Reply With QuoteI actually put that table together, and admit there are probably some inaccuracies. However, I did do some research, and (based on posts on our forums) found that the client performed better on Athlon CPUs than P4's.
I had thought SoB is one of those few P4 optimized projects (like Gimps, they are using almost the same code or so)? On the new Free-DC homepage I now read it is Athlon optimized.
Anyone able to clearify things?
I actually put that table together, and admit there are probably some inaccuracies. However, I did do some research, and (based on posts on our forums) found that the client performed better on Athlon CPUs than P4's.
Just a thought. I may have done my research while there was a problem with the client, and SSE2 support was ripped out for a while. During this time the Athlons were far outperforming the P4's.
Perhaps not, though. Let's see what other responses you get
Thanks for your anser, alpha.
It's of course possible you read of the SSE2 problems, but that still wouldn't make the P4 slower than the Athlon (just slower than it could be).
Perhaps another reply clarified things:
Are you sure what you read was about the core project of SoB and not about the sieving (or factoring)?When it comes to sieving Athlons rule though.
I only own P4s so I have to ensure they get all the honor they can get - GIMPS and SoB are the only P4 optimized projects I know of so I just have to ensure they are marked as such.
Thanks,
Wirthi
You can add md5crk to the list.GIMPS and SoB are the only P4 optimized projects I know of so I just have to ensure they are marked as such.
Originally posted by wirthi
It's of course possible you read of the SSE2 problems, but that still wouldn't make the P4 slower than the Athlon (just slower than it could be).
According to other people, you are incorrect. I remember just yesterday reading an old post where somebody claimed their XP 1700+ was thrashing a P4 2.4.
Are you sure what you read was about the core project of SoB and not about the sieving (or factoring)?
100% positive. I have contributed to all aspects of the SoB project and am therefore familiar with the concept of the sub-projects.
I only own P4s so I have to ensure they get all the honor they can get - GIMPS and SoB are the only P4 optimized projects I know of so I just have to ensure they are marked as such.
Understandable. I'm not entirely sure I'll own up to being wrong yet, though.
P4's used to be much faster than athlons running SOB.
The client was heavily optimised and the P4's running SSE2 ruled the roost.
Recently a bug was discovered when doing the higher ranges, and the 'Fix' was to disable SSE.
The current client has SSE optimisations disabled.
So its all raw FPU at the moment and the Athlons are able to perform better than the P4's.
This is only temporary until SSE is enabled again.
I think you are mistaken. I don't have time to find quotes/posts at the moment, but I'm sure SSE2 instructions have been renabled again now.Originally posted by PCZ
The current client has SSE optimisations disabled.
So its all raw FPU at the moment and the Athlons are able to perform better than the P4's.
Alpha
I checked the download page and v1.20 has SSE re enabled.
Folks,
The new client 1.20 does indeed have SSE2 instructions reenabled. For the values of N we are working on now(mid 55xxxxx range) a 3Ghz P4 with dual channel memory will run about 750,000cEM/s. A comparably rated Barton will do 525,000cEM.sec with the same client.
The Opterons were faster with the older 1.11 client with SSE2 disabled. A 2Ghz Opteron would reach about 450,000CEM/sec with the older client 1.10 client, and a bit faster with the 1.20 client, however, it was 25% faster with the 1.11 client. It seems that disabling SSE2 was advantgeous for Opteron or Athlon64 processing speed.
Fritz
I'm using the 1.2.5 client now, and I think the athlons are way better.
I have a Pentium4 2.4 Ghz (533fsb) with 512mb of pc2100, and it only does around 260,000cEMs/sec
I also have an AthlonXP 2000+ (266fsb) with 256mb of pc2100, and it does over 300,000cEMs/sec pretty consistently, and that is only like ~1800Mhz
Emporor,
Your results are consistent with benchmarks I have run. For the P4s to be fast on 17, they must have a fast memory bus. A P4 2.4 with the 845 chipset and PC2100 memory is fairly slow. However, take a P4C-2.4 with the 800MHz FSB and PC3200 and you get between 600,000 and 700,000cEM/sec.
Fritz
Posting Permissions
