Xeon benchmarks for 3D rendering?

[redlancer87]

I'm trying to find some benchmarks on various multi-socket, multi-core systems for some rendering machines for our renderfarms (we use 3dsMax). Our IT dept is looking to get us some new systems, and I'm worried they may be spending too much for what they think they're getting. (They're very good at networking, but just not enough of an enthusiast to care about specific processors)

Right now they're looking at some quad - quad systems, probably a 4x X7350 or 4x E7440. While I'm sure these monsters would work amazingly, they run about $16K + for each one. I'm thinking a dual W5580 system would probably be the same speed for rendering (3dsMax is very multithreaded) but would be thousands and thousands less per unit, allowing us to get a lot more total machines.(their thinking is 4 processors is fastest because 4 is more than 2)

I can't find any bechmarks on Tigertons, Dunningtons, or really any of the higher end Xeons comparing an x7350, e7440, or w5580.

Any educated guesses on which would be fastest? I'm thinking the hyperthreading and extra clock speed of a 55XX series would dominate the 73XX and 74XX series Xeons for rendering power.

SJ

 

[Evilsizer]

I'm trying to find some benchmarks on various multi-socket, multi-core systems for some rendering machines for our renderfarms (we use 3dsMax). Our IT dept is looking to get us some new systems, and I'm worried they may be spending too much for what they think they're getting. (They're very good at networking, but just not enough of an enthusiast to care about specific processors)

Right now they're looking at some quad - quad systems, probably a 4x X7350 or 4x E7440. While I'm sure these monsters would work amazingly, they run about $16K + for each one. I'm thinking a dual W5580 system would probably be the same speed for rendering (3dsMax is very multithreaded) but would be thousands and thousands less per unit, allowing us to get a lot more total machines.(their thinking is 4 processors is fastest because 4 is more than 2)

I can't find any bechmarks on Tigertons, Dunningtons, or really any of the higher end Xeons comparing an x7350, e7440, or w5580.

Any educated guesses on which would be fastest? I'm thinking the hyperthreading and extra clock speed of a 55XX series would dominate the 73XX and 74XX series Xeons for rendering power.

SJ

a system with two w5580, will be the fastest. you are talking a difference of core 2 arch vs i7. the latter system will get bottlenecked by the NB's limited fsb speeds. not to mention alot of data will be flowing thru it if you start rendering.

not only will the W5580 system be faster but use less power in some cases while doing work. the advantages of the w5580 system will be it has QPI. this is a interconnect to allow cpus to send data to each other. where as on the other types of cpu's you listed have to go thru the limited NB.

QPI for multi cpus really does make a difference just look at at a single cpu Core 2 quad vs i7 920. in that case of a simple i7 920 run, it does take the intel Extreme quad core running at 3.2ghz to match it. now think that the i7 setup is running at a mere 2.66ghz. there are no benches of the systems you asked about. as the price of the systems is not something a review would be able to cover. well some might if they get intel to sponsor it.

In your case tell your guys they are wrong even if the cpu has more cores. the biggest problem like i covered is the NB on the old FSB based systems. this is and has become a problem for multi-cpu systems with a FSB. as there is no direct comm's to the ram or a fast way to send data to the other cpu. as data from one core on cpu 1 needs to get to cpu 2 for one of the cores. what happens is the data has to go down to the NB then back up to the cpu/core. it is a huge waste but is what intel has used until i7 came out.

to cover clock speed that is only part of it and it would move it self away from the others even more. per clock i7 is at min. 16.7% faster. throw in rendering with the new SSSE instructions used and it moves in some case upwards to 35% faster per clock. these numbers im throwing out are from past bench runs comparing core 2 to i7 arch.

 

[nzaneb]

Not that this is specific for rendering, but check out the Geekbench benchmark results here. They have multi-socket results, and may help you out in your quest.

 

[redlancer87]

Thanks guys. I figured the 55XX series would be the way to go, that nehalem core is just too good for rendering.

Now to convince the corporate IT they can save some money (and get me more machines )

SJ