i7-3930K vs Opteron 6272...

[MisterMac]

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,

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?

I can do a better job than what they did.
So that is exactly what i'm claiming.
As long as i'm allowed to use whatever language i want.
And i can undermine it all i want, cause the objective of Minecraft wasn't to make a damn good networking and memory engine for an MMO that can handle thousands of users and sockets.


Secondly, i'm SAYING THERE IS, a humoungous difference from efficiently written code from language to language, based on what the hell the language was designed for in the first place.

How can you not read? Seriously?
Your post are lol full of ignoring what i'm saying while sticking to your arguments - and not counterarguing me.

And no, there is NO multithreading in the server side.

Find me the code, even find me a closed build that claims to have it of a minecraft distro.


I'd wager you shouldn't post without actually reading my posts or your own before clicking the go button.

 

[dma0991]

And no, there is NO multithreading in the server side.

Server requirements

• The more memory your computer has, the better. It's recommended to have more than 2 GB, that way plenty of ram is left over for the operating system, and you have the possibility of allocating more to the Minecraft server.
• Minecraft servers, as of 1.0.0, can use multiple cores, so now they won't sit idle.

Source

 

[UberApfel]

Why is everyone so certain the i7-3930k outperforms the Opteron 6272? Further so... in Java?

This isn't an application where a single program is expected to scale across an entire processor. Each cores's simultaneous processing power added together -- the maximum throughput -- isn't that what should matter here?

 

[blckgrffn]

Why is everyone so certain the i7-3930k outperforms the Opteron 6272? Further so... in Java?

This isn't an application where a single program is expected to scale across an entire processor. Each cores's simultaneous processing power added together -- the maximum throughput -- isn't that what should matter here?

I think people are thinking about the VM benchmarks that were run here at anandtech where a 2P 6C Xeon walked all over a 2P 16C Interlagos in a number of scenarios (including VM tiles) while using less juice.

Interlagos is probably worth looking at if you are going to be awesome and get greater than 48GB of ram for the box.

I am still standing beside my 2x (or more) smaller computers running VMs. Don't pay for density unless you need it, IMHO. We aren't talking about building a high scoring folding rig here...

 

[MisterMac]

Source

Not multithreaded, reusing java api's for java VM to execute resources for multiple cores available for the operating system.

This means all execution speed gains from L2/L3 Cache are flushed, as it's handled by extra stage from the Java VM.


PS.:

I wanted to be shown in the specific source, cause i don't see it. And your mistake i also already mentioned btw.
You really don't read much other than definitions and assume it covers all.

 

[blckgrffn]

Not multithreaded, reusing java api's for java VM to execute resources for multiple cores available for the operating system.

This means all execution speed gains from L2/L3 Cache are flushed, as it's handled by extra stage from the Java VM.


PS.:

I wanted to be shown in the specific source, cause i don't see it. And your mistake i also already mentioned btw.
You really don't read much other than definitions and assume it covers all.

Is Minecraft open source at all? Do you really expect to see any source code here?

I am not really doubting your expertise here, but going around on the Internet and demanding to see source code when conversations like this come up must get you ignored more often then not?

 

[MisterMac]

Is Minecraft open source at all? Do you really expect to see any source code here?

I am not really doubting your expertise here, but going around on the Internet and demanding to see source code when conversations like this come up must get you ignored more often then not?

Minecraft has been relased to the opensource after a big snahoozle regarding a small insignificant fee for svn access(apparently minecraft people are cheap?).
Or atleast my google-fu picks that up on newssites.


Your telling me he released a complete free source snapshot and then made it private?


That sounds weird, considering all the mod distro's out there.

 

[dma0991]

There are no source codes, even if there were it was during the alpha, beta and release candidate stages which you could definitely not find in the 1.0.0 which is the launch release. Minecraft is a semi open source and only open source after the game has passed its prime.

If you're searching for a certain part of the source code to prove that it is multithreaded you may do it yourself, I don't have the time to scan thousands of lines just to prove a point. You may do it at your own leisure, I'm done here.

 

[LOL_Wut_Axel]

Oh, for Christ's sake. Stop recommending the Opteron out of ignorance and read.

 

[ed29a]

Oh, for Christ's sake. Stop recommending the Opteron out of ignorance and read.

That's really golden. You complain about ignorance.

