Xeon E5-2620

Hi guys

I'm planning a server based on the Xeon E5-2620 (6 cores, 12 threads)

It does

2500 MHz (1 or 2 cores)

2400 MHz (3 or 4 cores)

2300 MHz (5 or 6 cores)

according to http://www.cpu-world.com/CPUs/Xeon/Inte … -2620.html">http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E5-2620.html

Rig specs are

E5-2620 (C2 stepping)

16gb ram

300mbps down/100mbps up

Planning to run some TF2 servers at 66 tick. Would I be able to put 12 x 20 slot servers without them lagging?

I currently am running 6 servers at 20 slots each on a Phenom II x6 1055T and get an average of 45% cpu per core usage when the servers are active. I'm hoping to know how this translates to the Xeon

Cheers!

18 Replies

^ it's like reading a post in another language - lol.

Here's a transcription with annotations:

@jed:

I'd benchmark it. Start one server, fill it with bots (computer-simulated players) and let them play payload (a gameplay type), watch graphs, start two servers, watch graphs, and so forth. I wouldn't play tricks like pinning an instance of hl2ds (the Half-Life 2 Dedicated Server) to a core: Linux is pretty good at scheduling by itself in most cases.

MVM (Machine Versus Machine, a gameplay type) is a bit more resource-intensive because there are a shitload of projectiles whizzing around. You end up network-bound, which is why MVM is so laggy even on Valve servers. I've noticed that using the new projectile shield (a powerup) and catching shots from, say, Major Crits (a weapon that fires a large number of projectiles simultaneously), is also a lag magnet. I've wiped (died) a bunch due to that lag.

I have never run a TF2 server before, but based on the stats you provide here, I'd expect it to do just fine. If you haven't purchased parts yet, you might consider the E5 2620 v2 instead of the previous generation one. Regardless, both should greatly outperform a Phenom II thread for thread. Here's the CPU benchmark results for each:

Phenom: http://cpubenchmark.net/cpu.php?cpu=AMD … 55T&id=390">http://cpubenchmark.net/cpu.php?cpu=AMD+Phenom+II+X6+1055T&id=390

E5 2620: http://cpubenchmark.net/cpu.php?cpu=Int … Hz&id=1214">http://cpubenchmark.net/cpu.php?cpu=Intel+Xeon+E5-2620+%40+2.00GHz&id=1214

What about HT? If I have 2 similar applications running on the same physical core, set by affinity to the two logical cores, how will they impact each other? Will the 2ghz be effectively be divided by half? or would the scheduler just push in as much instructions from both threads into the space where the other thread did not request cpu time?

I've also found this (http://www.amdzone.com/phpbb3/viewtopic … 2&t=138958">http://www.amdzone.com/phpbb3/viewtopic.php?f=532&t=138958) to compare the architectual differences between sandy and thuban

Don't they have TF2 forums?

You'd probably get a better answers from people that actually run TF2 (I had to google "TF2" - had no clue what it was) then asking random people on a VPS board.

They do but I want to know from the pure CPU side of things. They don't really get technical over in the tf2 forums.

Common wisdom says you'll get on average about 10-20% increased performance from hyperthreading. It all depends on what kind of calculations you are doing. I did find this, though it doesn't effectively answer the question: http://facepunch.com/showthread.php?t=1106600

It would be best to test it yourself and see, though I would doubt that HT would cause performance to be any lower. The CPU scheduler is generally pretty smart about where to schedule things - I would only fiddle with affinity if you test and see that it can provide definite improvements.

that looks very promising. he is getting 10 to 18% CPU usage with 20 to 24 players in each server. my servers will average 16 players a server so i guess that helps.

@ntp:

that looks very promising. he is getting 10 to 18% CPU usage with 20 to 24 players in each server. my servers will average 16 players a server so i guess that helps.
For what it's worth, I've run TF2 at a similar tick on a straight Linode 512, when those existed. Didn't have an issue doing 12v12 payload. And that's on a Linode host machine shared with who knows what.

From all that I've been able to observe hl2ds is fairly light on the resource ask. Just watch the mods, as you probably know.

~~![](<URL url=)http://i.imgur.com/OPjC3JJ.png" />

Good luck, have a sandvich.~~

@jed:

@ntp:

that looks very promising. he is getting 10 to 18% CPU usage with 20 to 24 players in each server. my servers will average 16 players a server so i guess that helps.
For what it's worth, I've run TF2 at a similar tick on a straight Linode 512, when those existed. Didn't have an issue doing 12v12 payload. And that's on a Linode host machine shared with who knows what.

From all that I've been able to observe hl2ds is fairly light on the resource ask. Just watch the mods, as you probably know.

~~![](<URL url=)http://i.imgur.com/OPjC3JJ.png" />

Good luck, have a sandvich. Thank you for the tasty sandvich!

Would your saying of hl2ds being light on the resource ask would be the because it does not eat cpu cycles all the time?

I'm not really sure what to make of the htop cpu usage graphs but that sounds like I can get away with running 2 hl2ds running on the same physical core (hyperthreaded) so that they each fill the pockets of cpu time left over by the other.

I'm also trying to grasp http://facepunch.com/showthread.php?t=1106600 where he says "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%:" The cpu utilization graph shows each HT core being above 50%.~~

I'd benchmark it. Start one server, fill it with bots and let them play payload, watch graphs, start two servers, watch graphs, and so forth. I wouldn't play tricks like pinning an instance of hl2ds to a core: Linux is pretty good at scheduling by itself in most cases.

MVM is a bit more resource-intensive because there are a shitload of projectiles whizzing around. You end up network-bound, which is why MVM is so laggy even on Valve servers. I've noticed that using the new projectile shield and catching shots from, say, Major Crits, is also a lag magnet. I've wiped a bunch due to that lag.

Mann vs Machine. Wiping is the death of a whole group, too – usually on the harder MvM maps, the medic (often me) dying = wipe.

NOOB

TF2 is so different to me from TF, at least DoD Source didn't change all that much from DoD.

Whups, my bad. I stopped playing TF2 long before Mann vs Machine. Long before there were hats, actually. Back when TF2 was Team Fortress ;)

I'm slowly starting to like TF2, adding that to my collection of favs.

I'd love to benchmark it but I'm unable to until I actually buy the server. However, I also got offered an i7 3930k based system which has a much higher clockspeed over the Xeon. I also have a current E3-1230 v3 (haswell) based server running comfortable with 2 instances per physical core (1 per logical) so I'm guessing that it scales the same with the i7 3930k (sandy).

The E3 Haswells are pretty beast. I'm running a 1230 v3 with 32 GB of RAM, and I'm running out of RAM way before I'm running out of CPU power. If only Intel would make them able to handle 64 GB of RAM….

The Xeon haswells look like desktop chips with ECC enabled to me. And since some of the desktop chips have ECC enabled, I'm not even sure if they have that going for them.

Then again, the only i7-class chip with ECC enabled is the 4700EQ, which is only available in a soldered-to-motherboard configuration. I don't think any non-soldered desktop chips have ECC above 2 cores.

Reply

Please enter an answer
Tips:

You can mention users to notify them: @username

You can use Markdown to format your question. For more examples see the Markdown Cheatsheet.

> I’m a blockquote.

I’m a blockquote.

[I'm a link] (https://www.google.com)

I'm a link

**I am bold** I am bold

*I am italicized* I am italicized

Community Code of Conduct