P106 for Folding@home?

[Furkell]

I guess everyone has watched the latest LTT video:

 

This made me think, can these GPUs be used for folding? I don't see why it shouldn't have the exact same performance as a normal GTX 1060 6GB, and the price difference is well worth it.

I have searched online but haven't seen anyone confirm that they work. Someone with one of these that can test it?

[Furkell]

@GabenJr do you have any free time to test it?

[Nicnac]

I've seen a few of those for sale here as well and thought about it. could be well worth it. Only thing to keep in mind is that they will have no resale value pretty much.

[Bananasplit_00]

should be able too yah, they were used for mining, F@H is pretty much the same thing but for reserch

[Emily Young]

23 minutes ago, Furkell said:

@GabenJr do you have any free time to test it?

I haven't tested it, nor am I sure when I'd have time to, but I would go so far as to say it would probably work even without the modified driver, since the mining cards should still have CUDA / OpenCL compute enabled by default. With the driver, the cards work in WDM mode and can be used with DirectX / OpenGL rendering, just obviously without any native outputs.

[Toru_The_Guardian]

Heyo,

 

I happen to have a p104-100 (1070 version) for the exact purpuse that LTT made a vid about. (In a VM, now use it for space heating)

 

I installed Folding@ home and can confirm that you don't need any bios or driver mods to fold.

 

Hope this image make more sense to you than it does to me.

 

If you wan't me to test something else, feel free to ask :)

 

__ David

[Guest]

17 hours ago, GabenJr said:

I haven't tested it, nor am I sure when I'd have time to, but I would go so far as to say it would probably work even without the modified driver, since the mining cards should still have CUDA / OpenCL compute enabled by default. With the driver, the cards work in WDM mode and can be used with DirectX / OpenGL rendering, just obviously without any native outputs.

Make @jakkuh_t test it.

We know he has nothing better to do.

[Emily Young]

39 minutes ago, King Poet said:

Make @jakkuh_t test it.

We know he has nothing better to do.

It's true though.

[RollinLower]

2 hours ago, Toru_The_Guardian said:

Heyo,

 

I happen to have a p104-100 (1070 version) for the exact purpuse that LTT made a vid about. (In a VM, now use it for space heating)

 

I installed Folding@ home and can confirm that you don't need any bios or driver mods to fold.

 

Hope this image make more sense to you than it does to me.

 

If you wan't me to test something else, feel free to ask

 

__ David

Can you maybe check the PPD this card is putting out? So I can compare it to my regular 1070.

You should be able to see the PPD when you go to the advanced control (should be in your icon tray, right click and you should be able to open the advanced one)

[Toru_The_Guardian]

17 hours ago, RollinLower said:

Can you maybe check the PPD this card is putting out? So I can compare it to my regular 1070.

You should be able to see the PPD when you go to the advanced control (should be in your icon tray, right click and you should be able to open the advanced one)

Just checked, the estimated PPD says: 86878 (the GPU runs @ stock settings 1860mhz consistent clockspeed)

 

It is also worth noting that the card has 4gb gddr5x instead of 8gb gddr5.

 

Edit: and now it says 127200.... Will leave it running for a while to see where this goes

Edited January 16, 2019 by Toru_The_Guardian

[RollinLower]

1 hour ago, Toru_The_Guardian said:

Just checked, the estimated PPD says: 86878 (the GPU runs @ stock settings 1860mhz consistent clockspeed)

 

It is also worth noting that the card has 4gb gddr5x instead of 8gb gddr5.

 

Edit: and now it says 127200.... Will leave it running for a while to see where this goes

The VRAM is kinda interesting. My 1070 gets about 550,000 to 600,000 PPD depending on the WU.

Please let me know what your card puts out!

[Toru_The_Guardian]

25 minutes ago, RollinLower said:

The VRAM is kinda interesting. My 1070 gets about 550,000 to 600,000 PPD depending on the WU.

Please let me know what your card puts out!

I could test again but with the same settings that you are using to see if there is a big difference.

