Xeon comparison

[pubairplanes]

Hey all,

This should be an easy question. For a linux server, which is faster? A dual core Xeon 3060 or quad Xeon 5310?

Thanks

 

[brian]

well you cant look at them the same way. the quad has great features like being a quad core and the 8 MB cache but the dual core is technically faster. For a server, i would go with the quad because it is more important to be able to put more processes through.

 

[pubairplanes]

That makes sense. It will be used in a server.

So even though I'm loosing a lot of clock speed the extra cores are more valuable on a busy server?

 

[Vizy]

Yes. Clocks wouldn't matter drastically for a server.

 

[brian]

well mostly..

you can think of it like this

Lets say we have a car. this car represents a processor.
This car can go 2.6 MPH but only has room to cary one person.
so to go from the start 2.6 miles would take a hour and the car then drops off the person and has to go back to get a new one.

However a quad core is like having 4 cars going 2.6 MPH
They go the same speed, however you are able to get more people across in the same about of time. Its not faster, but more efficient.

 

[Vizy]

lol,

Very well put brian!

 

[brian]

thanks, used that once before. long time ago. (i hear by licence this under the GNU public license )

 

[Vizy]

Lol. That has to be the best way I've ever seen it put.

 

[TrainTrackHack]

In general, adding up the frequencies of the cores is incorrect; however, when talking about applications that can actually utilize all the cores, you can add the frequencies up to get the effective performance of the CPU... so, for example, let's say we have an application that runs 4 or more threads. It can utilize all cores of a dual-core CPU, so if we have a 3GHz dual-core, it means that the program would perform equally well on a 6GHz single-core CPU (providing it's based on the same architecture). However, if the program was run on a 2.4GHz quad-core, it would run faster, since there are 4 cores clocked at 2.4GHz, totalling 9.6GHz. Therefore, assuming that the application can utilize 4 or more cores (and any server application should be able of doing that), a slower-clocked quad will be your best bet.

 

[lovely?]

yeah we could technically do that but because of the diminishing returns rule it is unpractical because it can be off by a good bit.

but still, it could work here.

 

[TrainTrackHack]

yeah we could technically do that but because of the diminishing returns rule it is unpractical because it can be off by a good bit.

And what's that?

 

[lovely?]

And what's that?

well the higher your mhz the higher your voltage inevitably is, but soon you need many more volts to gain a few mhz, so you get diminishing returns per mhz you overclock.

 

[TrainTrackHack]

well the higher your mhz the higher your voltage inevitably is, but soon you need many more volts to gain a few mhz, so you get diminishing returns per mhz you overclock.

...but how does it relate to this topic (did I miss something?)

 

[lovely?]

no you didn't i was merely mentioning that the way you said it cannot apply to all situations, or many situations at all except this one.

 

[simpletron]

the law of diminishing returns can deal with as you increase the number of cores, you don't get proportionally more performance. think about sli or crossfire. when you have two cards, is it twice as fast as one? ... no. when you have 3 cards is it 3 times faster than one or 50% faster than two? ...no. there is a similarly for cpus, but the affect isn't as larger in good code. this is because there is an overhead associated with creating the threads and this doesn't increase linearly.

to use brian's car analogy, the overhead is like every car towing a trailer with each trailer being bigger than the last one(Car1's trailer is the smallest, car2's trailer is bigger than car1's, car3's trailer is bigger than car2's,...). the bigger the trailer is the more the car slows down like car1 going at 2.6mph, car2 going at 2.5mph, car3 going at 2.4mph,... in very good code the starting size is small and doesn't increase much, so it is very close to porportionally faster. like car1 going at 2.6mph, car2 going at 2.5999mph, car3 going at 2.5998mph,... server code tends to be well written for a lot of cores(4+), so going with the quad core is your best bet.

 

[pubairplanes]

Ok thanks for the input everyone. I'll go with the quad.

that was a ridic awesome analogy brian haha.

 

[TrainTrackHack]

the law of diminishing returns can deal with as you increase the number of cores, you don't get proportionally more performance. think about sli or crossfire. when you have two cards, is it twice as fast as one? ... no. when you have 3 cards is it 3 times faster than one or 50% faster than two? ...no. there is a similarly for cpus, but the affect isn't as larger in good code. this is because there is an overhead associated with creating the threads and this doesn't increase linearly.

No, it doesn't work like that for CPUs. A well-coded application can -and will- gain 100% more perormance when the number of cores is doubled, depending on how many cores it can utilize; maybe even more, since the sheduler overhead is lowered (less switching between active threads). In this case, "bad code" simply doesn't utilize the other core(s); the power not used, however, is available for other applications and threads. Unless a program is specifically coded to take advantage of multiple cores, it won't, no matter how many cores you have; unlike with SLI/CFC, where a game doesn't have to be made for a multi-GPU setup to realize performance gains. Of course there are exceptions, a game can be SLI/CFX-optimized and gain better performance, or a poorly coded game can actually perform worse in a multi-PU setup.