Help! Staggeringly poor performance with P4 640 CPU

[bm2222]

Hello everybody,

This is tricky and I need some smart advice. I have a PC set up with the following specs:

Pentium 4 640 CPU
Zalman CNPS9500 cooler
Asus P5LD2 2.0 mobo
1 GB Corsair DDR2 667
Asus EN6200LE video
Antec TPII 480W pwr
Seagate WD3000JB sata disk

When I run any performance benchmark *other than* Sandra, e.g. CPUBench2003, I get extremely poor performance (good integer ALU, pathetic FPU, poor overall). This corresponds to the poor real-world performance of the machine (we purchased it to run some single-thread math simulations).

The confusing part is this. I also have a HP nw8240 laptop with the following specs:

Pentium M 755 2.0 GHz CPU
2 GB DDR2 500, dual channel
Stock everything else

When I run the same performance benchmarks, other than Sandra, this laptop performs AS WELL or BETTER than the first system.

Providing further confusion is my old system:

Pentium 4-C 3.2 GHz Northwood cpu
2 GB PC3200 DDR dual-channel
Asus P4P800 mb
Quadro FX3000 graphics
Raptor 74 GB SATA disk

This system, midly overclocked, performs TWICE as fast as either the laptop or the first system on our real-world application.

I can't believe that using all newer technology would produce a system that is twice as slow as my old generation box.

So I appeal to all the tech gurus out there to advise me on what to do.

1. Is the 640 just a bum processor with known "real world" performance issues?

2. If so, what can I buy cheaply that will perform at least as well as the 3.2C Northwood on single-thread code but also fit LGA775?

3. Is there a possibility I am seeing this bottleneck in the first box due to running only in single-channel DDR2? Though I tried taking one of the sticks out of my laptop and that didn't affect results.

4. I tried looking through the bios menus in the first system and disabling any kind of thermal protection which could slow the thing down, but that didn't make any difference. Perhaps I am missing something?

5. One thing I forgot to mention, the new box is running Windows 2000 Professional, while the laptop and old box are running XP Pro. Could this possibly be the cause? Does Win2k not support these latest processors performing at their best?

Many thanks in advance,

Ben

 

[mechBgon]

Disable Hyperthreading in the BIOS, if that option is in there, and see if that fixes it. Win2000 isn't fully SMT-aware.

 

[bm2222]

Hi mechBgon, thanks for your advice. I just tried disabling HT, and although now the CPUMark2003 score is considerably higher, our simulation application is just as slow.

An update:

One other important thing I've just noticed: I enabled the Q-Fan technology on the motherboard, which apparently adjusts the fan speed to match CPU temperature. When I did this, the fan became silent (1500 RPM), and the temp hovered around 105F. Now when I run my simulation which is 4+ minutes, the fan speed never changes, NOR does the temperature. It stays at 1500 RPM and 105F. As a consequence the simulation takes 9 minutes, where before, when the fan was running faster, it took 6 minutes.

I think this must be the culprit. On my Northwood, when the PC was running at 100% utilization, the fan got much louder and the temperature increased. On this machine, it doesn't do that. What temperature-limiting feature must be engaged, and how do I disengage it?

I disabled every temperature control feature in the bios that I could find. Again, could this possibly be a Win2k issue? Or do these CPUs automatically slow down over 100F and that cannot be changed? If so what CPU can I buy that won't exhibit this behavior?

Thanks again,

Ben

 

[cheesehead]

105F is not too bad - 40 degrees celsius is par for the course.

I would see if the SCSI disk is helping - for some applications, a seperate controller for storage can drop CPU use a great deal.