![]() |
|
|
#1 |
|
Jun 2005
1011111102 Posts |
I'm sure this has been discussed before, but I don't know where.
I just discovered that my instance of Prime95 is operating at only 50% utilization. Is there a way to change this? It's a hyperthreaded CPU, so I don't know if it's effectively the same as a dual-proc, or what? Thanks, Drew |
|
|
|
|
|
#2 |
|
31·107 Posts |
There's no way that I can find to goose Windows to give you the whole CPU. That's because of Intel's shadow registers (for hyperthreading).
You can run a second copy of Prime95. But ... I just ran a little test consisting of generating 10^8 random dfloats. The test program alone takes 13 seconds. With one copy of Prime95 running, the test program takes 19 seconds. With two copies of Prime95 running, the test program takes 19 seconds. So, it's likely that two copies of Prime95 crank along about 1.5X one copy, even though you have "fully" utilized the CPU. So much for hyperthreading. - rll |
|
|
|
#3 | |
|
Jun 2005
2×191 Posts |
Quote:
Drew |
|
|
|
|
|
|
#4 |
|
Jul 2004
Nowhere
32916 Posts |
HT is not useful in prime95 because the program uses only 1 cpu / logical unit. Its best to test weather your getting more work done running 2 copies of prime95 if you arnt then you should just turn off ht in the bios.
|
|
|
|
|
|
#5 | |
|
Jun 2005
2×191 Posts |
Quote:
Drew |
|
|
|
|
|
|
#6 | ||
|
"Richard B. Woods"
Aug 2002
Wisconsin USA
22·3·641 Posts |
Quote:
After a quick review of titles and threads, I can't find an earlier instance of this topic in "Hardware", so you're probably the first to post it here. It seems like a reasonable choice to me -- I was surprised not to find an earlier "50% utilization" thread here. Quote:
A hyperthreaded CPU is not effectively the same as a dual-processor CPU. An HT CPU has just 1 real CPU, but pretends that it has another, virtual CPU that can execute a second instruction stream. It has a few extra real components (extra registers) beyond what a non-HT CPU has, but does not duplicate most components. The HT CPU can rapidly switch back and forth between two instruction streams, using separate components (e.g., ALU for integer arithmetic, FPU for floating-point) to overlap some instructions of the separate streams in the only real CPU. The HT CPU also fools monitor software into thinking that there are two CPUs, so when the one real CPU is really 100% utilized by a single instruction stream, the monitor is tricked into telling you that your system is only 50% loaded. But an HT CPU can't fool either Mother Nature or Father Mersenne; an L-L test requires real processor cycles; virtual cycles do it no good. :-) |
||
|
|
|
|
|
#7 | |
|
Jun 2005
38210 Posts |
Quote:
Drew |
|
|
|
|
|
|
#8 |
|
Oct 2005
22·32 Posts |
Some time ago I carried out a few experiments on my 3EGHz P4 Prescott running Prime95 with priority 10 on a 10,000,000 digit candidate and discovered the following:
With HT turned ON and no affinity set, per iteration time of 0.077; With HT turned ON and affinity set to 0, per iteration time of 0.074; With HT turned ON and affinity set to 1, per iteration time of 0.075; (These all show an overall CPU usage of <51%; 50% is used by Prime95, <1% is used by the system, anti-virus, firewall, etc.) With HT turned OFF the per iteration time INCREASES to (I've forgotten! I know it nearly doubles. I'll check it again and come back shortly and fill this in. What an idiot!) CPU usage is at 100%, nearly all used by Prime95. Because of these results I have HT turned ON and Prime95 affinity set to 0. I don't understand these results: Why, with HT turned ON, is processor 0 faster than processor 1? Why, with HT turned OFF, does the per iteration time INCREASE? |
|
|
|
|
|
#9 |
|
Oct 2005
22·32 Posts |
(I thought I'd be able to EDIT my previous post. Never mind.)
What a load of rubbish that was. The actual per iteration time with HT turned OFF was 0.079. BUT ... With priority set to 10 it slowed everything else down beyond anything reasonable. For example, it took over two minutes to load a simple performance monitor that usually takes two seconds. In conclusion, my experience dictates that if you have an HT enabled processor you should turn HT ON. |
|
|
|
|
|
#10 | |
|
Jun 2005
1011111102 Posts |
Quote:
I actually had better results when idle with HT off, but it took a much bigger hit when I was using the PC. Drew |
|
|
|
|
|
|
#11 |
|
Oct 2005
22·32 Posts |
From VERY unreliable memory my per iteration time INCREASED when I ran two instances of Prime95, not decreased as you have experienced.
I will experiment with this again after Christmas: I have a 10,000,000 digit prime candidate due to complete LL testing on Christmas day. Incidentally, if one runs two instances of Prime95, do they each require their own User/Computer ID? I'm guessing they do. Last fiddled with by MS63 on 2005-12-07 at 21:37 |
|
|
|
![]() |
| Thread Tools | |
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| Hyperthreading | TheMawn | Hardware | 12 | 2013-08-15 00:03 |
| Hyperthreading | Primeinator | Information & Answers | 13 | 2010-05-20 15:15 |
| Hyperthreading | Jud McCranie | Information & Answers | 11 | 2009-03-05 06:41 |
| Should hyperthreading be used? | Electrolyte | Hardware | 5 | 2006-11-08 01:29 |
| Hyperthreading | dave_0273 | Hardware | 5 | 2003-12-12 13:22 |