SETI@home
Wednesday 2nd August, 2006 19:48 Comments: 1
I was looking at my position in my SETI@home team, which is still exeter university as I\'ve been a part of it since I started back around the beginning of 2001, and noticed that I\'m in second place. If you look at the number of results returned recently, I\'ve returned 65, 25 and 16 (a total of 106). Nick Johnson has returned 72. That's only 68% the number of results I\'ve returned. In theory, I should expect to see about 2/3rds the amount of recent average credit that I have. My recent average credit is 535.79, Nick's recent average is 615.65. Yes, that's more than mine. So can anyone explain why someone is returning fewer units and getting a lot higher average credit? I decided to dig a little deeper in search of answers.
My Three PCs
silentbob2 Intel Xeon 2.40GHz - 4 CPUs, floating point speed 1034.91 million ops/sec, integer speed 804.58 million ops/sec
silentbobx Intel Pentium D 2.66GHz - 2 CPUs, floating point speed 1748.72 million ops/sec, integer speed 1495.74 million ops/sec*
silentbob3 AMD Athlon XP 2800+ - 1 CPU, floating point speed 1968.45 million ops/sec integer speed 3305.22 million ops/sec
Nick Johnson's PC
Intel Xeon CPU 3.20GHz - 4 CPUs, floating point speed 667.82 million ops/sec, integer speed 1228.54 million ops/sec
In terms of raw clockspeed, if we assume that the Athlon rating is fairly accurate compared to the P4 (and it typically has been over the years) we get:
4 x 2.4GHz
2 x 2.66GHz
1 x 2.8GHz
= 17.72GHz
4 x 3.2GHz
= 12.8GHz
So why aren\'t I ahead? Well the most obvious reason might be because my main machine (2.66GHz) isn\'t on all the time. In fact it's not on very often at all. However, I do have it overclocked, so instead of being the 2.66GHz it claims to be, it's actually running at a very impressive 3.6GHz. That makes it difficult to compare raw clockspeed.
It\'d probably be more accurate to compare floating point operations, as I believe that's pretty much what you need for crunching SETI workunits, and those figures should be there based on the CPU benchmarks or the work returned or something useful, consistent and accurate. I don\'t think integers really affect SETI, but they normally go hand in hand. Note: the Xeon processors that Nick and I have use HyperThreading so it's really two CPUs each with with two integer units and a shared floating point unit, rather than being two dual core CPUs (the Pentium D I have is a dual core CPU). Assuming my main machine is never on (even though it is) and the ops/sec are overall and not per CPU, that's 3003 and 4110 versus 668 and 1229. What happens if we just compare his one machine against just one of mine (the fileserver)?
silentbob2:
Measured floating point speed 1034.91 million ops/sec
Measured integer speed 804.58 million ops/sec
Average turnaround time 3.27 days
Results 65
Nick's PC:
Measured floating point speed 667.82 million ops/sec
Measured integer speed 1228.54 million ops/sec
Average turnaround time 0.57 days
Results 71
So the number of floating point operations is about 2/3rds of mine (my CPU is 3/4 the speed of his), his integer speed is 50% faster (more than I expected). Comparing the two hosts, he's only done slighty more results (1.09x the number of mine). The big differences? His turnaround time is a lot faster - perhaps because I queue a lot of workunits up as I don\'t trust my ISP, router or the SETI@home servers; he runs Linux. I\'m not switching to Linux just to get more credit, but I could see if fiddling with the queue makes any difference (even though it shouldn\'t).
Other possible differences? I don\'t think he's overclocking his machine, otherwise I\'d expect to see faster ops/sec. Perhaps he uses an optimized client? I thought I\'d switched to an optimized client a few days ago (P4-SSE2 for the Xeons, P4-SSE3 for my Pentium D) but I hadn\'t restarted the services for it to take effect (I hibernate my main machine). That means I could crunch even more workunits and get more credit, but probably still not get as much credit. Time will tell, now that I\'ve just started crunching with the optimized clients.
