pyrit CUDA nvidia Tutorial + Nvidia overclock instructions

[kidFromBigD]

So it’s possible to have a box full of 8400 GS 512MB “PCI” cards. I have this old AMD 939 board laying around with 2gigs of ram on it and has like 6 PCI Slots, plus a PCI EXPRESS SLOT. So in theory if I buy these 8400 GS PCI cards and a Cheap PCI Express Card I will have a total of 7 cards.

i found the PCI 8400 GS cards on new egg for around 60 bucks each which isn't too bad for a super computer.

Seems possible.
hxxp://en.wikipedia.org/wiki/CUDA#Supported_GPUs
Is a good reference for the nVidia line of cards supporting CUDA, and the GeForce 8400M GS is indeed listed.

If you follow thru and purchase the cards, kindly share your results to this thread. Like you, I have some very old hardware with only PCI slots on the motherboard. I'd like to know how best to run pyrit on my limited budget.

Thanks in advance.

[intertan]

anybody run this on a gtx 280 or 285? what were your pkms?

[purehate]

here is a list of benchmarks for cards tested
http://code.google.com/p/pyrit/

[liteo]

In the CUDA_INSTALL I think there is a small mistake I think. It's supposed to be:

dpkg --force-overwrite -i /opt/cuda/nv_20090130-1_i386.deb

not

dpkg -force-overwrite -i /opt/cuda/nv_20090130-1_i386.deb

it's missing a "-".

It took me a while to figure it out.

Even with that modification I can't get it to work !
I downloaded the WmWare image just to try that tuto ; it's the first thing I did so I can't have messed with anything ?
To the persons who had this problem, did you fixed it ?
I've got a 8800GT so it's sure the problem's not the vidcard.

[purehate]

You cant do it in vmware because the virtual machine doesn't have access to the hardware.

[Barry]

You cant do it in vmware because the virtual machine doesn't have access to the hardware.

Heh, one more thing to add to the "can't do with vmware" list....

I bet if I stopped X and these 4 big old monitors I could harness the full power

Now you're just rubbing it in....

[ochlocracy]

Had the same FATAL ERROR when trying to install with a 8600M GS.

Changing the line in CUDA_INSTALL solved it.

dpkg --force-overwrite -i /opt/cuda/nv_20090130-1_i386.deb

but overclocking doesn't work with either method mentioned. The Card will not accept any new values. Had the same problem with windows until I found the "perfect driver".

root@bt:~# nvclock -s
Card:           nVidia Geforce 8600M GS
Card number:    1
Memory clock:   399.600 MHz
GPU clock:      540.000 MHz

root@bt:~# nvclock -c 1 -m 450 -n 600 -f
Requested memory clock:         450.000 MHz
blaat: 450 0x8063380 1
Requested core clock:           600.000 MHz

Adjusted low-level clocks on a nVidia Geforce 8600M GS
Memory clock:   399.600 MHz
GPU clock:      540.000 MHz

Available cores: 'Standard CPU', 'Nvidia CUDA'
Testing CPU-only core 'Standard CPU' (2 CPUs)...  273.34 PMKs/s

Testing GPU core 'Nvidia CUDA' (Device 'GeForce 8600M GS')...  654.12 PMKs/s

[digitalfriction]

Just wanted to offer my thanks for this guide, I havent ran it yet, but it looks very easy to follow, looking forward to trying it.

[jpm11]

Big thanks for the post.

[spacemonkey]

Benchmark from my box

Available cores: 'Standard CPU', 'Nvidia CUDA'
Testing CPU-only core 'Standard CPU' (8 CPUs)...  1452.31 PMKs/s
Testing GPU core 'Nvidia CUDA' (Device 'Quadro FX 1700')...  1120.05 PMKs/s

The GPU sucks - it's based on a G84 core (which is basically a watered down G80). The CPU is a Xeon X5450 - 4 cores @3Ghz with 12MB of L2 cache. Surprised to see that it's only a tiny bit better than pureh@te's CPU result...

Testing CPU-only core 'Standard CPU' (4 CPUs)...  1294.58 PMKs/s

pureh@te, what were you CPU specs? It would be interesting to see how well pyrit scales with the number of cores.


[edit] never mind, i need to learns me to read mo better..

pureh@te's CPU:

Q6600 quad core chip running at 3.6 GHz

[/edit]