The new DNET CUDA client is really something. Instead of using your CPU to crack RC5 keys, you use your graphics card’s processing power. You might not realize it, but in some respects your graphics card probably has a more powerful CPU than the one on your motherboard.
It only works with recent NVIDIA graphics chip sets, but man, it is really moovin along on my test computer! I have a couple year old NVIDIA GeForce card in this machine, pumping out some excellent key rates:
Projected ideal time to completion: 0.00:53:20.00
[Jan 31 14:28:08 UTC] RC5-72: 4 packets (256.00 stats units) are in
buff-out.r72
[Jan 31 15:22:32 UTC] RC5-72: Completed CB:F75B7435:00000000 (64.00 stats units)
0.00:54:23.00 – [84,240,854 keys/s]
[Jan 31 15:22:32 UTC] RC5-72: Loaded CB:F75B7476:00000000:64*2^32
[Jan 31 15:22:32 UTC] RC5-72: Summary: 17 packets (982.00 stats units)
0.13:35:24.89 – [80.07 Mkeys/s]
[Jan 31 15:22:32 UTC] RC5-72: 0 packets remain in buff-in.r72
[Jan 31 15:22:32 UTC] RC5-72: 5 packets (320.00 stats units) are in
buff-out.r72
[Jan 31 15:22:32 UTC] The keyserver says: “distributed.net / cracking keys in
idle time / many years to go”
[Jan 31 15:22:32 UTC] Refreshed project state data from server. (cached)
[Jan 31 15:22:32 UTC] RC5-72: Retrieved 5 packets (293.00 stats units) from server.
[Jan 31 15:22:33 UTC] RC5-72: Sent 5 packets (320.00 stats units) to server.
[Jan 31 16:16:56 UTC] RC5-72: Completed CB:F75B7476:00000000 (64.00 stats units)
0.00:54:23.29 – [84,233,187 keys/s]
[Jan 31 16:16:56 UTC] RC5-72: Loaded CB:F870A3E6:00000000:49*2^32
[Jan 31 16:16:56 UTC] RC5-72: Summary: 18 packets (1046.00 stats units)
0.14:29:48.18 – [80.33 Mkeys/s]
Newer NVIDIA graphics cards can run circles around my old card, so if you are considering a new system or upgrading your video for gaming purposes, give the NVIDIA cards a look.
The CUDA client needs a special driver loaded on your windows box to allow it to use the graphics card’s many microprocessors to crunch RC5 keys. You can download it for free from the NVIDIA web site.
Unfortunately, the CUDA client can’t crunch OGR work units, but if you are already concentrating on RC5-72, this could substantially increase your work rate. If you are currently doing OGR-NG exclusively, you can help Team Warped on the RC5-72 challenge and not take away from our OGR stats.
I use both the standard DNET client as well as the CUDA client on Samovar, our friendly neighborhood stats server. The two clients can run simultaneously, and since they use separate hardware, each can run full speed – you get the best of both worlds!
The only downside is that the CUDA client makes your graphics card run a bit hot and it vastly slows down the screen drawing speed. If I need to do actual work on Samovar, I have to pause the CUDA client to make the screen usable. I don’t have to pause the standard client, as this one runs at idle priority and is unnoticeable.
The computer screen looks perfectly normal when running CUDA – it just runs really slow.
I made desktop links to a couple of batch files that pause and unpause the CUDA client. That seems to work well for me. Pause; Do some work; Unpause.
The CUDA client really picks up speed if you overclock the graphics card. My graphics card came with a utility that can shift the GPU clock speed and monitor its temperature. My card heats up another 10-12C when running the CUDA client at its top clock speed. This is not too bad, as I have a lot of cooling fans in Samovar’s computer case.
So, if you already have an NVIDIA card or are thinking of upgrading, give the CUDA client try!
Kevin