 

If you wan't me to test it, just let me know the settings/workload and I'll see what I can do

[RollinLower]

3 hours ago, Toru_The_Guardian said:

I could test again but with the same settings that you are using to see if there is a big difference.

 

If you wan't me to test it, just let me know the settings/workload and I'll see what I can do

right now i am getting 630174 PPD on the 1070, running +150 on the core and +550 on the memory.

i didn't configure anything in F@H tough, so it just grabs the next WU it can find.

[Curlay]

Just got my hands on a P106-6G today! Will fire it up tonight and let you know how it goes. Interested in performance vs the P104

[Gorgon]

22 hours ago, Toru_The_Guardian said:

Just checked, the estimated PPD says: 86878 (the GPU runs @ stock settings 1860mhz consistent clockspeed)

 

It is also worth noting that the card has 4gb gddr5x instead of 8gb gddr5.

 

Edit: and now it says 127200.... Will leave it running for a while to see where this goes

Those numbers seem really low. VRAM doesn’t really matter for folding and at that clock you should be getting 600,000 to 700,000PPD.

 

the weird thing is that 9031 Work Unit is a GRO_A4 core which are usually for CPU folding???

 

Very strange.

[Toru_The_Guardian]

3 hours ago, Gorgon said:

Those numbers seem really low. VRAM doesn’t really matter for folding and at that clock you should be getting 600,000 to 700,000PPD.

 

the weird thing is that 9031 Work Unit is a GRO_A4 core which are usually for CPU folding???

 

Very strange.

Interesting.... I was working on project 14127 if that matters.

 

I'll make sure I have all the latest drivers etc. installed and run it again.

 

It is also worth noting that I do not use my GPU in a standard environment. I've got it passed through to a Windows Server 2019 VM it could be possible that this has an adverse effect on the performance.

 

Also, I'm not completely sure if the card runs at pci-e3 16x. It could also be 4x, does this impact F@H perf?

[Toru_The_Guardian]

On 1/16/2019 at 1:18 PM, RollinLower said:

The VRAM is kinda interesting. My 1070 gets about 550,000 to 600,000 PPD depending on the WU.

Please let me know what your card puts out!

I think I found out why my card puts out so little PPD. The VRAM is pinned at 100% usage.

 

My assumption is that the F@H software thinks it is an regular 1070 with 8gb of ram and tries to use that.

 

Assuming this is the issue, I think that the 1060-6G will perform near native performance.

[Gorgon]

3 hours ago, Toru_The_Guardian said:

Interesting.... I was working on project 14127 if that matters.

 

I'll make sure I have all the latest drivers etc. installed and run it again.

 

It is also worth noting that I do not use my GPU in a standard environment. I've got it passed through to a Windows Server 2019 VM it could be possible that this has an adverse effect on the performance.

 

Also, I'm not completely sure if the card runs at pci-e3 16x. It could also be 4x, does this impact F@H perf?

14127? The screenshot showed a 9031. Must have been a later WU.

 

PCIe throughput does make a difference, more so under Windows than Linux. Only 4 Lanes may impact performance for a 104, especially virtualized. But I would expect, un virtualized, only a 5-10% max hit.

 

Are these 4 lanes also from the PCH rather than the CPU and is it possible a NVMe module might be also using some of those lanes?

 

Unconstrained, at 1800MHz, the 104 should be able to do 600-700KPPD.

 

The VRAM utilization at 100% is also suspicious. I’ve never seen more than 150 - 400MB used when folding but I’m running Ubuntu natively with current 300 and 400 series drivers. During the Folding Month when I was folding on Win10 64 Pro and using the system for light desktop productivity I was seeing 1 to 2.5 GB VRAM utilization.

 

What are you using as a Hypervisor?

[Toru_The_Guardian]

20 minutes ago, Gorgon said:

14127? The screenshot showed a 9031. Must have been a later WU.

 

PCIe throughput does make a difference, more so under Windows than Linux. Only 4 Lanes may impact performance for a 104, especially virtualized. But I would expect, un virtualized, only a 5-10% max hit.

