Xeon versus i7

[jstrob]

I'm trying to figure out how I could build a faster workstation for an upcoming project involving huge PhoenixFD sim and for sure vray rendering. I just realized that regular cpu are now more pricey that they were 3 years ago. When I built my current workstation in 2011, I bought 2 xeon x5690 3,47 Ghz for 1778$ each but now, 3 years later, the exact same cpu are 1925$!!! I am really wondering what's happening in the cpu market!

I also heard that dual xeon are faster for raytracing but slower for fluid simulation that i7. I can't remember where I saw this but I think it was someone from chaosgroup saying that. I would like to know more about this. I can't understand how the most high end dual xeon could be slower than the most high end i7. Most of the time, the latest xeon is way faster than the latest i7 and you can even put 2 in the same machine. for sure they are more pricey but they should be faster too!

[joconnell]

Yep, talk to the guys from www.bottleshipvfx.com on twitter - they tested the lot and came up with your conclusion, a single, high clocked cpu is better for fume, naiad and phoenix as a lot of speed gets lost in sending data across the bus between the two cpus. They settled on high speed i7 machines and then some dual cpu dell machines for render nodes.

[jstrob]

Yes. i know Hristo Velev from BottleshipVFX also wrote that somewhere. But I''m still sceptical... Anyway I ordered a dual xeon 10 cores. it will certainly be useful when rendering. I guess it depends on the config of the whole computer. I would like to know what they compared exactly. If they took account of the price that's for sure true that a i7 for the same price will be faster.

[Neilg]

phoenix needs raw clock speed, 10 cores or dual doesn't make a difference. something about the slowest part of the simulation not being mutithreaded (convection?). i was asking about this a year back
I've got a 3.5ghz i7 that's a few years old and it's faster than a 2.8 ghz xeon in phoenix. If we get another phoenix heavy job i'm going to pitch picking up a 4ghz haswell and overclocking it to 4.5, see if it makes a big difference.
You'll probably not notice much speed difference being that you got 3.5ghz, but they'll help with rendering a lot.

[joconnell]

Yeah it's the same as the issues that naiad has - the bus transferring data between the two cpus is quite a bit of an overhead so that really chokes the benefit you get from the second CPU. As you mentioned it'll help a lot of highly multithreaded bits like normal rendering but for simulation bits as Neil mentioned higher clock speeds will give you more benefit.