Folding Forum

Hello,

im curios if a WU with the same Base Credit but different Run, Clone, Gen are easy to compare or not.

Background info: I now own two GTX770, one running on a Linux- the other on a Windows-Computer.
Theyre both running a 17800 WU, same Base Credit. The 770 on Windows is around 2000 Points lower in estimated credit.
This translates into ~33000 PPD lost.

But is it even comparable to crunch 27, 118, 259 vs 45, 75, 303?


Regards,
Patrick

New here, but I think I know the reason, someone more skilled than me, plz correct me if I'm wrong.
The PPD difference is due to PCI-E bandwidth an the general "heavier" nature of Windows. As a general rule of thumb, performance on Linux is better than on Windows-as far as F@H is concerned.

I did read about this overhead.
But the 770 isnt close to use up the bandwith of my PCIe2 x4 Slot (max 15% GPU-Z says)
The system itself is at 50-60% CPU and has about 2GB RAM free.

So maybe Windows in general is slower, or there is a bottleneck i dont see, or the WUs are not comparable.

Foliant wrote:I did read about this overhead.
But the 770 isnt close to use up the bandwith of my PCIe2 x4 Slot (max 15% GPU-Z says)
The system itself is at 50-60% CPU and has about 2GB RAM free.

So maybe Windows in general is slower, or there is a bottleneck i dont see, or the WUs are not comparable.

By 15% you mean GPU usage? Or bandwidth speed? From what i can gather online, an 2.0 x8 should be enough to not limit the card.
Right now you are at about PCIe 3.0 x2 levels of bandwidth, so maybe there is performance to be gained from a wider bus.

Your motherboard also plays a role, if they are not identical, you should check the exact number of pcie lanes and their distribution across the board.

There are many reasons why WUs cannot be compared and PCIe bandwidth is only one of them. The FAH Bonus calculation puts a very heavy emphasis on WU duration so even a brief pause is "expensive." The number of atoms can change if 'the box' changes size and if you're trying to compare different GPUs, the number of shaders can be very important as well as the deadlines. Switching to/from CUDA (only on nVidia) from/to OpenCL matters, too.