Intel Xeon E5-2670 vs i7-3920XM

[mmr]

Which will render faster on vray? xeon e5-2670 or core i7-3920XM extreme.. which is recommended and why? a double xeon e5-2670 is worth it? Will it truly double rendering speed?

[Dimitris Tolios]

You are compairing apples with oranges, or more likely apples and strawberries...

The Xeon is a 115W workstation CPU, and the 3920XM a 55W laptop CPU of (very) roughly the same architecture - so there is no doubt that the mobile chip will be lacking in absolute VRay performance. In theory a lot.

 

Both chips are very expensive as those are nearly top of the line in their class - value for money in the top vs. the second or third from top of their line PC component products is almost always much worse.

 

For example the i7-3920XM is virtually the same CPU as the i7-3840QM. The XM offers same 4C/8T design, 100MHz higher base speed, same turbo frequencies and cache but the XM has the unique feature in its class of unlocked multiplier for overclocking - a feature that you will be pushed hard to be able to utilize in laptops. List price for the XM is twice that of the fastest QM. IMHO not worth it. Shave another 100MHz from the base speed and you get the i7-3820QM for nearly 1/3 the price of the top of the line XM, and not even 5% performance penalty (vs. the XM, not the slightly faster 3940). The QMs are 45W CPUs tho, and 10W (22% decrease) are much more important than the situational 5% performance advantage when we are talking mobile chips.

 

Now for the Xeon...given that it has to be paired with an expensive mobo, and that it doesn't clock as high as more aggressive 6C/12T i7s like the 3930K and the 3960X, a single E5-2670 will rarely have an advantage in a single CPU workstation. Most design/gfx applications are not properly threaded to really utilize more than one core/thread while doing anything but rendering/compiling the final image/video. That leaves the Xeons with slightly slower per core performance lagging behind the top i7s. That's for factory frequencies, as the i7s can overclock quite well. It is not hard to achieve 4.5-4.7GHz on a 3930K, which ofc will help it not only beat a 8C/16T in single threaded, but also multithreaded applications. No, 8C/16T Xeons are nearly impossible to O/C.

 

Adding a second Xeon for at 2P - 16C/32T machine won't change that. It will nearly double rendering performance, yes, but depending on the % of time you spend actually rendering a project, it might or might not pay off:

 

E.g.: say you have a complex exterior model that you need to render 2-3 stills out off. You have your material library tuned, lighting is not that complex i.e. few trials are needed. This scenario won't let the 2P system shine (enough)...while you are modeling / working on it, you will have like 3-7% CPU utilization (1-2 threads)...if modeling is a week a rendering 2 days on a 1P system, the 2P won't really be a game changer.

 

Say now it is the same project, but you will be rendering small walkthrough movies through it. That could be same 7 days of modeling and a couple of weeks for accumulating video footage for you to edit etc. Then the 2P really shows its teeth: the more % of rendering your project requires, the more cores you have @ your disposal the better.

 

It is not that a you will be disappointed by a 1P or 2P Xeon in absolute performance, under light threaded apps, but definately the relative performance in comparison with a 3930K or even 3770K system that will dent your wallet under $900 or $500 in comparison for CPU+Mobo respectively, is underwhelming in anything but the action of rendering itself. So for spending $2000 or even almost $4000 for the same components on a Xeon workstation, you have to be certain that you will see returns in productivity for your particular workflow / demands.

Edited October 22, 2012 by dtolios

[mmr]

i took the CPUs name from a benchmark chart.. so yes i m sorry it was the i7 3960x but again the 3930k gives approximately same performance for much lower price... as for xeon e5 2670 again too expensive compared to xeon e5 1660 is it woth it? anyway point is speeding rendering.. so double cpu is considered.

[Valtiel]

The first question we might be asking though, as you might derive from Dimitris' response is: What will the box be doing? Are you building a workstation or a render node? If you're building a render node, a pair of Xeon 26xx will give some pretty hefty performance in pure rendering jobs. Honestly, the way they're priced, the 2620's are probably what I'd recommend for price/performance as they begin to lose the pricing advantage of 2 cpu's per box as you go up the line.

 

In a workstation environment where you want a great deal of human-time, single threaded performance in your applications the 3770K has the crown these days. The 3960X will perform better in rendering but, considering the price, it's hard to recommend anything with the "X" suffix as it doesn't offer much of a performance gain over the next-in-line "___K" models. The 3930K is a decent choice if you have a bigger budget and plan to do all of your rendering on your workstation but mind that it will make your motherboard and memory choices more expensive as well, being socket 2011.

 

The short of my response:

Is this your workstation that will be supported by render slaves/nodes? 3770K

Is this your workstation and only render machine? Then 3930K

Is this a render node? Then 2x E5 2620's, 2630's if you REALLY want the clockspeed and have the budget.

[mmr]

Thank you both for the help.

[stevenarnold]

Hello Friends,

 

Chipmaker Intel will take the wraps off its Intel Sandy Bridge ES Xeons later today, so here we look at the difference between the firm's ultra high-end Core i7-3960X and a pair of Xeon E5-2670 processors. The Xeon E5 2670 we tested has a TDP of 115W, the top end of Intel's Xeon E5 power usage range. Although the Core i7-3960X has a TDP of 130W, it is used in single processor configurations. For the purpose of this comparison we will ignore the overclocking ability of the Core i7-3960X despite its unlocked multiplier, because we simply cannot say with any statistical accuracy whether the overclockability of one chip is representative of a batch.

 

Best Regards

Steven Arnold