There's a bit of a discussion in the top management of my community (just me and some other dude) about disabling hyperthreading on on our dedicated server (sporting a Intel Xeon 3450) on which we host several TF2 / GMod / CSS servers. We're not sure if hyperthreading is providing extra performance, or if it's in the way.
From what I've read, hyper threading gives about a 10% performance boost if you have lots of applications running. But apparently it's better to disable it when you got a single process being a resource hog. In our case the TF2 servers (usually having 20 to 24 people per server these days, yay f2p update) use about 10 to 18% CPU according to task manager, our GMod server is constantly 15 to 25%, although with HT enabled these numbers are a bit less straightforward apparently. The total average CPU usage is around 50-70%:
[img]http://i.imgur.com/nkcb6.png[/img]
There seems to be plenty of debate around hyperthreading on the web but I can't find extensive benchmarks to help me out. Do you guys have some knowledge to share?
(hoping for that gman dude to reply)
HT can help with very parallel complex programs like encoding and rendering because it's not really time dependent. While one thread is halted another one is thrown into the core to start processing.
If you are dealing with latency then HT could add in some delay I think and cause some hindrance for response dependent applications.
For task manager, every 2nd bar is the HT core. So graph 0, 2, 4, 6 are the actual cores I believe, and graphs 1, 3, 5, 7 are the HT logical cores. That's my understanding from my reading. You may notice that graphics 0,2,4,6 sometimes run higher overall.
________________
But I want to hear was gman has to say. My post is based on my experience and my findings.
[QUOTE=Brt5470;31071803]If you are dealing with latency then HT could add in some delay I think and cause some hindrance for response dependent applications.[/QUOTE]
Used to be a thing with Hyperthreading back in the day, but Intel's pretty much killed the added latency nowadays.
Why don't you just do benchmarks like a normal person? If it's worth doing it'll be immediately obvious. The effect hyperthreading has depends on many things but generally it's best to leave it enabled.
From what I can see, having HTT on can't decrease performance. In a single-thread situation, it may not actually increase performance, but I don't see any realistic way it could decrease it. I'm not 100% sure of this, though - feel free to run some benchmarks, find out for yourself.
There is a [i]potential[/i] way it could decrease performance, but that only applies to Windows 2000 (or earlier (or extremely old versions of Linux)), and then only on processors with multiple physical cores. Not exactly relevant anymore.
[QUOTE=Catdaemon;31073225]Why don't you just do benchmarks like a normal person? If it's worth doing it'll be immediately obvious. The effect hyperthreading has depends on many things but generally it's best to leave it enabled.[/QUOTE]
It's a remote dedicated server with gameservers and other services running on it 24/7. I rather just discuss the theory instead of throwing everything out of the window just to run some benchmarks.
[QUOTE=Clavus;31079159]It's a remote dedicated server with gameservers and other services running on it 24/7. I rather just discuss the theory instead of throwing everything out of the window just to run some benchmarks.[/QUOTE] is this the only computer you have running servers?
[editline]13th July 2011[/editline]
also if you're not renting out servers it doesn't hurt to bring it down for an hour or two in the wee hours of the night to do some benching
HyperThreading, in my experience, provides a massive boost to SRCDS performance on Linux. No idea about Windows.
This is with an i7-2600k however
I've run very busy servers with and without HT, they both perform the same. If anything, HT gives you the advantage of being able to allocate each server it's own core (since srcds is single-threaded).
Sorry, you need to Log In to post a reply to this thread.