[NASsoccer]

you can find this thread by clicking here if you wish. saw this posted over at [H], it was posted by JCH, thought some of you would like to give it a look...

Originally posted by JCH
I'm giving this it's own thread, since it will be easier to find if other people are interested, plus it will give them a place to post their own results.

I promised in AtomicMoose's thread about "Intel upping the ante" that I'd post the results of folding with Hyperthreading turned on and off on my new server. Ok, here ya go.

Test platform:
- Dell Poweredge 1600SC server, BIOS ver. A00
- Single Intel Xeon 2.4 GHz CPU w/512K cache, 400 MHz FSB,
- 512 MB DDR SDRAM,
- Dell PERC3/SC SCSI RAID controller, three 18 GB 10K rpm SCSI drives, RAID 5
- Windows 2000 Server, SP3, all critical updates applied
- member server, not domain controller
- basic Win2K services running, plus Symantec Corporate Antivirus 8.0
- server isn't doing anything else, currently on my lab bench for *cough* burn-in before I move it into production.

I installed FAH3Console 3.14 in two separate folders. I ran each instance solo, and set up the config with different machine IDs. One instance got a p184 protein, the other got a p185, both 9-pointers.

I disabled Hyperthreading in the BIOS and ran the P185 instance for a while, saved the log, then ran the P184 instance and saved the log. I shut the system down, enabled Hyperthreading, and started both instances simultaneously. After it had run for a while, looked at the logs again and I manually calculated the frame times from the logs. Here are the results (times are minutes:seconds):

WITHOUT HYPERTHREADING (so only running one instance at a time):
p184 - 11:53 per frame average
p185 - 11:53 per frame average

WITH HYPERTHREADING (running two instances at the same time):
p184 - 15:38 per frame average
p185 - 15:38 per frame average

These proteins run on TINKER cores. No way to test GROMACS at the moment, since they aren't handing them out right now.

The result for these 9.0-point proteins is 10.91 points per day with Hyperthreading OFF and 16.58 points per day with Hyperthreading ON, so about a 52% increase in folding output per day.

This means that, in this specific case, enabling Hyperthreading is like adding another half of a CPU to your system.

Remember, this is on a 400 MHz FSB P4-based Xeon running in a server, not a regular P4, so you may or may not see similar behavior in a desktop system.

FOLD ON
NAS

[gtsimmo]

Where do i get me one of those xeon thingies????? drooollls Bet they cost alot too.

Looks like hyperthreading makes a sizable difference there. Something for all intel fans to look forward to

[Daemonfly]

I think Newegg has Xeons available, but yes, they are a lot more expensive.

Basically "server-class" Intel processors.

[Loyalty]

Which Xeons have HT?

[Yodums]

FYI, the new P4 3.06 has HT Just wait a month or two for the things to drop in price and that would be a good folder.

[Loud]

I think the MoBo has to support HT as well as the CPU.

This could get 'spensive.

[overdoze]

Hehe .. I'm so glad to see this as my XP1600 OC to XP2200 is producing 18 points per day on tinker core and about 40 points a day folding gromacs core. Hyperthread or not still can not beat my $55 cpu

I'm running 2 instance folding on the box.

[SickBoy]

I did an experiment a while back with a Hyperthreaded dual Xeon 1.8 at work....I was thoroughly unimpressed. My TBird 1 GHz machine and my dual P3 Xeon 550 -almost- outfolded it over a period of about 36 hours. My test was distinctly less scientific though.

I'm pretty sure too that my dual Duron rig can crank out more WU's than that P4 Xeon rig if I tried it again.

http://forum.oc-forums.com/vb/showth...ht=hyperthread

[Wedo]

What we need is a test running an AMD MP dually folding two, then four, instances with different machine ID's. I'd like to see what an HT MP can do.

Great find NAS.

Wedo

[Wommalong]

Originally posted by Wedo
What we need is a test running an AMD MP dually folding two, then four, instances with different machine ID's. I'd like to see what an HT MP can do.

Great find NAS.

Wedo

That would be something to see... i wonder how it would pan out?

p.s. WTG with the chip (1.4GHz Athlon) you so kindly donated, it is good to see things like that are happening round here

[Yodums]

Originally posted by overdoze
Hehe .. I'm so glad to see this as my XP1600 OC to XP2200 is producing 18 points per day on tinker core and about 40 points a day folding gromacs core. Hyperthread or not still can not beat my $55 cpu

I'm running 2 instance folding on the box.

Elaborate to this folding newbie about this and how he can use the rig in the sig more efficently. Also, when you say 2 instances, do you mean running two folding clients (Console I'm assuming or EM3, I don't know), off of two different directories or?

Yodums

[overdoze]

I never thought you are folding newbie
Yes I would run 2 folding clients in a single cpu box and 3 folding clients on a dual cpu box. Very often one WU might get stuck and not using the cpu due to various reason such as can not download WU, can not send WU or worse yet getting a bad WU which does not use cpu time. While one WU is busy sending or getting new WU the other could fold 100% of the cpu. Hence you would never waste any cpu cycles.

Each client must has its own directory and also setting its own Machine ID ie.. 1,2 or 3, Usage is set at 100% on each client. I use EIII to manage this more efficiently which called the text FAH3Console.exe as well.

PS: the above points I get from my XP1600 at 1.8GHz were measure using P184, P185 and the discontinued Gromacs Pala.

[RCtruckguy]

umm i read 11:53 without HT and 15:38 and With HT and the last part of the post said HT is faster??? did every one skip the last part of the post? i'm jsut confused ~RCTG

[plizzo]

Originally posted by RCtruckguy
umm i read 11:53 without HT and 15:38 and With HT and the last part of the post said HT is faster??? did every one skip the last part of the post? i'm jsut confused ~RCTG

if im not mistaken the times with ht on are the times per instance of folding and with ht on he had 2 instances running. so in essence he folded 2 proteins in 15:38 with ht on and one protein in 11:53 with it off. is this correct?

[belorsch]

Originally posted by plizzo


if im not mistaken the times with ht on are the times per instance of folding and with ht on he had 2 instances running. so in essence he folded 2 proteins in 15:38 with ht on and one protein in 11:53 with it off. is this correct?

Thats what it looked like to me.

[RCtruckguy]

then why are you guys celebrateing? that means HT is slower when folding....just seems an expesive feature to me if noting takes advantage of it (i know nothing always takes advantage of new features at first but if stuff dosen't eventually support it its worthless.), also once its out will it make every program faster or just certian ones. i can't wait for its relese because other CPUs prices will drop ~RCTG

[IceMan3928]

no, that means when HT was on he was running 2 folding clients at the same time, with HT off he was only running one. So even though the time went up a bit, performance is increased by a good deal.

[Deus]

just think of it as 7:49 with HT (15:38/2 if my math is right) and 11:53 without

[RCtruckguy]

oh i never caugt that i though he was running only one instance on each test. but i guess HT makes a big diffrence, now the quetion is how well will it do with other programs.~RCTG