is my Pascal Titan X being limited in F@H performance due to PCIe 2.0?

[RollinLower]

hello,

 

so the question is in the title really. my current build is in my signature. eventough my CPU and my motherboard support PCIe 3.0, a firmware update from intel limits all E5-V1 CPU's to be ran at PCIe 2.0 speeds.

 

my titan X is getting about 900K PPD running on 2075Mhz, while other users report 1.3M+ PPD on these cards with such overclocks. could it be that it is being choked through the PCIe bandwith? i have done some reasearch and in the folding process the GPU is being fed with blocks of data to process constantly over the PCI bus. so i guess it would make sense?

 

any help is greatly appreciated!

Edited March 30, 2018 by RollinLower
typos...

[Radium_Angel]

If you look at the raw throuoghput for PICe 2.0 here:

http://www.tested.com/tech/457440-theoretical-vs-actual-bandwidth-pci-express-and-thunderbolt/

 

It sort of seems unlikely, it's hard to saturate a PCIe bus, at least for video games (tests have shown, for example, that the difference between an x16 PCIe slot and an x8 slot, for gaming, is a rounding error. In other words, virtually nothing)

 

I'm not an expert on F@H so take this advice for what it's worth, but I would think the amount of data being sent to your client model would also be dependent on network bandwidth and, I presume, whatever else the system is doing at the time. In other words, I would be highly surprised that the PCIe 2.0 is the culprit here

[RollinLower]

well folding takes a lot more performance out of the PCI bus then a game, this is why i think it might be the limiting factor here.

network is through cable and not a problem at all, the network is only ever used to download the model and when the folding is complete send the results. only other things running on the system are speedfan and MSI afterburner, so really nothing at all.

 

the thing is, i only ever see these tests for gaming scenario's, while in a F@H scenario the results would be very different. i was hoping someone would have a link to an article about this without the focus exclusively on gaming peformance, but more on raw compute perfromance.

[cj09beira]

unless a huge part of the computation is done in the cpu there is no reason for the pcie bus to be a limiting factor, as all it needs to do is to load the data and retrieve the results, which should take much less time than the time the gpu takes computing the data,

now that i think about it there is loads of people running F@H and boinc machines with 4-6 cards per system which means that they have at least some cards running with less pcie lanes, as low as 1x connections, with almost no perf hit. 

[RollinLower]

5 minutes ago, cj09beira said:

unless a huge part of the computation is done in the cpu there is no reason for the pcie bus to be a limiting factor, as all it needs to do is to load the data and retrieve the results, which should take much less time than the time the gpu takes computing the data,

now that i think about it there is loads of people running F@H and boinc machines with 4-6 cards per system which means that they have at least some cards running with less pcie lanes, as low as 1x connections, with almost no perf hit. 

well, according to the Folding@Home forums:

Quote

You are misunderstanding how the GPU is used during folding. The folding core running on the CPU prepares blocks of data to be processed on the GPU, receives them back from the GPU and sends new blocks for processing almost continuously over the PCIe bus. How frequently the data is transferred depends on how fast the GPU processes each batch. The bandwidth is used all of the time, and not just during the initial stage. More bandwidth is used by a fast GPU like a GTX 980 than by a slower one such as a GTX 750 ti.

so according to this logic my fast GPU should use tons of bandiwth

[cj09beira]

8 minutes ago, RollinLower said:

well, according to the Folding@Home forums:

so according to this logic my fast GPU should use tons of bandiwth

you can do 2 things one is look for some sort of benchmark the other is to try on another slot with even lower bandwidth and see how much difference it makes 

[RollinLower]

55 minutes ago, cj09beira said:

you can do 2 things one is look for some sort of benchmark the other is to try on another slot with even lower bandwidth and see how much difference it makes 

did the first thing, all benchmarks i found are with slower older GPU's or gaming oriented.

the titan is currently in a custom waterloop, so switching slots isn;t really an option either  

 

i was hoping someone would have some concrete info on this

[RollinLower]

@cj09beira @Radium_Angel

i mean, the card is running at PCIe 2.0 X8 right now, getting only 700K PPD. that is pretty much 50% of what it should be getting. 

i guess the jump to a Xeon E5 V2 is needed?

[RollinLower]

Bump?