Hyperthreading again / Mixing Prime 95 TF / NFSNET / ZetaGrid

[TauCeti]

Well, i really do not want to bore you with more Hyperthreading issues, but i see a good chance to improve the overall throughput in GIMPS and specific other DC-Projects.

The following suggestions apply (at least with the given numbers) only to GIMPS-members who own a Hyperthreading P4 and are willing to do TF in GIMPS and are willing to sacrifice 10-20% off their GIMPS-throughput to other projects.

In return GIMPS will get about 80% from P4-HT owners of the other projects. It is a win/win situation if you consider the projects as a whole and not individual stats. If you participate in both projects, it's a win/win for you :)

Idea:

If you mix Prime 95 SSE2-Trial factoring and the NFSNET-client (www.nfsnet.org) or the ZetaGrid client (www.zetagrid.net), you get a huge overall thoughput improvement.

Data:

a) ZetaGrid & SSE2-TF

TF: Works at 83% compared to single thread
ZetaGrid: Works at 78%

Overall throughput 157%

b) NFSNET & SSE2-TF

TF: Works at 87%(!)
NFSNET: Works at roughly 70%

Overall throughput also 157%

If some P4-Users decide to do TF (maybe up to a higher cutoff point), they could help NFSNET to gain some more power (desperatly needed there). Because NFSNET only has about 100 machines right now, they really could use some help. Hehe.

Zetagrid is running on about 8500 machines. With the number of P4-HT machines increasing, there could be a noticable gain for both projects if users from both projects decide to run both clients.

Tau