SRCDS Steam group


srcds performance boost with new CPU?
#1
Hi!

How much would i gain with a P4 3Ghz HT CPU against an Celeron D 3Ghz? I'm planning to host multipile servers on my machine. I have monitored the CPU usage so far and it's at resonable levels.

I wan't to host 3-4 TF2 servers if it's possible.
RAM and bandwidth won't be a problem.
The server is running Windows Server 2003.
Reply
#2
I wouldn't go past 2 servers with a single core P4 3ghz.
Reply
#3
Disable HT if you can, it basically splits the processor in half, so you get a dual core 1.5ghz... I hate ht...
[Image: b_350x20_C002748-004880-FFFFFF-000000.png]
Reply
#4
pengy Wrote:Disable HT if you can, it basically splits the processor in half, so you get a dual core 1.5ghz... I hate ht...

sorry, but bullshit. HT will create a virtual second cpu that can use parts of the cpu that are otherwise unused (e.g. some pipelines that are not needed for the code that it currently executes). Of course you will not really get 2 cpus, it's more like 1.5 cpus.


Btw: a P4 is a very old CPU. The Celeron is very similar to the P4 and only has a smaller cache. If you want half decent servers you will need completely new hardware. Core 2 CPUs are much faster per core (despite the have less GHz -> that is not a quantity to compare CPUs). I would not run more than 2 servers per core on a modern Core2 system, or maybe 3 servers if you don't need such high quality. On your P4 the limit is (to my experience) one public and one war server (that was before orange box, so maybe it's got worse!).
http://www.fpsmeter.org
http://wiki.fragaholics.de/index.php/EN:Linux_Optimization_Guide (Linux Kernel HOWTO!)
Do not ask technical questions via PM!
Reply
#5
Okay, it seems to be a bad idea to run many servers on that machine. And yes, i know that a Core 2 has much more performance per core than an P4/Celeron CPU. What i really wonderd was how much i would gain on the Hyper Threading.

At least i'm gonna try to run 2 servers on it when I get my upgraded connection, and see how it works out. My 8vs8 Arena server use about 30%-40% of the CPU when it's full, so I hope that i can run a additional server on it with decent performance.
Reply
#6
I ran a TF2 server on that exact CPU. Keep HT ON, it actually does help. However, don't expect to run much more then a single 24-man TF2 server. I would run a 24-man TF2 server and a small CS or other server that didn't get traffic.
~ Mooga ...w00t? - SRCDS.com on Twitter
[Image: 76561197965445574.png]
Please do not PM me for server related help
fqdn Wrote:if you've seen the any of the matrix movies, a game server is not all that different. it runs a version of the game that handles the entire world for each client connected. that's the 2 sentence explanation.
Reply
#7
I have noticed now that if i have srcdsfpsbooster running in the background the server drags about 20-40% (depending on connected players), but whitout it it only consume a couple % of the CPU power, 2-4% with 8 players connected. Can that really be true? I have checked the load with both Task Manager and Processexplorer, and both report the same thing.
Reply
#8
BehaartesEtwas Wrote:
pengy Wrote:Disable HT if you can, it basically splits the processor in half, so you get a dual core 1.5ghz... I hate ht...

sorry, but bullshit. HT will create a virtual second cpu that can use parts of the cpu that are otherwise unused (e.g. some pipelines that are not needed for the code that it currently executes). Of course you will not really get 2 cpus, it's more like 1.5 cpus.


Btw: a P4 is a very old CPU. The Celeron is very similar to the P4 and only has a smaller cache. If you want half decent servers you will need completely new hardware. Core 2 CPUs are much faster per core (despite the have less GHz -> that is not a quantity to compare CPUs). I would not run more than 2 servers per core on a modern Core2 system, or maybe 3 servers if you don't need such high quality. On your P4 the limit is (to my experience) one public and one war server (that was before orange box, so maybe it's got worse!).

P4's have high pipeline queue depth. They are generally very crappy, and HT only affects a very small amount of workloads. Disabling HT also decreases cacheline pingpongs and other things, which is generally more desirable for gameservers. P4's have very high syscall latency as well.

HT is NOT SMP, even those it installs APIC interrupts to almost have SMP support, it's NOT SMP, period.
http://leaf.dragonflybsd.org/~gary

“The two most common elements in the universe are hydrogen and stupidity.”








Reply
#9
I don't get what either of you are talking about pipelining, but turning off HT for multiple game servers was better for me...
[Image: b_350x20_C002748-004880-FFFFFF-000000.png]
Reply
#10
More or less that is what I was trying to say. HT on P4's is junk.
http://leaf.dragonflybsd.org/~gary

“The two most common elements in the universe are hydrogen and stupidity.”








Reply


Forum Jump:


Users browsing this thread: 3 Guest(s)