i7-3930K vs Opteron 6272...

[UberApfel]

... for hosting 30+ minecraft server instances. Impossible to find on google. Thoughts?

If your reasoning is "not bulldozer cause amd is stupid"; you need not apply your bigotry.

 

[BallaTheFeared]

I'd assume the Opty would work best, but honestly I have no idea.

2.1GHz bulldozer cores seems dreadful though, but I have no idea the kind of load 30+ minecraft servers would put on either processor so ignore my post I guess

 

[Smoblikat]

Id say the opteron, for two reasons:
1) Its cheaper and is designed for server use.
2) I run a couple minecraft servers off of my 2500K at stock and i never have any CPU limitations while running GTAIV and the servers at the same time, so that tells me that its core count not speed that impacts performance, since the opty has 10 more physical cores i would go with that.

 

[dma0991]

Why 30 instances, are you providing some hosting services or some sort?

My choice goes to the Core i7 3930K as MC prefers stronger single threaded performance, as with any game for that matter. Also you would not gain much from having a 16 core CPU handling all of them unless you're setting core affinity to each running server instance. The cheapest LGA2011 board can be had for less money than a Socket G34 motherboard.

 

[Accord99]

... for hosting 30+ minecraft server instances. Impossible to find on google. Thoughts?

If your reasoning is "not bulldozer cause amd is stupid"; you need not apply your bigotry.

If you're comparing 1 socket systems, the i7 is capable of more throughput and thereby will do more work regardless of the number of CPU intensive threads running.

 

[Grooveriding]

My guess would be Intel. Game server hosting companies are all using 4c/8t xeons, that should be an indicator of what best does the job. Many use Xeon E3-1270/1275s on their best boxes, a $350 CPU and a motherboard can be had for about $150-$200.

I think that would be your best bet. I have no idea what sort of load these boxes handle though and if 30 minecraft servers falls into that amount of load. If you check with RSPs and their dedicated box hosting they will give you an indication of what each machine can handle. I know a premium dedicated box can handle about 300 player slots of BF3, no idea how that translates to Minecraft though of course

The way the dedicated boxes work in general is each server runs on a thread afaik and they are running some version of linux.

So a direct answer would be the 3930K is quite likely the best choice to being faster IPC and faster than one of the xeons I mentioned earlier due to a better clock speed and more threads. It could be overkill though in both performance and cost depending on what demands a minecraft server places on the hardware. The load is also going to be variable depending on how many users are on the servers of course.

 

[Diogenes2]

Can't really find any side by side tests but here are a couple of items from openbenchmarking.org

They show the results for the Apache Benchmark among other things.
I don't know how much difference the operating system makes, but the 3930K is running Centos and the Opteron Linux

http://openbenchmarking.org/result/1201155-AR-TEST1985626
http://openbenchmarking.org/result/1111154-LI-TEST0215431

The 3930k is beating the Opteron 6272 by about 40% ..

 

[ed29a]

... for hosting 30+ minecraft server instances. Impossible to find on google. Thoughts?

If your reasoning is "not bulldozer cause amd is stupid"; you need not apply your bigotry.

People run Minecraft servers on laptops and old machines, so I doubt any serious IPC is required to run it. IMHO, you need two things: as many physical cores as possible and a lot, and I mean a lot of RAM.

 

[dma0991]

People run Minecraft servers on laptops and old machines, so I doubt any serious IPC is required to run it. IMHO, you need two things: as many physical cores as possible and a lot, and I mean a lot of RAM.

It is definitely possible to run Minecraft on laptops and old PCs but I've tried playing and hosting a server at the same time on a C2E laptop and it lags severely with/without core affinity set to each task. The lowest end 'old PC' server that I've seen so far should be no less than a C2D, for a single server.

 

[MisterMac]

Look at all the minecraft server distros and look at the code.
Look at the documentation.
All state the same.

Minecraft is an amateur project, and is SHITTILY coded (in all comparison to a professional project) - it is NOT multithreaded and runs single thread instances.


Therefor the FASTEST possible single thread performance will be the clear winner, in this case CLEARLY the 3930k will outperform.
IPC and Clockspeed will matter, and here 3930k shines bight.

Even with 4 more threads, the bulldozer wouldn't have an advantage with 30 single connected instances(Culprit being java/single threaded code).
Possibly 16 vs 16 instances bulldozer would have a slight advantage, but it would require benching and heavily depend on workload on each instance.
A bit far fatched at most, id wager.
The 3930k has monster clock advantage + ipc advantage, so it'd most likely even win in such a benchmark.

From experience running various "private server" games leaked and emulated, the bottleneck always occurs based on wether it's real professional code or emulated java things.

