• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Server GPU for F64 Compute?

burninatortech4

Senior member
I've been noticing FirePro S9170/S9150 GPU's showing up on eBay for ~200 USD and am wondering if there's any way to repurpose these for distributed compute?


These look to be fully-enabled Hawaii GPU's with unlocked FP64 with 16GB (S9150) and 32GB (S9170) VRAM. I know the VRAM is probably wasted on projects like Milkyway but is it possible to use these in a second-hand BOINC rig? 2.619 TFLOPS (1:2) FP64 compared to the ~840 GFLOPS on my Vega 64. Nothing to sneeze at.

Considerations:
1) There doesn't seem to be an display connector on any of these cards. If I used a RX 460 I have lying around as primary would BOINC recognize a S9170 connected elsewhere in the system and use it for compute?
2) I did a quick Sketchup model and it looks like I could mount 2x 40mm server fans (high RPM) on the rear of each for "blow through" cooling. Could be a fun project - assuming they would work with BOINC.
3) The Firepro workstation version of these cards (W8100 and W9100) are still much too expensive to be worth it.

Thanks
 
I tried that route once, and I could not get the performance expected, but my seller had a generous return policy. If the seller takes returns, give it a go!
 
They work well with Milkyway@home.

One of the guys at [H] has a bunch of them, but he used 1200mm fans to cool them. He would pair them up so 1 120mm fan would handle both.

tayunz is his name, not sure if he has an account here, but he'll gladly answer any questions you may have about them if you hit him up on [H] forums.
 
Last edited:
So before I take the plunge can anyone elaborate on whether BOINC will detect the GPU if it's not connected to a monitor? It will be in the second PCI-e slot below my RX 460.

Looks like I'm able to pick up an S9150 for $125.
 
Last edited:
Oh yes, many of us have dual GPU rigs, even triple and quad GPU rigs. Only one card is connected to the monitor, but all are recognized.
 
So before I take the plunge can anyone elaborate on whether BOINC will detect the GPU if it's not connected to a monitor? It will be in the second PCI-e slot below my RX 460.

Looks like I'm able to pick up an S9150 for $125.
Where ?
 
My old AMD GPUs seemed to do pretty well on Seti. I think those old FirePros could be an interesting experiment, though I would think that they would pair best with an HD 7970 or R9 280x 290 or 290x to present similar GPU architectures to the driver.

Got the wrong GPUs in there.
 
Last edited:
And how would it do on F@H ?
S9150 is identical with R9 290X, except
  • it has 4x as many FP64 units enabled,
  • 4x as much VRAM on the card,
  • ECC enabled,
  • it is clocked slightly lower,
  • it has no own fan on the cooler, and
  • no video output port.
None of these points have any impact at all on Folding@home performance, or boinc performance in most projects. Only Milkyway@home should profit from the increased FP64 width, and probably Einstein@home too.

OCN's F@h PPD database lists R9 290X with ~260 kPPD, taken from 17 reported work units.
 
Back
Top