Greater production with a single client - HT P4

Medlock

I noticed as I was setting up both my console clients run as services that one of the clients folded much faster than it did when the other was running. Now I'm starting to see more clearly why people say hyperthreading is so overrated. Though both of them comined when dual folding seems to produce slightly more, 200-400 PPW. Perhaps it would fold even faster if I had disabled hyperthreading in the BIOS?? Thoughts/opinions please. Thanks...

See these screenshots.

-Rick

Thrax

Well, yeah. It's not true SMP.. It's more like a bastardized, cobbled-together-because-the-p4-sucks, quasi-symmetric multi-thread(ish) execution engine.

witenoiz

TheGr81 wrote:

I noticed as I was setting up both my console clients run as services that one of the clients folded much faster than it did when the other was running. Now I'm starting to see more clearly why people say hyperthreading is so overrated. Though both of them comined when dual folding seems to produce slightly more, 200-400 PPW. Perhaps it would fold even faster if I had disabled hyperthreading in the BIOS?? Thoughts/opinions please. Thanks...

See these screenshots.

-Rick

I came to the same conclusion when I got my 1st P4 HT and just ran 1 console for about a week - then I thought what the heck let's go with 2 and decided that it is almost like having 1 3/4 boxes running - SO I built another one. They are both running 2 consoles each, I know I am generating more points with 4 instances running than with 1 on each - even though I guess that would be about the equivilant of 3+ pc's gads that even confuses me - but I know what I mean. Jack

Medlock

witenoiz wrote:

I came to the same conclusion when I got my 1st P4 HT and just ran 1 console for about a week - then I thought what the heck let's go with 2 and decided that it is almost like having 1 3/4 boxes running - SO I built another one. They are both running 2 consoles each, I know I am generating more points with 4 instances running than with 1 on each - even though I guess that would be about the equivilant of 3+ pc's gads that even confuses me - but I know what I mean. Jack

Lol. Seems more like 1 1/2 to me.

mmonnin

Some have said for folding to improve about 25% overall. I guess it depends on the WUs. I would think the higher clocked the more 2 clients would help.

dragonV8

Hyper Threading is the reason we run with intel for folding. We dont play games as such, else the story would be different.

The advantage in our case as i see it is: Less computers means more room to expand.
Know nothing about farms with diskless nodes so to us oldies it seems the simple way out. Did a bit of reading/searching about diskless nodes and got totally lost.
Have enough trouble with these normal puters running through a switch.

Just my thoughts and, yes Jack, i do understand, lol.

csimon

How much trouble is it to disable hyperthreading and then re-enable it?

muddocktor

Yeah, I'd guess on the order of 15-25% more production using 2 instances folding instead of 1 on a HT capable rig. I don't think that you would see any difference between folding with 1 instance with HT enabled or disabled but it would be an interesting experiment for you to perform, TheGr81.

Chris, you should be able to enable or disable HT in bios options on a normal mobo but I'm not too sure on those Dell boxes you have at school though.

Squat

Hey TheGr81, I know this is off topic, but from the screen shots, your wallpaper looks sweet, were did you get it.

And for something a little more on topic, Hyperthreading is only making better use of your processer by allowing more than one thread at a time to addressed by the CPU, you still don't have two CPUs, and therefore it won't be abel to run twice as fast (or two programs at the same spead as it would run one, in your case). At least that is the way I understand it. Hope that made sense.

witenoiz

Squat wrote:

Hey TheGr81, I know this is off topic, but from the screen shots, your wallpaper looks sweet, were did you get it.

And for something a little more on topic, Hyperthreading is only making better use of your processer by allowing more than one thread at a time to addressed by the CPU, you still don't have two CPUs, and therefore it won't be abel to run twice as fast (or two programs at the same spead as it would run one, in your case). At least that is the way I understand it. Hope that made sense.

What he said!! That's what I meant. And I like the wallpaper too. And I want one of Jon & Sally's thingies in the corner too. Jack

Medlock

Wallpapers I get from here. You click wallpapers up in the corner. Then I agree at the right. The username and password is guest/access. Lots of cool stuff to see there. Resoltions range from 800x600 to 1600x1200 for nearly all the wallpapers. Except it's 1280x960 instead of 1280x1024. I think it's an aspect ratio thing. All the available resolutions are 4:3.

Well I've been getting quite a few double gromacs WUs lately and I'm just producing points faster and faster. I love it. I may or may not experiment with turing hyperthreading on and off. I think I'll just do better with it on anyway. Thanks for helping me better understand it.

-Rick

Leonardo

Some have said for folding to improve about 25% overall. I guess it depends on the WUs. I would think the higher clocked the more 2 clients would help.

Indeed! My System No. 1 is simply a Folding monster.

How much trouble is it to disable hyperthreading and then re-enable it?

On/off switch in the BIOS.

I've said it before: HT is overrated. But there are few situations, such as Folding and significant multitasking, where HT really shines. Were I a gamer, I probably wouldn't consider a Pentium 4 over an Athlon. If I were to build a box with the primary consideration being Folding - no question about it - P4 Northwood in HT mode w/2 Folding clients.

Medlock

Leonardo wrote:

I've said it before: HT is overrated. But there are few situations, such as Folding and significant multitasking, where HT really shines. Were I a gamer, I probably wouldn't consider a Pentium 4 over an Athlon. If I were to build a box with the primary consideration being Folding - no question about it - P4 Northwood in HT mode w/2 Folding clients.

I've been thinking the same way since joining Short-Media. I've learned so much since then.

I've finally gotten two double gromacs WUs again. But it's not the same as last time. I think there may be something wrong, because before, when only one client was working on a dgromacs WU while the other was doing gromacs, the dgro folded TWICE AS FAST as it is now! I'm not sure if there's something wrong, or if it's this particular WU. But both of them now are averaging >450 PPW!! Take a look at this screenshot ...again.

Could it have anything to do with both using SSE2 instruction? Normally (with one gro and one dgro WU) one unit uses SSE and the other uses SSE2. Just taking a wild guess... I'm gonna stall one and see how the other does.

entropy

Yeah, it makes sense that if the dgromac utilizes the p4 a lot more, stretching it for all it's worth, then two of them would have slightly longer times than one that stresses the whole cpu and one that only does a bit...

csimon

muddocktor wrote:

Chris, you should be able to enable or disable HT in bios options on a normal mobo but I'm not too sure on those Dell boxes you have at school though.

Oh the school boxes are staying hyperthreaded ...especially the server! The 2.4/400's (20) aren't ht'd anyway but my 2.53/533 is.

Leonardo

Don't worry about anomolous changes in your work unit assignment. Earlier today, my Athlon and office machines were crunching Gromacs, while my System 1 Gromacs Destroyer was assigned a Tinker. Go figure. It really has to do with what is 'hot' at Stanford at the time of your computer turning in a WU and receiving a new one. There are trends though: my Pentiums typically draw more Gromacs than Tinkers, and vice versa for the XP2800+.