How many lines of server application code you have written?
How many servers do you regularly maintain?
How many server applications have you designed?
What is your server background?

Please enlighten us of your server experience since you like to make a lot of blanket statements that obviously show you have no clue what you are talking about.

PS: seasonal job at CompUSA selling crap doesn't count as server experience.

PPS: From Anand: Again, the Opteron 6276 delivers a very respectable performance per dollar, delivering 96% of a Xeon that costs almost twice as much. Any more brilliant suggestions or server benchmarks that you fail to understand?

 

[LOL_Wut_Axel]

That's really golden. You complain about ignorance.

How many lines of server application code you have written?
How many servers do you regularly maintain?
How many server applications have you designed?
What is your server background?

Please enlighten us of your server experience since you like to make a lot of blanket statements that obviously show you have no clue what you are talking about.

PS: seasonal job at CompUSA selling crap doesn't count as server experience.

PPS: From Anand: Again, the Opteron 6276 delivers a very respectable performance per dollar, delivering 96% of a Xeon that costs almost twice as much. Any more brilliant suggestions or server benchmarks that you fail to understand?

Missing the point entirely. A single of the Xeon X5650 or X5670 they used for testing is much slower than a 3930K because it uses the Nehalem architecture and runs at much lower clock speeds. It also has 6 cores/12 threads, just like the 3930K. They have two of them, and as we all know core scaling decreases as you add cores so that means SB-E looks even better given all AMD can do is add cores to compensate for their CPU department's lazy engineering.

I still don't get how people keep recommending it because it has "MOAR COREZ!!!" when each of those cores is over 50% slower clock-for-clock than a Sandy Bridge core plus it'll be running at 50% lower clock speed. Each core of the 3930K can execute more than twice the number of instructions than a single Opteron 6272 can at any given time, and your argument regarding price is completely null because the Xeon was used to drive home the point that even in heavily multi-threaded scenarios more cores won't save you when your competitor can do much more per clock cycle and they can still get more performance for very little increase in power with SMT. It wasn't used to compare the Xeon to an Opteron, but to compare Nehalem to Interlagos, and we all know Sandy Bridge-E>Nehalem.

Oh, and please spare me the crappy insults. Reported for flaming.

 

[blckgrffn]

I'd like to point out here that even Facebook uses AMD for their caching servers due to the ability to more a have great memory density per system.

Since this is a Java Application, it would be very advisable to get as much ram as possible. An Opteron platform will obviously allow to pack more ram into a given server.

I still don't think one server is the cost effective solution in this case

 

[ed29a]

as we all know core scaling decreases as you add cores so that means SB-E looks even better

Got it. You have no clue.

given all AMD can do is add cores to compensate for their CPU department's lazy engineering.

I see you are a genius CPU designer, what's your background? It's easy to be armchair CPU designer. Isn't? The fact is that server sides, Interlagos is competitive with Xeons and costing far less is pretty damn nice. Is it the holy grail of CPUs? Hell no, it has many issues, but it's not the piece of monkey crap that you are constantly complaining about it.

I still don't get how people keep recommending it because it has "MOAR COREZ!!!" when each of those cores is over 50% slower clock-for-clock than a Sandy Bridge core plus it'll be running at 50% lower clock speed.

Got it. You have no clue. Ever wonder why Microsoft is begging Intel for 8-16 core Atoms? When you understand that quite a lot of server processes wait for disk IO, network IO or user IO don't need high IPC, you can maybe grasp the simple concept of this problem. OP wants to run thirty, let that sink in, 30 instances of a server application. If one of those instances is lagging or stuck waiting on IO, semaphores or some really bad code, guess what the core is going to do (especially since people say that MC code is crap)? It will wait. If a core waits at 1Mhz or at 100000000Ghz it matters not. It. Just. Waits. Thus when you got more physical cores, it helps the entire system be much more fluid and the other 29 instances won't be stuck waiting for processor time and processor won't be spending its time context switching. Kapish?

