Page 1 of 1
Points allocation question
Posted: Wed Apr 08, 2020 10:21 pm
by DieHardEazyE
Hi.
I am running project ID 14190 (21, 65, 10), its ETA is 1.25 days (originally 4 days) and its credit allocation is 5,500
I am also running project ID 11777 (0, 9805, 15), its ETA is 2 hours 23 minutes (originally about 6 hours) and its credit allocation is 9,405
What determines credit allocation if the quicker running processes have bigger allocations of credit than the much slower ones ?
Thanks
Re: Points allocation question
Posted: Wed Apr 08, 2020 10:41 pm
by JimboPalmer
https://stats.foldingathome.org/project?p=14190
https://stats.foldingathome.org/project?p=11777
https://apps.foldingathome.org/psummary
Project 11777 is a 'brute force' project running on your GPU.
Project 14190 is searching for subtle changes on your CPU.
I am just guessing, but I am betting your GPU is more powerful than you CPU and so earns more points.
There are projects that are run on the CPU because the GPU has more power, but lacks subtlety.
[Or one researcher just assigns higher points, just like grades]
Re: Points allocation question
Posted: Wed Apr 08, 2020 11:31 pm
by DieHardEazyE
Thanks Jimbo.
Here are my specs. I have an 8-core AMD cpu with a great fan so no overheating issues. My GPU is a radeon RX 460.
Based on this spec I wouldnt have thought the GPU would far outweigh the CPU such as to greatly adjust points.
Thoughts ?
Build Version is 7.5.1
System:
CPU: AMD FX-8370 Eight-Core Processor
Memory: 32 GB (22 GB free)
O.S: Windows 10
OS Arch: AMD64
GPU0 AMD:5 Baffin XT (Radeon RX 460) - actually, it reports this as being a 460 but its actually an RX 560 4GB GDDR5 OC edition
CUDA: Not Detected: Failed to open dynamic library 'nvcuda.dll': The specified module could not be found
OpenCL device: Platform:o Device:0 Bus:1 Compute:1.2 Driver:3004.8
Win32 Service:False
Re: Points allocation question
Posted: Wed Apr 08, 2020 11:38 pm
by JimboPalmer
https://www.techpowerup.com/gpu-specs/r ... -460.c2849
This GPU has 896 cores running at 1 to 1.2 Ghz
https://www.techpowerup.com/cpu-specs/fx-8370.c1785
This CPU has 8 cores running at 4.3 Ghz
so 1/112 as many cores, running 3.6 to 4.3 times as fast.
Re: Points allocation question
Posted: Wed Apr 08, 2020 11:49 pm
by DieHardEazyE
Thanks Jimbo - now I see the math. Thanks for putting me straight on that
BTW, here is the Radeon STRIX RX 560 OC card that I have...
Graphics Engine: AMD Radeon RX 560
Core Name: Polaris 21 XL
OpenGL: OpenGLĀ®4.5
Video Memory: GDDR5 4GB
Engine Clock: Up to 1285 MHz (OC Mode)
Up to 1275 MHz (Gaming Mode)
Stream Processors: 1024
Memory Clock: 7000 MHz
Memory Interface: 128-bit
Resolution: Digital Max Resolution: 5120 x 2880
so 1024 cores running at 1.2 GHz with a clock speed of 7 Ghz.
I see now why my GPU might be finishing stuff off quicker. It doesnt even phase it - I am folding and playing steam games at the same time with no change in speed.
Thanks again.
Re: Points allocation question
Posted: Wed Apr 08, 2020 11:53 pm
by bruce
A CPU core is not exactly the same as a shader. In fact, the CPU is more versatile because it has a more complex set of instructions ... including SSE and AVX which speed it up by a copule of factors of 2. but when you've got to do the same 3D force calculations and motion calculations for 40,000 atoms (or 400,000 atoms) the ability to do those calculations in parallel makes the GPU's 3D engine really shine.
And some of the newer GPUs aren't as limited as one with (only)
1000 shaders. I've seen them with upwards of 4000 shaders
but the cost does goe up.
Re: Points allocation question
Posted: Thu Apr 09, 2020 3:26 am
by iceman1992
JimboPalmer wrote:There are projects that are run on the CPU because the GPU has more power, but lacks subtlety.
Oh, is that why? How does that work? Any sources where I can read up on this? Sounds interesting.
Re: Points allocation question
Posted: Mon Apr 13, 2020 12:53 am
by JimboPalmer
As bruce mentions, your CPU task can be thought of as a AVX engine doing an operation on 64 data items at a time. Considered this way, it has an entire x86 CPU as an auxiliary to handle special cases.
While your GPU has the ability to do an operation on 1000s pieces of data, it has no general purpose CPU also on the card.
Back when I thought I knew what the code did, GPUs had to consider the water surrounding the protein as a continuous fluid, called implicit solvation. CPUs could solve for water as individual molecules, explicit solvation. I never knew why this was better, but the biochemists did.
This Wikipedia article goes over my head real quickly!
https://en.wikipedia.org/wiki/Implicit_solvation
This one does too, but you can almost hear me saying explicit would like a real CPU.
https://en.wikipedia.org/wiki/Solvent_model