| Author |
Message |
|---|
 BiBi Volunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
                   
|
|
I am running PFGW on three multi-core systems and am noticing some performance degradation using all cores.
I cannot figure out what is causing the degradation. Is it the memory, is it the stepping modes of the CPU or is there another reason.
For example: The tasks I am running on a C2D T5600 take 3600 seconds when one core is busy but it takes almost 5400 seconds when both cores are busy.
Did someone notice this as well?
Regards BiBi |
|
|
|
|
I am running PFGW on three multi-core systems and am noticing some performance degradation using all cores.
I cannot figure out what is causing the degradation. Is it the memory, is it the stepping modes of the CPU or is there another reason.
For example: The tasks I am running on a C2D T5600 take 3600 seconds when one core is busy but it takes almost 5400 seconds when both cores are busy.
Did someone notice this as well?
Regards BiBi
LLR is cpu cache intensive so this cause the slowdown. I can see this also with Riesel Sieve LLR or SR5.
____________
|
|
|
mfbabb2Volunteer tester
Send message
Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 17,211,226
RAC: 38,181
              
|
|
Also to consider -- are these real cores or hyper-threaded virtual cores?
____________
Murphy (AtP)
 |
|
|
|
|
Also to consider -- are these real cores or hyper-threaded virtual cores?
The C2D has no HT ;) |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1249
ID: 12001
Credit: 18,565,548
RAC: 0
 
|
Did someone notice this as well?
Regards BiBi
LLR is cpu cache intensive so this cause the slowdown. I can see this also with Riesel Sieve LLR or SR5.
gwnum, the library behind LLR and PFGW is memory intensive. I have seen the same behavior. According to George (the author of gwnum) the memory bus is the bottleneck, so even though you might have the same physical CPU (a Core 2 Duo for example), two different models of computer from the same manufacturer can have vastly different performance. |
|
|
|
|
Did someone notice this as well?
Regards BiBi
LLR is cpu cache intensive so this cause the slowdown. I can see this also with Riesel Sieve LLR or SR5.
gwnum, the library behind LLR and PFGW is memory intensive. I have seen the same behavior. According to George (the author of gwnum) the memory bus is the bottleneck, so even though you might have the same physical CPU (a Core 2 Duo for example), two different models of computer from the same manufacturer can have vastly different performance.
Sandy brigde should have be a better performance of the memory controller. Can anybody confirm this? |
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
     
|
Sandy brigde should have be a better performance of the memory controller.
Not only Sandy Bridge, all Intel-CPUs with integrated memory controller.
____________
Best wishes. Knowledge is power. by jjwhalen
 |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Previous timing tests with PFGW '3.4.4.32BIT.20101104.Win_Dev'
New timing tests with PFGW '3.4.7.32BIT.20110404.Win_Dev'
1 core ~3150 seconds
2 core ~5950 seconds
It must be the memory. I'll be using one core for the PFGW tasks. |
|
|
mfbabb2Volunteer tester
Send message
Joined: 10 Oct 08
Posts: 510
ID: 30360
Credit: 17,211,226
RAC: 38,181
|
Previous timing tests with PFGW '3.4.4.32BIT.20101104.Win_Dev'
New timing tests with PFGW '3.4.7.32BIT.20110404.Win_Dev'
1 core ~3150 seconds
2 core ~5950 seconds
It must be the memory. I'll be using one core for the PFGW tasks.
With 1 core, you get 2 WUs done in ~6300 seconds (3150 x 2).
With 2 core, you get 2 WUs done in ~5950 seconds.
2 cores is better. (Or did I misinterpret your data?)
____________
Murphy (AtP)
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
With 1 core, you get 2 WUs done in ~6300 seconds (3150 x 2).
With 2 core, you get 2 WUs done in ~5950 seconds.
2 cores is better. (Or did I misinterpret your data?)
You're right, but with two cores running the laptop get's really hot. Besides that there is still the possibility to sieve. The other core is doing that right now.
I should have invested that extra money for some faster ram in the 1055T based system that I use :( (It has PC1333 instead of PC1600)
Does someone know how to check the stepping of the CPU in Windows? (In linux I use cpufreq-info)
|
|
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
|
Does someone know how to check the stepping of the CPU in Windows? (In linux I use cpufreq-info)
http://www.cpuid.com/softwares/cpu-z.html
____________
Best wishes. Knowledge is power. by jjwhalen
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
Does someone know how to check the stepping of the CPU in Windows? (In linux I use cpufreq-info)
http://www.cpuid.com/softwares/cpu-z.html
Thanks for reminding me to get a recent version. It immediately showed that the processor was stepping down (probably because of temperature)
I cannot say the slowdown is caused by memory on this C2D. I am not sure about the 1055T system. But I'll wait checking this until next winter :D |
|
|