Each core of the 3930K can execute more than twice the number of instructions than a single Opteron 6272 can at any given time,

Got it. You got no clue. If super duper Xeons does 1000000 NOPs per cycle, vs Opteron that does 1000 NOPs, does it matter? Or when one waits for IO callback, does it matter? Oooooh, it will wait 'faster'. /golfclap

and your argument regarding price is completely null because the Xeon was used to drive home the point that even in heavily multi-threaded scenarios more cores won't save you when your competitor can do much more per clock cycle

Got it. You have no clue. Point of your graph is that for half the price you get 96% of the performance of that Xeon. My other point was is that you don't understand jack about servers because you used an image to invalidate your own point. Once you understand that IPC is not the holy grail of severs, maybe you'll se why an Opteron would be better for OP.

Oh, and please spare me the crappy insults. Reported for flaming.

Crappy insults? I was wondering what was your server background. I see now: nada, zilch, nothing.

 

[LOL_Wut_Axel]

Got it. You have no clue.


I see you are a genius CPU designer, what's your background? It's easy to be armchair CPU designer. Isn't? The fact is that server sides, Interlagos is competitive with Xeons and costing far less is pretty damn nice. Is it the holy grail of CPUs? Hell no, it has many issues, but it's not the piece of monkey crap that you are constantly complaining about it.


Got it. You have no clue. Ever wonder why Microsoft is begging Intel for 8-16 core Atoms? When you understand that quite a lot of server processes wait for disk IO, network IO or user IO don't need high IPC, you can maybe grasp the simple concept of this problem. OP wants to run thirty, let that sink in, 30 instances of a server application. If one of those instances is lagging or stuck waiting on IO, semaphores or some really bad code, guess what the core is going to do (especially since people say that MC code is crap)? It will wait. If a core waits at 1Mhz or at 100000000Ghz it matters not. It. Just. Waits. Thus when you got more physical cores, it helps the entire system be much more fluid and the other 29 instances won't be stuck waiting for processor time and processor won't be spending its time context switching. Kapish?


Got it. You got no clue. If super duper Xeons does 1000000 NOPs per cycle, vs Opteron that does 1000 NOPs, does it matter? Or when one waits for IO callback, does it matter? Oooooh, it will wait 'faster'. /golfclap


Got it. You have no clue. Point of your graph is that for half the price you get 96% of the performance of that Xeon. My other point was is that you don't understand jack about servers because you used an image to invalidate your own point. Once you understand that IPC is not the holy grail of severs, maybe you'll se why an Opteron would be better for OP.


Crappy insults? I was wondering what was your server background. I see now: nada, zilch, nothing.

Repeating the same crap ad-nauseum doesn't make you right, especially when you've yet to provide one single fact.

Also, we're comparing Sandy Bridge-E to Interlagos, in case you missed the memo. The Xeon only works for pointing at the Nehalem architecture and the OP won't buy one.

 

[ed29a]

Repeating the same crap ad-nauseum doesn't make you right, especially when you've yet to provide one single fact.

I am repeating because you don't want to or are incapable to understand certain basic server loads. If you have an application that waits a lot (Minecraft server) for various IO operations, IPC isn't a prime quality you want in a CPU. My facts are decades of computer engineering.

Fact: Minecraft server CPUs don't spend 100% of time at max CPU usage. They don't do massive number crunching. CPUs for MC servers wait. A lot.
Fact: When CPU waits for IO, it doesn't matter if it's a super duper mega hyper Xeon or a slow Atom, it waits.
Fact: Embarrassingly parallel tasks (30 Minecraft instances) that don't require intense IPC, benefit far more from more cores than IPC. You can even assign affinities to each process and glue them to a specific core, because you have 16.
Fact: 250$ Opteron boards support 16 core chips (550$) and up to 128 GB of RAM. RAM is one of the key ingredients to run a lot of instances of this application.

If Minecraft was doing nothing more than crunching numbers, I would agree, get the highest IPC possible. But it's a damn server application that runs fine on a 4 year old laptop because all it does it waits on sockets, read something from a data store, write stuff to it and send responses back to client. All of this is network and disk IO latency limited that is independent of CPU speed.

 

[jsedlak]

I don't see 30 instances needing more than 64gb, which the 3930k can support (with all 8 slots used). I guess you would need more if you assign more than 2gb to each instance though.

If they wait a lot, isn't their less of a chance of them all needing to hit the CPU hard for an extended time?

 

[gorydetails]

wow,,,a java server,,,,u mite need a superpc to run that,,,

 

[Ferzerp]

PPS: From Anand: Again, the Opteron 6276 delivers a very respectable performance per dollar, delivering 96% of a Xeon that costs almost twice as much. Any more brilliant suggestions or server benchmarks that you fail to understand?

