April 2012 News from the DAB

[ChasR]

THere has been a lot of back and forth on BA ppd. I think PG has concluded that it is so high as to be detrimental to the project, hence the release of p8101 from beta. THe discussion has led to a new topic, the benchmark machine. I proposed that it be a 4P M-C or IL machine and that all WUs be benched on it. Uniprocessor on 1 core, regular smp on 1 processor and BA on 2 processors. THat seems to be taking hold among the DAB members, though PG has yet to comment.

 

[harlam357]

That would definitely help Intel cpus on Uni and SMP work since even a faster 8-core G34 chip won't be able to hold a candle a 2nd or 3rd gen Core cpu. I say an 8-core G34 because that would be more apples-to-apples in terms of OS core count vs. what is probably the most abundant Intel cpu, a quad w/HT.

 

[ChasR]

You would need 12 MC cores to approximate an i7. probably 16 IL cores. Normalization would even things out so that the 2600Ks make about the same ppd as they do now on regular SMP work. Other rigs would fall where they may.

I am still considering the Intel MPs as an alternate, THey are comparatively expensive and I'm unsure of 4P availability. They do have the advantage of better per core performance and a more direct relationship to the more popular Intel rigs, Using an Intel rig would more likely avoid even greater BA ppd when Intel introduces their next round of server chips.