Slow iteration times with 23.7

[smoffat]

I am using V23.7 on a P4,3G. I am consistently getting .041 iteration times which is slower than on my P4,2.5G machine @0.40 per iteration and 100% utilization. Task manager says that the CPU is only at 50% utilization on the newer machine. I know that George says that Hyper Threading may be a partial cause for the CPU utilitzation, but whats up with the iteration times. I am running XP Home with the latest patches on both.

I am pretty sure that I started out with V22.? when I first got the machine and iteration times were much lower.

[QuintLeo]

Try disabling HyperThreading in your BIOS (by report, some have an option to do so).

There are a LOT of applications that run SLOWER when HyperTreading is enables than when it is disabled - Intel's PR to the contrary, HyperThreading does NOT help on a majority of applications, and RARELY helps if the application is coded for maximal efficiency on a P4 to begin with.

[Xyzzy]

That said, I run HT on my P4 and as long as I run just 1 instance of Prime95 I don't see a difference...

What I do like is how the "second" CPU picks up the slack for when I do other stuff so the "first" CPU can run uninterrupted...

[smh]

Does XP home support dual cpu's?

[Xyzzy]

Originally posted by smh
Does XP home support dual cpu's?

I don't know about dual CPUs, but XP Home supports HT...

http://www.intel.com/support/platform/ht/os.htm

[smoffat]

I tried disabling HT and found that the iteration time went up slightly. Also tried an older version of Prime95 with same results. Still baffled as to why my P4 2.5Ghz machine has lower iteration times than this P4 3.0Ghz machine, even when I make sure that nothing else is running on both machines????

[Rastus]

Does the P4 have onboard video? That could be your problem.

[lycorn]

Are you sure you are comparing exponents in the same FFT range? If the one you are testing on the P4 3G belongs to a higher range that may explain why iteration times are higher then on the 2.5.

[Jim]

Ive noticed a similar thing. Just got a new P4 3Ghz with XP Prof. Installed 23.7.1 which seems to run ok, but when i look at the list of processes, Prime95 is taking 50% and the system sits idle for the other 50%.

I havnt installed any other programs yet. I installed and am running as an administrator.

Iteration times are about 0.051

But cant see why the processor sits idle for half the time.

Jim

[ET_]

Ive noticed a similar thing. Just got a new P4 3Ghz with XP Prof. Installed 23.7.1 which seems to run ok, but when i look at the list of processes, Prime95 is taking 50% and the system sits idle for the other 50%.

Maybe HyperThreading is on?

Luigi

[nucleon]

I've been thinking on how would be a good way to explain hyperthreading. I think I have it.

My suggestion is to think of a traditional (non-HT) CPU as having three components:
1) 1x Owner
2) 1x Manager
3) A number of workers

The owner is the Operating System, the manager is the scheduler or the microcode of the CPU, and the workers are the instruction units - FPU/INT etc..

The owner gives the workload to manager to get done. The manager reports to the owner how busy the workers are. The manger coordinates all the tasks to the different workers.

Now in a normal CPU if the manager and their workers are 100% busy then 100% is reported to the owner. No issues there.

In a HT CPU, there are two managers, BUT the same number of workers.

In the HT CPU, the same task is given to one manager, the manager needs most of the workers, the manager reports to the owner that they are 100% busy. The other manager reports to the OS that his/her team is 0% (as manager no2's virtual 'team' hasn't been allocated any work to do). The OS reports (100% + 0%) / 2 = 50%. So the OS reports 50% even though there aren't any workers left, so the CPU is technically at close to capacity.

What about two LL tasks on a HT CPU? One LL is given to one manager, he grabs the FP workers and the LL process starts. But when the other manager is given an LL task to complete, all of the FP workers are already in use. So there is contention over the workers.

So now as the FP workers are on two tasks and have to report to two managers, there is more "adminstration" overhead. So two tasks require more than twice the time.

So more highly optimised code for a single task is better on non-HT environment. But a HT environment is better if the two tasks don't have common instruction requirements. (Say general OS requirements, internet explorer and 1x LL test)

Also probably why some of the lesser optimised tasks can run in two instanaces on HT cpu and get twice the work done in the same time. (As each tasks isn't using all the workers)

Just a thought, let me know if it was useful...

-- Craig