GPU > CPU? (9800gt superclock > i7?)

Methal · Apr 15, 2009

I just installed my first GPU folding er...thinger. I noticed that the first package it DLs to fold is one with 20,000 things (pardon lack of terminology.) the one I have on my CPU (i7 920 @ 3.7ghz) is doing a 125 packet and is at 112 of. from the time I took the GPU to go from 0 to 600 my cpu went from 100 to 112.

what gives? Am I missing something?

SniperXX · Apr 15, 2009

I think you are refering to frames? The cpu and gpu work units are very different, so they will have different amounts of frames. With your i7 plus notfreds vm you should get more points on the i7 I think. Most other cpus will be outproduced by the gpu tho.

StrategyFreak · Apr 15, 2009

if you haven't already, install fahmon for easier monitoring of ppd.

Methal · Apr 15, 2009

StrategyFreak said:
if you haven't already, install fahmon for easier monitoring of ppd.

Love it =D

Methal · Apr 15, 2009

SniperXX said:
I think you are refering to frames? The cpu and gpu work units are very different, so they will have different amounts of frames. With your i7 plus notfreds vm you should get more points on the i7 I think. Most other cpus will be outproduced by the gpu tho.

by that definition yes. I guess you are right. My i7 is running 8. while my GPU is running 1.

what is a frame if you don't mind me asking?

turbohans · Apr 15, 2009

LOL If you havent folded with you Gpu yet before your in for a treat!! I do know that If you really want to get the most out of your I7 you will have to read the sticky on installing the milticore cpu client,, but you will be happy with gpu!!

Methal · Apr 15, 2009

turbohans said:
LOL If you havent folded with you Gpu yet before your in for a treat!! I do know that If you really want to get the most out of your I7 you will have to read the sticky on installing the milticore cpu client,, but you will be happy with gpu!!

why such a huge difference between them? i mean isn't the i7 supposed to be top dog? how is a measly 9800gt running circles around it? (or is it?)

Outback_Jon · Apr 16, 2009

Basically it's because the Folding program is a simulation of a 3D model. And which part of your computer is optimized for 3D work? Yep, the GPU. That's essentially all it's designed to do.

If you're running an Nvidia GPU, some of the CUDA demonstration downloads from Nvidia give a really good idea of the difference. The Dark Basic PhysX Soft Body Demo available here lets you change from CPU to GPU to see the difference.

StrategyFreak · Apr 16, 2009

Methal said:
Love it =D

then it should make it a lot easier to see the rate of production of the gpu vs. the cpu in terms of ppd, so you don't have to describe things in terms of the number of steps completed

scientifically, the gpu does computations a lot faster but does not have the same flexibility as the cpu.

Methal · Apr 16, 2009

what if anything is the advantage of running CUDA?

Roisen · Apr 16, 2009

If you are running a GPU client, then you are running CUDA.

CUDA is an extension of the C programming language that allows the CPU to tell your nVidia 8xxx series and above GPU to do work. With nVidia graphics cards, this will typically only take 2% of your CPU time, meaning that in addition to running a GPU client with CUDA, you can also run a CPU client as well.

Edward2 · Apr 16, 2009

ABSTRACT. We describe a complete implementation of all-atom protein molecular dynamics running entirely on a graphics processing unit (GPU), including all standard force field terms, integration, constraints, and implicit solvent. We discuss the design of our algorithms and important optimizations needed to fully take advantage of a GPU. We evaluate its performance, and show that it can be more than 700 times faster than a conventional implementation running on a single CPU core.

The above quote is from Vijay Pande on his blog, found HERE.

Surferseth · Apr 16, 2009

Something else to note, the SMP client can only use 4 cores (please correct me if I am wrong). The SMP client on a stock core i7 920 produces about 2kPPD (going from what I see on my i7). If you run VMWare server and drop in a couple NotFreds VMs you will maximize your CPU usage and see the CPU will outproduce the GPU.

All things equal the GPU outperforms the CPU, but if you use all 8 cores (HT on) of the i7 the CPU will outproduce the single core of the GPU.

ChasR · Apr 16, 2009

@ Methal

A frame, as it is refered to in this forum, is 1% of the WU's steps. 1 frame from the file FahLog.txt:

[21:13:03] Completed 270000 out of 500000 steps (54 percent)
[21:33:51] Writing local files
[21:33:51] Completed 275000 out of 500000 steps (55 percent)

WUs have different numbers of steps, depending on how the researcher set up the simulation, but all current WUs have 100 "frames".

GPU > CPU? (9800gt superclock > i7?)

Methal

Member

SniperXX

Member

StrategyFreak

Registered

Methal

Member

Methal

Member

turbohans

Member

Methal

Member

Outback_Jon

Member

StrategyFreak

Registered

Methal

Member

Roisen

Member

Edward2

Member

Surferseth

Member

ChasR

Senior Member

Similar threads