Should I turn off HT in my farm?

agrikk

Gawd
Joined
Apr 16, 2002
Messages
933
I have a stack of 2.8GHz P4 HT rigs that I'm putting into a farm. I plan on running XP on them with minimal services running in the background except for FAH4 that I'll run as a service using FireDaemon.

Should I turn off hyperthreading and have only one session of FAH4 running? What are the performance implications of this?
 
I believe if you leave HT on and run 2 instances of folding you will see about a 15% gain in productivity. If I am wrong on the numbers someone will correct me. :D

T-3
u_T-3.gif
 
T-3 said:
I believe if you leave HT on and run 2 instances of folding you will see about a 15% gain in productivity.
That is my understanding. I could be wrong too though. :)
 
T-3 said:
I believe if you leave HT on and run 2 instances of folding you will see about a 15% gain in productivity. If I am wrong on the numbers someone will correct me. :D

T-3
u_T-3.gif
That sounds about right
 
I have 2 Identical P4 Systems and ran 1 on 2 instances HT and one not, didn't do the % math but I definetly saw more production on HT enough to put both on 2 instances. :)

JonasSteel
750.gif

876 Points and Climing
 
if u run one instance on a HT machine of 2.8GHz then it should work at about 3.1-to-3.2GHz...
if u run two instances then it should fold like two 1.6GHz systems...i think...:)
it is definatly a boost on HT machines.
 
arunabh said:
if u run one instance on a HT machine of 2.8GHz then it should work at about 3.1-to-3.2GHz...

umm, no.

If you run one instance of FAH on a 2.8Ghz HT Machine you will get 2.8Ghz of performance. It takes two instances to take advantage of HT. HT allows multiple threads to be worked on at one time, hence the advantage. Running a single thread (instance of FAH) is only going to run on one virtual CPU and give you no advantage.

I posted a question on here a few months ago regarding Folding on HT processors. The question I raised was if there was an overall performance impact to the system when running FAH. It’s always been the status quo that running a distributed computing client on your computer doesn’t significantly impact performance because it is set at the lowest priority, thus giving up CPU power to any other thread that requests it. This breaks down in an HT environment because thread priorities only apply to threads on an individual CPU.

In a situation where you were running two instances of FAH, and then proceeded to launch a CPU intensive app, the FAH client on the same virtual CPU as the application would certainly give up it’s CPU resources, however the FAH client on the 2nd virtual CPU would continue to run at full power (because even though it’s low priority, nothing is fighting for the CPU power on that particular virtual CPU). The end result would be that the CPU intensive application you launched would only be able to take advantage of about half of your CPU power.

The ultimate question here is whether an OS being “HT aware” also means that it is capable of handling thread priorities across multiple Virtual CPU’s. If it is not, then you end up in a situation where a high priority app is forced to share the CPU with a low priority app because they are each running off a different virtual CPU (though very much the same REAL cpu).

Unfortunately, no one was really able to come up with a definitive answer in that particular thread. I’d sure like to know though, because when I borg someone’s computer, and look them in the eye and tell them it won’t really reduce their systems performance, I’d hate to be lieing.
 
What I've done on clients PCs is to enable HT on their rig and only run one instance of FAH, so at least they have half a processor free at any given time. If the user then complains about performance, I shit down the client for good.

Frankly, I've found that running a distributed client on a non-HT rig causes it to run sluggish, even when the client is set to the lowest priority.

I guess no one really has a good answer about two instances on a dedicated FAH HT rig... :(
 
agrikk said:
What I've done on clients PCs is to enable HT on their rig and only run one instance of FAH, so at least they have half a processor free at any given time. If the user then complains about performance, I shit down the client for good.

Frankly, I've found that running a distributed client on a non-HT rig causes it to run sluggish, even when the client is set to the lowest priority.

I guess no one really has a good answer about two instances on a dedicated FAH HT rig... :(

Well, the problem doesn’t really go away when you only run one instance. It will still keep one virtual CPU fully utilized, and the other program you open will just run on the available virtual CPU (only being able to use 50% of the real CPU because the other virtual CPU is in use by the distributed client).
 
agrikk said:
If the user then complains about performance, I shit down the client for good.
Back when we had the SSE bug on AMD's, the occasional lockups made me want to rip off FAH's head and shit down it's neck. Now I only shit it down when I'm playing games. I never shit it down at work and I've been running RC5, FAH, GAH and UD for many years with no noticeable performance degredation. :p :)
 
Mattman said:
Back when we had the SSE bug on AMD's, the occasional lockups made me want to rip off FAH's head and shit down it's neck. Now I only shit it down when I'm playing games. I never shit it down at work and I've been running RC5, FAH, GAH and UD for many years with no noticeable performance degredation. :p :)

Yeah sometimes FAH decides it wants all of my preciousssss CPU cycles and I can hardly move the mouse.
 
There has to be something else going on with your computer. Have you changed the priority of the task from it's default? The computer I'm using right now just turned in its 1648th WU and outside of the SSE lockup bug, it hasn't complained once about running FAH. Give your PC a good checkup and see if you find anything.
 
Back
Top