Nvidia Pascal GPU cards???

[JimboPalmer]

No.

How could we know how much of a performance boost they will have if they are not yet for sale?
Once they ARE for sale we will know what they are like with the current cores and drivers. In six months (to two years) we will know what they are like with optimized cores and drivers.

As to where they will be on sale, my guess is retailers.
The exciting part is not how much they will cost but how much cheaper they will make the current cards.

[7im]

The folding forum is hardware agnostic. And until the product has been released to the market, any answers would be purely speculation and more unhelpful than helpful.

NVidia product details are found at their site: http://www.nvidia.com

Web search tools might also be helpful if you are looking for speculative answers.

[Sven]

Here some specs of the GeForce GTX 1080

http://www.geforce.com/hardware/10serie ... e-gtx-1080
http://www.anandtech.com/show/10304/nvi ... -1080-1070

Looks interesting, except for the price.

[DocJonz]

Stanford have worked closely with nVidia in the past, though I don't know if that is still the case.
Would be nice to know whether these could be used for Folding on release, or whether months of FAH code development would be needed - I expect the latter.

[toTOW]

We don't know until someone gets a card and gives it a try ...

[kiore]

Here some specs of the GeForce GTX 1080

http://www.geforce.com/hardware/10serie ... e-gtx-1080
http://www.anandtech.com/show/10304/nvi ... -1080-1070

Looks interesting, except for the price.

It is impressive, both in performance and price. It'll take me a long time to save up for one, I doubt there'll be a price break anytime soon. I do hope and pray it'll increase my point production! I only wish everyone had one of these (if a frog had wings...).

Less shaders than a GTX 980ti or Titan, and a much lower launch price than either of these. All theoretical at present.

[beer]

One thing that disappoint me a bit is the TDP has gone up from 165 W to 180 W

[toTOW]

True ... but you can also compare it with 980Ti/Titan X and consider that it went down from 250 to 180W ...

[MarkyMark7]

The rumor mill at http://wccftech.com/nvidia-geforce-gtx-1080-launch/ says $599 for the GTX 1080 (reported in several other sources too), available May 27, 2016, and $379 for the GTX 1070 available June 10, 2016.

Code: Select all

              
                   GTX 1070   GTX 1080
CUDA Cores         2048       2560
Memory             GDDR5      GDDR5X  (way faster in theory)

Now I need to crack open my PC to see if I have an 8 pin power tap. I know I have some power tap hooked to my GTX 645, don't remember how many pins.

[7im]

Less shaders than a GTX 980ti or Titan, and a much lower launch price than either of these. All theoretical at present.

And how shaders from one generation to the next can change in performance significantly (or not) so even shader counts may not be a good way to make estimates. Just have to wait two weeks for the real thing. Then let the GPU chart at Anandtech tell the tale.

[FldngForGrandparents]

I have 980 and 980 Ti OC's and am purchasing a 1080 as soon as they are available. Maybe even two of them if I can. The clock speeds are impressive over both of my current cards and the lower power is a plus. The good thing about this is it will drive the 980/980 Ti's into the mid to low $300 range. Based on my calculations I would not be surprised to see 1M PPD. I will be testing these in PCI-E 1x with riser cables as well to see if there is any major performance loss. I hope to run six per server in the coming months.

[PS3EdOlkkola]

@ FldngForGrandparents Of course, we can speculate all day long about 1080 performance, but if history is any guide, the performance of a high-powered GPU is curtailed when limited by a 4x PCIe bus, never mind a 1x PCIe bus. You may want to wait before you commit to purchasing more than two 1080 GPUs to first determine how much of a load they place on your PCIe bus. For Nvidia GPUs, one of my systems has 3 Titan X's installed and three systems each have three 980ti's installed. All are using motherboards with 40 PCIe lanes (2011 and 2011-v3) and a minimum of 4 core/8 thread 3.5GHz CPUs. There was a drop in overall PPD of about 5% to 8% of what I would have expected out of each GPU when the third GPU was put in each machine, and those had a minimum 3.0 PCIe lane width of 8. Putting 6 GPUs in one machine, each on a 1x PCIe interface may not get you the performance your expecting, if history repeats itself.

[FldngForGrandparents]

@ FldngForGrandparents Of course, we can speculate all day long about 1080 performance, but if history is any guide, the performance of a high-powered GPU is curtailed when limited by a 4x PCIe bus, never mind a 1x PCIe bus. You may want to wait before you commit to purchasing more than two 1080 GPUs to first determine how much of a load they place on your PCIe bus. For Nvidia GPUs, one of my systems has 3 Titan X's installed and three systems each have three 980ti's installed. All are using motherboards with 40 PCIe lanes (2011 and 2011-v3) and a minimum of 4 core/8 thread 3.5GHz CPUs. There was a drop in overall PPD of about 5% to 8% of what I would have expected out of each GPU when the third GPU was put in each machine, and those had a minimum 3.0 PCIe lane width of 8. Putting 6 GPUs in one machine, each on a 1x PCIe interface may not get you the performance your expecting, if history repeats itself.

Well I can tell you what your performance issue is. A motherboard with 40 PCIe lanes is just what it supports. Intel's PCI-e lanes are based on CPU. And a 4 core CPU more likely only has 16 lanes. Maybe even less with a lower CPU. Thus 3 cards plus whatever else share that. So that max you could do is 16/0/0 or 8/4/4 speed, etc. I use AMD which has it's PCI-e lanes handled via the chipset instead of the CPU. I have the Gigabyte 990Fx-UD5 which has 32 lanes. So 16/4/4/4/4 or 8/8/4/4 and other combinations. I have tested with 1x and it does work with some performance hits. But this was with Nvidia GPU which uses more CPU communication than AMD it seems. I have some more hardware coming in today to do some additional testing and will post updated info.

[PS3EdOlkkola]

All 2011 and 2011-v3 motherboards have CPUs supporting 40 PCIe lanes. Configurations are generally 16/16/8. I also have one remaining AMD system using an FX-8350 with exactly the motherboard you mention, and on Titan X or 980ti Nvidia GPUs, it performs at best at 85% of the PPD of a 2011 based Intel system. It is currently relegated to running three R9 Fury X GPUs, as the performance hit is far less than when running Nvidia GPUs. I've experimented with many different combinations of motherboards, CPUs and GPUs, and have 14 systems running 41 slots reliably and with very good performance. Just trying to give you the benefit of some hard won and expensive experience.

[Foxbat]

After the new Pascal cards become available, I'm assuming Stanford needs to put them through some sort of validation testing procedure to ensure good science. They might make awesome gaming cards, but does that mean that they will necessarily make just as awesome Folding devices?

As someone who has held off getting a GTX 980 Ti card since Pascal was just around the corner, being the first on the block with the newest toy may not translate to leaping to the top of your Folding Team's stats, especially if the GP104 doesn't make it on to the GPUs.txt Whitelist for a few months.