Your fukked on the emulated side by:
A.ANY managed language(.NET/C#, JAVA, python blabla).
B. Piss poor coding from amateurs.
(Coded to work, not to work efficiently, smartly or in scalable conditions).

Any professional project however usually gives you headaches on:
a. Diskspeeds (Lets face it, any dedicated server renter on the web generally has shitty quality hardware).
b. Monsterous Cachching of Ram (In case tons of players log on, and it's needed) - In short, MOAR RAM!
(Coded to the opposite of the above, with expectations of quality grade hardware).

Not knowing minecraft or bothering to look at it's source, i will gaurentee you, if you plan to run 30 instances of some java based server, your gonna get bottlenecked with either CPU due to the Architecture and Code.

But if you were run to several instance heavily full of workload, the 3930k should emerge as a CLEAR winner.
And in general should give you the advantage.

Your scenario however assuming same level with some degree of workload will most likely give you a headache no matter what.

EDIT ONE OH ONE OH ONE OH ONE:

After fumbling around the net, some Minecraft distro's seem to have used some of java's own libs to create their own "Scheduling system".

One thing makes them all common and that is they by design give you access to the main thread, which they're own documentation clearly states:
"Can lead to server freezes".

I wouldn't trust that with a heartache.

go IPC, go Clock.
Can't go wrong then.

 

[dma0991]

^
Actually Minecraft is quite multithreaded and will run off more than a single core at any given time, I play it so I should know. Maximum amount of cores it is able to utilize I am not sure but it definitely is capable of using 2 threads or more. Most games that are 'professionally' coded are not even well threaded and we're still stuck with games that perform optimally with 4 cores.

I don't know if Java coders could be defined as shitty as compared to the norm of using C++ for bigger games but bugs exists in any games, Skyrim is no exception and requires quite a few mods here and there to optimize its performance. Minecraft may have its slight drawbacks with using Java but it is definitely not lacking in community made mods which expands its possibility or as you would prefer, improve its overall performance via OptiFine.

Every world is randomly generated and new parts of the world will be generated as you move further so there will only be a requirement during the world generation phase. As it is a sandbox game, I could crash it easily by spamming it with something till it crashes, with other games, you wouldn't be able to crash it unless it is a bug.

 

[cytg111]

... and is SHITTILY coded ..

- Factisize(tm) please.

 

[cytg111]

....
Any professional project however usually gives you headaches on:
a. Diskspeeds (Lets face it, any dedicated server renter on the web generally has shitty quality hardware).
b. Monsterous Cachching of Ram (In case tons of players log on, and it's needed) - In short, MOAR RAM!
(Coded to the opposite of the above, with expectations of quality grade hardware).
....

- I like your posts and insight, make no mistake, but seriously, a "professional project" is hardly a recipe for "proffesional code"... anyone with the ability to hit "compile" and get "success" is ready to do damage *anywhere* .. and i've seen it anywhere (maybe even done a little myself)..

 

[blckgrffn]

Maybe a compromise? A 12 core Opteron with better clock speeds?

I would really consider ECC ram.

Thought #2 - with the opteron you can buy a two proc board and just use one socket. If/when you decide you need more oomph, just buy another CPU and re-balance/buy more ram.

Are you going to be using a HyperVisor? ESXi 5 would work great but is limited to 32GB of ram in the box. Not sure if that is a hard limit or EULA enforced though... Anyway, that could make your life simpler, ie if one host OS needs bouncing you aren't trashing all the environments. I sincerely hope you'll consider using virtualization, this is a text book case for it Run three instances per VM or something to minimize your risk exposure.

And if you find one box isn't cutting it, you move some VMs around and no one loses their stuff. So, how do I vote for two Thuban/BD boxes or two 2x00k boxes? Compute density costs money, you may be able to do it cheaper by spreading it across two or more cheaper platforms.

 

[LOL_Wut_Axel]

Core scaling decreases as you add cores, and the 3930K has 60%+ higher IPC than the 6272. It can also overclock with very little additional voltage to 4.3-4.4GHz, while the Opteron can overclock very little if any. That means 60%+ faster per-core at the same clocks, and the 3930K can be easily clocked at 100% higher frequency. Just from the additional single-threaded performance you're looking at removing any and all deficiencies in multi-threaded and actually gaining performance. For multi-threading the HT helps it achieve anywhere from 20-25% higher performance than without.

TL;DR: The 3930K will be faster.

 

[Kevmanw430]

I run a single Minecraft server off a P4 HT 3GHz, and 2 GB RAM. I would say either would be fine.