 

Are these 4 lanes also from the PCH rather than the CPU and is it possible a NVMe module might be also using some of those lanes?

 

Unconstrained, at 1800MHz, the 104 should be able to do 600-700KPPD.

 

The VRAM utilization at 100% is also suspicious. I’ve never seen more than 150 - 400MB used when folding but I’m running Ubuntu natively with current 300 and 400 series drivers. During the Folding Month when I was folding on Win10 64 Pro and using the system for light desktop productivity I was seeing 1 to 2.5 GB VRAM utilization.

 

What are you using as a Hypervisor?

Oh yeah, I just noticed the difference yeah. Gues it switched to another project as my server was restarted.

 

As for the pci-e lanes, it is in a 16x slot that is controlled by cpu1 no PCH or NVMe problems.

 

Furthermore, there are some interesting quirks when running the GPU virtualized. For instance, starting GPU-z instantly crashes the VM. This also leads me to beleive that the readouts from MSI Afterburner are not accurate.

 

I might place the card in my desktop to see what kind of results I would get. Doing this would also eliminate the virtualization layer and give me more direct controll over the GPU. (and readouts) I remember being able to run GPU-z that way, perhaps I can find out some more info about PCI-e lanes etc.

 

I use Proxmox as a Hypervisor.

 

[Gorgon]

7 hours ago, Toru_The_Guardian said:

Oh yeah, I just noticed the difference yeah. Gues it switched to another project as my server was restarted.

 

As for the pci-e lanes, it is in a 16x slot that is controlled by cpu1 no PCH or NVMe problems.

 

Furthermore, there are some interesting quirks when running the GPU virtualized. For instance, starting GPU-z instantly crashes the VM. This also leads me to beleive that the readouts from MSI Afterburner are not accurate.

 

I might place the card in my desktop to see what kind of results I would get. Doing this would also eliminate the virtualization layer and give me more direct controll over the GPU. (and readouts) I remember being able to run GPU-z that way, perhaps I can find out some more info about PCI-e lanes etc.

 

I use Proxmox as a Hypervisor.

 

Thanks, if these cards run well and can be had at a reasonable cost I’ll likely pick up a couple to replace some GTX 1060s which I can sell.

[Toru_The_Guardian]

I finally had the time to put the card in my desktop and was able to grab some info via GPU-Z. I think the following screenshot basically explains the low PPD as it shows to run @ PCIe1.1 4x....

Running it in my desktop has the same performance as in the server....

[RollinLower]

2 hours ago, Toru_The_Guardian said:

I finally had the time to put the card in my desktop and was able to grab some info via GPU-Z. I think the following screenshot basically explains the low PPD as it shows to run @ PCIe1.1 4x....

Running it in my desktop has the same performance as in the server....

That's strange, it's a mining card tough and they really don't need the bandwidth, so maybe this is intentionally put there to drop manufacturing costs?

Can you maybe force it to run on a higher PCIe in the bios?

[Gorgon]

21 hours ago, Toru_The_Guardian said:

I finally had the time to put the card in my desktop and was able to grab some info via GPU-Z. I think the following screenshot basically explains the low PPD as it shows to run @ PCIe1.1 4x....

Running it in my desktop has the same performance as in the server....

That’s weird as if it is electrically, on the card, wired for PCIex4 then it would be missing most of the gold fingers towards the back of the card, or, if they are there, would not have traces connected. Most manufacturers, however, when doing this omit the fingers to save on the gold for the plating. And the 1.1 would normally be a limit on the motherboard not the card as I can’t see GigaByte gimping the PCI bus controller or using a 5 or 6 year old I.C. Intentionally.

[Gorgon]

Any further results?

[ZexMaxwell]

I bought a P106-090 to up my folding@home game. without any modifications to drivers and registry, F@H controller sees the GPU and works. 

 

its already known that Nvidia disable driver support on newer versions. but what i would like to do is to re-enable support for physX, since i already have a dedicated GPU for gaming. it would be even better to get the P106 onto Precision X1 for fan control and/or temperature monitoring.