![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Hey guys,
As some of you might know/remember, I am very active with Folding over with the LinusTechTips team. One of the things that comes up on the forum frequently, and there seem to be no definitive answers on, is around GPU Folding. I am hoping some of the seasoned vets here can add their two cents to the following questions:
1 - What impact does GPU Folding have on the CPUs?
I know, when installing the FAH client, it leaves one core free "for the GPU" when the conditions are correct. But how much of an impact will this make, not having any cores free? As far I know and have been able to determine, have a core free per GPU does make a slight performance increase on the GPU side, but not that much to definitively say "You must have one core free per GPU..."
2 - The more modern GPU Cores (17 and 18) are very CPU intensive as well?
3 - The more modern GPU Cores (17 and 18) need more bandwidth, in regards to the slots?
Previously I had built GPU folding rigs where I used x1 to x16 expanders to get GPU's folding in the x1 slots in my board along with the x16. There are some saying that now, with these new cores, you need the bandwidth of at least a 8x if not a 16x to get the full performance. Thoughts?
4 - Overall how much of an impact does GPU Folding have on the CPU?
It was always my understanding that 99% of the work on a WU for GPU folding was done, on the GPU. Period. Yes, having a core free could give a bit of performance boost, but not much. Slot bandwidth wasn't an issue as the WU's were actually pretty small in size. Basically a GPU WU was for the GPU. Rest didn't matter, a whole lot.
There seem to be some that believe the newer GPU cores need a lot of CPU power, bandwidth, etc. and I am just trying to clarify what people are seeing out there, as there doesn't seem to be a lot of official stuff on this. If there is some floating around there, links would also be greatly appreciated!
Thanks!
As some of you might know/remember, I am very active with Folding over with the LinusTechTips team. One of the things that comes up on the forum frequently, and there seem to be no definitive answers on, is around GPU Folding. I am hoping some of the seasoned vets here can add their two cents to the following questions:
1 - What impact does GPU Folding have on the CPUs?
I know, when installing the FAH client, it leaves one core free "for the GPU" when the conditions are correct. But how much of an impact will this make, not having any cores free? As far I know and have been able to determine, have a core free per GPU does make a slight performance increase on the GPU side, but not that much to definitively say "You must have one core free per GPU..."
2 - The more modern GPU Cores (17 and 18) are very CPU intensive as well?
3 - The more modern GPU Cores (17 and 18) need more bandwidth, in regards to the slots?
Previously I had built GPU folding rigs where I used x1 to x16 expanders to get GPU's folding in the x1 slots in my board along with the x16. There are some saying that now, with these new cores, you need the bandwidth of at least a 8x if not a 16x to get the full performance. Thoughts?
4 - Overall how much of an impact does GPU Folding have on the CPU?
It was always my understanding that 99% of the work on a WU for GPU folding was done, on the GPU. Period. Yes, having a core free could give a bit of performance boost, but not much. Slot bandwidth wasn't an issue as the WU's were actually pretty small in size. Basically a GPU WU was for the GPU. Rest didn't matter, a whole lot.
There seem to be some that believe the newer GPU cores need a lot of CPU power, bandwidth, etc. and I am just trying to clarify what people are seeing out there, as there doesn't seem to be a lot of official stuff on this. If there is some floating around there, links would also be greatly appreciated!
Thanks!