase the memory speed on 390/390X. And Tahiti wouldn't beat Tonga.

And I am only talking 1080p.
https://tpucdn.com/reviews/AMD/R9_Nano/images/perfrel_1920.gif[/img.][/QUOTE]

Didn't say there was zero gain from more bandwidth and there have also been instances were Tonga beats Tahiti because of better tessellation for example. There certainly is more to the performance difference in those skus than just bandwidth.

 

[ShintaiDK]

Given the 980 still sells north of $450, at what price does this card make sense to you and Nvidia both?

Well considering there is no transistor cost savings with 14/16nm I am not going to fool myself on pricing. So 500-600$ if the Polaris 10(assume GDDR5X) or GP104 can deliver that.

 

[ShintaiDK]

Didn't say there was zero gain from more bandwidth and there have also been instances were Tonga beats Tahiti because of better tessellation for example. There certainly is more to the performance difference in those skus than just bandwidth.

Yes, we can always find some cases. But lets be honest, its not going to work without the bandwidth. And to say AMD doesn't gain from more memory bandwidth was outright silly.

 

[Adored]

Well considering there is no transistor cost savings with 14/16nm I am not going to fool myself on pricing. So 500-600$ if the Polaris 10(assume GDDR5X) or GP104 can deliver that.

You would pay over $500 for 15%-20% over a 980? Surely these people already own 980's...

 

[ShintaiDK]

You would pay over $500 for 15%-20% over a 980? Surely these people already own 980's...

Fury X/980TI is more than 15-20% over a 980 when resolution goes up. And remember, 125W. So in short yes, I would pay 500-600$ for a 125W Fury X/980TI with 8GB.