I think the 3930K would be better without any tweaks or anything, but if you set affinity for all of the instinces to use the different cores, it could work better on the Opteron. Regradless, I'd say go with the 3930K, just faster overall, and you could OC if you needed to.

 

[Arachnotronic]

3930K. No contest.

 

[Smartazz]

I'm surprised the 3930K can beat one of AMD's highest end server processors. I'd imagine the 8 core Sandy Bridge Xeon is going to be amazing.

 

[RavenSEAL]

Forget the 3930K, get a cheap opty and buy a lot of ram.

 

[MisterMac]

^
Actually Minecraft is quite multithreaded and will run off more than a single core at any given time, I play it so I should know. Maximum amount of cores it is able to utilize I am not sure but it definitely is capable of using 2 threads or more. Most games that are 'professionally' coded are not even well threaded and we're still stuck with games that perform optimally with 4 cores.

I don't know if Java coders could be defined as shitty as compared to the norm of using C++ for bigger games but bugs exists in any games, Skyrim is no exception and requires quite a few mods here and there to optimize its performance. Minecraft may have its slight drawbacks with using Java but it is definitely not lacking in community made mods which expands its possibility or as you would prefer, improve its overall performance via OptiFine.

Every world is randomly generated and new parts of the world will be generated as you move further so there will only be a requirement during the world generation phase. As it is a sandbox game, I could crash it easily by spamming it with something till it crashes, with other games, you wouldn't be able to crash it unless it is a bug.

Id reply to you, but you make no sense and don't read what i wrote nor what the original OP wrote.

Your just trying to defend and sound smart.
I .....heavily dislike people.... like that.

Your comparing languages by saying because games written in both languages have bugs, they're practicly the same.
Then your comparing a Server application to a Client one.

Not really starting of well, are ya?

So here's what you'll do:

Find me a minecraft distro, where you show me native object/que management, async/sync management of threads that's not just plying on Java VM's garbage management.

I dare you.



@cytg11:

Yes, i am generalizing alot.
I apologize for that.
But i'm also right in 90% of the cases.
Any complicated workload with lots of inheritance and data variable management in memory is going to suck on any managed language.
Cause the language was never meant for that, period.

Minecraft is a fantastic idea and game, and obviously not a big budget project - so the tools used to create it reflect this.
This is not a problem of minecraft but more a problem of the tool's ultimate limitations.
And as say yourself, there's a difference from professional project to professional code - and with a non-constant workload oriented language as base - you can see how it's easier to run into "performance issues" depending on who has written what


Still TL;DR
Clockspeed/IPC for anything written in a managed language.
I highly doubt minecraft servers carries 1000+ sockets with tons of complex mods that require 100+ active new variables per user, so the chances of running into performance issues with some decent hardware(the procs mentioned should qualify as that), are minimal.

 

[dma0991]

I dare you.

I'm not preaching to be an expert in Minecraft nor Java. To say that Minecraft was poorly written in the first place isn't true either as the server client can be just as multithreaded. I'm not even defending what I'm saying, I'm just very displeased when you're undermining the efforts the developers as though you could do a better job than they did.

Replying back to the statement that you've claimed,

Your fukked on the emulated side by:
A.ANY managed language(.NET/C#, JAVA, python blabla).
B. Piss poor coding from amateurs.
(Coded to work, not to work efficiently, smartly or in scalable conditions).

If you're saying that there are no difference in terms of the language it was written, then what's wrong with using a managed language in the first place? How did you come to a conclusion that it was piss poor coding done by amateurs?

 

[Diogenes2]

Forget the 3930K, get a cheap opty and buy a lot of ram.

The OP asked:

[URL="http://forums.anandtech.com/showthread.php?p=32922242#post32922242"]i7-3930K vs Opteron 6272...

?

[/URL]The 3930K offers better performance.

If you want to go cheap, without considering performance then yes, there are other choices.

 

[Smartazz]

The OP asked:

[/B]The 3930K offers better performance.

If you want to go cheap, without considering performance then yes, there are other choices.

The Opteron 6272 also isn't that much cheaper than the 3930K. For the OP's purpose I also recommend the 3930K.

 

[UberApfel]

IPC or single-threaded performance shouldn't matter here unless we're under the assumption that the scheduler is retarded.

The Opteron 6272 has 16 'cores'; 30 minecraft instances... ideally the most needy will get their own cores and the less needy will be coupled to an average of 3 per core. Perhaps each module gets 1 needy java instance and has a couple idlers on the module's other core.

 

[Diogenes2]

The 16 cores don't matter if the 6 cores ( 12 threads ) of the 3930K gets the job done faster/more efficiently..