That was a problematic statement because when you buy servers, there is near price parity between the two configurations used in that test. Looking at the retail cost of that singular component, that was an accurate statement, but when you quote out the two systems used, you end up with prices within 2-3% of one another.

I know, because when I read that article, I immediately did so. The discount levels we get are decent, but are not weighted in favor of intel over amd at all.

 

[blckgrffn]

I don't see 30 instances needing more than 64gb, which the 3930k can support (with all 8 slots used). I guess you would need more if you assign more than 2gb to each instance though.

If they wait a lot, isn't their less of a chance of them all needing to hit the CPU hard for an extended time?

I would think then we'd be doing a lot pipeline flushes/context switches if the process doesn't actually stall the pipeline (as you pointed out). This is expensive, time wise, as well, and having more cores would allow you to wait more often without this unloading, I would think.

Java will use however much memory is available, more is better Also, this is why Java apps can suck. They are certainly a bummer in virtual environments...

 

[LOL_Wut_Axel]

I am repeating because you don't want to or are incapable to understand certain basic server loads. If you have an application that waits a lot (Minecraft server) for various IO operations, IPC isn't a prime quality you want in a CPU. My facts are decades of computer engineering.

Fact: Minecraft server CPUs don't spend 100% of time at max CPU usage. They don't do massive number crunching. CPUs for MC servers wait. A lot.
Fact: When CPU waits for IO, it doesn't matter if it's a super duper mega hyper Xeon or a slow Atom, it waits.
Fact: Embarrassingly parallel tasks (30 Minecraft instances) that don't require intense IPC, benefit far more from more cores than IPC. You can even assign affinities to each process and glue them to a specific core, because you have 16.
Fact: 250$ Opteron boards support 16 core chips (550$) and up to 128 GB of RAM. RAM is one of the key ingredients to run a lot of instances of this application.

If Minecraft was doing nothing more than crunching numbers, I would agree, get the highest IPC possible. But it's a damn server application that runs fine on a 4 year old laptop because all it does it waits on sockets, read something from a data store, write stuff to it and send responses back to client. All of this is network and disk IO latency limited that is independent of CPU speed.

The key ingredient you forgot is that for it to support those 128GBs of RAM you need to buy ECC RAM. Those single socket server motherboards are limited to 32GB when running non-ECC, meaning they officially only support 4GB DIMMs of non-ECC RAM in each of the eight slots. That's an additional cost, and for what it'll be used for there's no benefit in running ECC other than enabling official support for 128GB. With Sandy Bridge-E you can still run 64GB of RAM via 8x8GB DIMMs and you don't need to be limited to ECC RAM either to get that much.

Also, single-threaded performance is still a key factor no matter how much you want to deny it. Just as an application will take advantage of many CPU cores to work in parallel, it will also take advantage of a CPU that can deliver more per clock cycles. If this wasn't true, you might want to explain to me the graph I linked above. Perhaps I'm crazy enough to think that running various VM instances isn't multi-threaded.

 

[jsedlak]

So... who is actually going to build and test these two machines? I kind of want to know and see it be worked out.

 

[ed29a]

Perhaps I'm crazy enough to think that running various VM instances isn't multi-threaded.

Wow. You lack even the basic knowledge of threads, processes and processors.

And I am crazy enough to argue about server loads with you.

Wow.

I am speechless.

 

[Arachnotronic]

I'd go with the Opteron if you're looking to optimize performance/$ for Minecraft servers.

 

[LOL_Wut_Axel]

Wow. You lack even the basic knowledge of threads, processes and processors.

And I am crazy enough to argue about server loads with you.

Wow.

I am speechless.

No, you're clueless. It was made to point out how absurd what you're saying is. If you didn't catch the sarcasm then perhaps that's why you're having issues.

Also, nice job at not counter arguing ANYTHING at all and attacking something that was clearly sarcasm. Bravo!

Now go back under your bridge.

Pot calling the kettle black ? You need to knock this off. If anyone is trolling here, its you.
Next warning is an infraction
Markfw900
Anandtech Moderator

 

[Arachnotronic]

Now go back under your bridge.

Most enthusiasts these days are using...sandy...bridges. Har Har Har...