[lukers9]

It's my first post on the forums, so hello everyone

I have a dedicated server with Intel Core i3-4130 @ 3.40GHz which has 2 physical cores and 4 threads when Hyper-Threading is enabled. I want to run 2 128TR CS:GO servers and get a maximum performance out of this machine. Will disabling Hyper-Threading help with boosting the performace or will it decrease?

Also I don't have access to the bios and I've read that Hyper-Threading can be disabled in the system. Will it perform just the same as disabling it with bios?

Is it better to assign each server to a certain core or just let them run normally?

Do you know any more tweaks to get more performance with SRCDS like kernel optimization or something similar? I'm currently running Debian 8 Jessie.

[Neuro Toxin]

If windows.

Dont disable HTT and dont assign affinities.

[Puppetmaster]

Yeah I would agree for the most part. Dont mess with HT and let the scheduler assign the threads. You probably wont see any improvement by messing with it.

[Neuro Toxin]

I receive significant performance decreases for assigning affinity or disabling htt.

This is measured in netgraph

[LeetDonkey]

I've been messing around with in in CentOS.
I came to the conclusion that I get better performance from using taskset to assign cores when running multiple srcds instances than letting the scheduler handle it itself.
The scheduler is overly eager in attempting to move threads to a single core, and a bit slow at assigning a new core when the first core gets overloaded.
It's the energy efficient way to do it, but latency sensitive applications might get degraded performance from this.
This leads to short VAR spikes while it's shuffling the threads around, by assigning 1 core manually to each instance I do not experience this issue anymore.
I did notice however that it depends a great deal on number of players and tickrate.

I have HTT enabled and assign two cores on my 4770 to each of my 4 SRCDS instances (I assume 1 core+1 HT scheduler)

[Puppetmaster]

Originally Posted by LeetDonkey I've been messing around with in in CentOS. I came to the conclusion that I get better performance from using taskset to assign cores when running multiple srcds instances than letting the scheduler handle it itself. The scheduler is overly eager in attempting to move threads to a single core, and a bit slow at assigning a new core when the first core gets overloaded. It's the energy efficient way to do it, but latency sensitive applications might get degraded performance from this. This leads to short VAR spikes while it's shuffling the threads around, by assigning 1 core manually to each instance I do not experience this issue anymore. I did notice however that it depends a great deal on number of players and tickrate. I have HTT enabled and assign two cores on my 4770 to each of my 4 SRCDS instances (I assume 1 core+1 HT scheduler)

This probably works better for smaller servers with low player numbers ;)
When my servers near their capacity I usually prefer to let the scheduler move stuff around. It probably also depends on if you are also running a website and DB server