I wouldn\'t accuse Nick Johnson of cheating, especially as SETI are meant to be good at spotting that kind of thing, but something must be up. I know they plan/planned on switching to credit based on the number of floating point operations, but based on the evidence above I have 50% more floating point operations per second so that should be in my favour. So WTF is going on? All I want are answers.
My Three PCs
silentbob2 Intel Xeon 2.40GHz - 4 CPUs, floating point speed 1034.91 million ops/sec, integer speed 804.58 million ops/sec
silentbobx Intel Pentium D 2.66GHz - 2 CPUs, floating point speed 1748.72 million ops/sec, integer speed 1495.74 million ops/sec*
silentbob3 AMD Athlon XP 2800+ - 1 CPU, floating point speed 1968.45 million ops/sec integer speed 3305.22 million ops/sec
Nick Johnson's PC
Intel Xeon CPU 3.20GHz - 4 CPUs, floating point speed 667.82 million ops/sec, integer speed 1228.54 million ops/sec
In terms of raw clockspeed, if we assume that the Athlon rating is fairly accurate compared to the P4 (and it typically has been over the years) we get:
4 x 2.4GHz
2 x 2.66GHz
1 x 2.8GHz
= 17.72GHz
4 x 3.2GHz
= 12.8GHz
So why aren\'t I ahead? Well the most obvious reason might be because my main machine (2.66GHz) isn\'t on all the time. In fact it's not on very often at all. However, I do have it overclocked, so instead of being the 2.66GHz it claims to be, it's actually running at a very impressive 3.6GHz. That makes it difficult to compare raw clockspeed.
It\'d probably be more accurate to compare floating point operations, as I believe that's pretty much what you need for crunching SETI workunits, and those figures should be there based on the CPU benchmarks or the work returned or something useful, consistent and accurate. I don\'t think integers really affect SETI, but they normally go hand in hand. Note: the Xeon processors that Nick and I have use HyperThreading so it's really two CPUs each with with two integer units and a shared floating point unit, rather than being two dual core CPUs (the Pentium D I have is a dual core CPU). Assuming my main machine is never on (even though it is) and the ops/sec are overall and not per CPU, that's 3003 and 4110 versus 668 and 1229. What happens if we just compare his one machine against just one of mine (the fileserver)?
silentbob2:
Measured floating point speed 1034.91 million ops/sec
Measured integer speed 804.58 million ops/sec
Average turnaround time 3.27 days
Results 65
Nick's PC:
Measured floating point speed 667.82 million ops/sec
Measured integer speed 1228.54 million ops/sec
Average turnaround time 0.57 days
Results 71
So the number of floating point operations is about 2/3rds of mine (my CPU is 3/4 the speed of his), his integer speed is 50% faster (more than I expected). Comparing the two hosts, he's only done slighty more results (1.09x the number of mine). The big differences? His turnaround time is a lot faster - perhaps because I queue a lot of workunits up as I don\'t trust my ISP, router or the SETI@home servers; he runs Linux. I\'m not switching to Linux just to get more credit, but I could see if fiddling with the queue makes any difference (even though it shouldn\'t).
Other possible differences? I don\'t think he's overclocking his machine, otherwise I\'d expect to see faster ops/sec. Perhaps he uses an optimized client? I thought I\'d switched to an optimized client a few days ago (P4-SSE2 for the Xeons, P4-SSE3 for my Pentium D) but I hadn\'t restarted the services for it to take effect (I hibernate my main machine). That means I could crunch even more workunits and get more credit, but probably still not get as much credit. Time will tell, now that I\'ve just started crunching with the optimized clients.
I wouldn\'t accuse Nick Johnson of cheating, especially as SETI are meant to be good at spotting that kind of thing, but something must be up. I know they plan/planned on switching to credit based on the number of floating point operations, but based on the evidence above I have 50% more floating point operations per second so that should be in my favour. So WTF is going on? All I want are answers.
Robert - Thursday 24th August, 2006 23:13 It all seems to be better now, as I'm getting about double the amount of average credit.