![]() |
|
|
#1 |
|
Jan 2009
310 Posts |
I noticed that 2 Sandy Bridge CPUs(E3-1230,2600K), both 4 cores w Hyperthreading enabled are only using 50% of CPU.(Taskmgr Windows7 x64 v26.6 Prime95) Both show pairs of logical CPUs(1-2,3-4,5-6,7-8) for each physical CPU. If I run torture test with 8 workers, CPU usage is 100%. I believe CPU usage was 100% before v26.6 upgrade, which reduced 8 workers down to 4. Any ideas?
|
|
|
|
|
|
#2 |
|
Romulan Interpreter
"name field"
Jun 2011
Thailand
41×251 Posts |
i think that is right, you have 4 "physical" cores, but 8 "logical", because of HT. I had same problem in the past with an atom cpu, doing factoring, saw that only 50% is busy and it was no way to convince it to start 4 threads of factoring (2 physical cores, taskmgr reporting 4 because of HT). After some playing around with the settings I convinced prime95 to start 4 workers, cpu became 100% busy, but the speed of each worker decreased to about 55%, so in fact you won't get much gain in the speed, and beside of it, your computer (talking here only about my atom, no idea how your computer will behave) became a bit slower (longer response for other programs). So, I believe the Prime95 correctly detected my 2 physical cores, and the best way should be to let it run only 2 workers, in this case I have the HT ("rest of 50%") free to do my daily stuff. Forcing the prime95 to open 4 workers is a bit of "too much" for 2 physical cores, either if taskman reports 4. Same as running two or more workers on a single core, you won't get much gain, with or without HT, the difference would be quite small.
|
|
|
|
|
|
#3 |
|
Account Deleted
"Tim Sorbera"
Aug 2006
San Antonio, TX USA
11·389 Posts |
LaurV is right: the extra HT logical cores don't significantly help LL testing, and in this case the task manager saying it's 50% is still using your CPU as much as possible. I think the HT detection was improved recently, so the new version decided to go down from 8 to 4 worker threads because you only have 4 physical cores.
Last fiddled with by TimSorbet on 2011-06-12 at 14:46 |
|
|
|
|
|
#4 |
|
Jun 2010
17 Posts |
Yes. I dont know if it is the same as on Nehalem, but enabling the helper threads on that arch caused the per iteration time to be about 5ms longer, dependent on FFT size.
It also burned 15W more power and ran hotter as a result. |
|
|
|
![]() |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| So Sandy Bridge Xeons are now launched | fivemack | Hardware | 6 | 2012-03-14 11:27 |
| Overclocking, Sandy Bridge-E : Don't | firejuggler | Hardware | 6 | 2012-03-08 19:38 |
| 2 disabled cores in new Sandy Bridge-E :( | stars10250 | Hardware | 8 | 2011-11-16 13:55 |
| New Sandy Bridge Computer Help (Built - WOW!) | Prime95 | Hardware | 104 | 2011-05-24 00:32 |
| Sandy Bridge benchmarks are out. | nucleon | Hardware | 0 | 2011-01-04 11:41 |