• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Render performance systems

Associate
Joined
22 Nov 2010
Posts
10
Hi, Just wondering what recommendations people have for CPU based rendering hardware?........ what today's performance, density value equation looks like?

I assume LGA 2011 v3 LGA is the way to go. Either 1 or 2 processor motherboards is what I have concluded. Peripheral features like card slots etc are not relevant at the moment (I don't think) as GPU rendering is not used at present, however features like RAM connection to processor to influence speed might be relevant....(although is is just a nano second difference?!)... GPU might come into play in future so it is a lesser consideration.

The main question is which CPU to go with, as there are such variety of cores vs GHz available at various wide ranging prices, and I am unsure as to what the stats online at benchmark ( ) mean in real terms?

For instance listed on https://www.cpubenchmark.net/high_end_cpus.html

Here are a few cpu options.

8Core:
i7 5960X : 3.0 - 3.5 GHz : scores 15974 : power 140W : Price £840.
E5 2667 v3 : 3.2 - 3.6 GHz : scores 16274 : power 135 W : Price £2114.
E5 2640 v3 : 2.6 GHz : scores 14072 : power 90W: Price £856.
E5 1680 v3 : 3.2 - 3.8 GHz : scores 17166 : power 140W : Price £1533.

10 Core:
E5 2687W v3 : 3.1 - 3.5 GHz : score 17734 : power 160W : Price £1775.
E5 2650 v3 : 2.3 - 3.0 GHz : score 15230 : power 105W : Price £964.
E5 2660 v3 : 2.6 - 3.3 GHz : score 16194 : power 105W : Price £1188

12 core:
E5 2670 v3 : 2.3 - 3.1 GHz : score 16686 : power 120W : Price £1316.
E5 2697 v2 : 2.7 - 3.5 GHz : score 17469 : power 130W : Price £2327.

14 core:
E5 2697 v3 : 2.6 - 3.6 GHz : score 21964 : power 145W : Price £2214.
E5 2695 v3 : 2.3 - 3.3 GHz : score 20711 : power 120W : Price £1962.

16 core:
E5 2698 v3 : 2.3 - 3.6 GHz : score 22309 : power 135W : Price £4157.

18 core:
E5 2699 v3 : 2.3 - 3.6 GHz : score 20594 : power 145W : Price £4109.

Prices are approx and inc. VAT.

So anything from score 14072 to 22309..........but what difference is there in real terms render time performance between those scores?

Perhaps another way to put it; what performance difference is there between 8 core score 15974 and 16274, and 12 core score 16686 and 17469, and 14 - 16 cores that score 20594 and 22309?
(The 18 core seems to score lower and yet cost more........ also drawing more power, so why would one buy it? Perhaps I have missed something or cpubenchmark source is an error?)

Any advice would be helpful in terms of real world results for rendering on these types of machine. Budgets run from £1500 - £4000, so unless the expense is not as many times as fast then it may not make sense to spend more, although higher density would be a consideration.

Are there any better sources of real world processor render performance?

I included quite a few processors in the list as 1. there is an issue of availability and 2. to illustrate the wide range to consider........ it might be helpful for others also asking themselves the same question.

Cheers
Steve
 
as you've not mentioned what software you will be using
a) are you sure it can support more than 4 cores
b) are you sure it doesn't support gpu ie cuda based rendering
 
Are there any better sources of real world processor render performance?

You should probably know already what scaling you can get out of your software, as well as cache, memory, I/O dependencies, etc.

If you don't know this, the best you can do is go with the highest (clock speed * number of cores) / £.

Since you're only looking at recent-gen Intel chips they all have approx identical architecture, the most relevant differences are the core counts and clock speeds.

(You could also take H.264/265 and cinebench with a pinch of salt for some guidance.)
 
Last edited:
A lot depends on which rendering app you're using, it could make more sense to setup multiple 4core i7s than one huge expensive xeon system. if nothing else this ensures you can quickly get a new main system back up quickly if your work pc fails.
 
Hi, thanks for your advice everyone.

It is just CPU rendering. It's 64 bit based and can use all cores thrown at it, one render project took 10G ram on 6 core single processor system (used the 12 threads), so chucking more ram wouldn't increase speed....... it's the processor number crunching ability that is the crucial factor.

The developer is not taking it to GPU - at least currently. It maybe a background consideration that perhaps a grapchics card could be added on a system in future if it were to go that way, especially as Intel are not increasing speeds but efficiecncy now.

A board considering would be small as possible, such as ASrock x99 4 Extreme, so to get as small as possible machine for the price, not having huge SLI graphic capcity, as opposed to a larger board with SLI. (the risk being if in future the number of ports restricted a future upgrade.) So maybe a MSI X99A RAIDER USB 3.1 Motherboard, would be safer, with larger GPU and RAM capacity. The other option would be the dual CPU boards, less space, but far more expensive currently - and little in the way of reviews apart from dissapointing - although I think people's expectation might be a little high to start with.

joejojo, yes pinch of salt with some quoted stats is why I posted the question - i supppose I was hoping for some definitive "time taken to render a frame by a range of processors"......... and then thus be able to work out the best for price point/density, as you suggest.

So clock speed * number of cores really works for comparison between the processors? I assume this will not work comparing cheaper old processors in the family as newer chips will have a gained speed beyond the GHz rating - or are they comparable too - this would change the price factor a bit (although Intel have not dropped the price of older CPUs much!)

Cheers
Steve
 
So clock speed * number of cores really works for comparison between the processors? I assume this will not work comparing cheaper old processors in the family as newer chips will have a gained speed beyond the GHz rating - or are they comparable too - this would change the price factor a bit (although Intel have not dropped the price of older CPUs much!)

The former. All but one of the Xeon models you mentioned are Haswell (the v3s that is, the v2 is Ivy Bridge) so the only differences between them is the number of cores and clock speeds. You won't find much new for sale older than Haswell.

You're right that when comparing between versions you have to take into account the small improvements between them, but they're small (typically several %).
 
I would compare Cinebench scores for the different CPUs.
Although Cinema 4D's render engine is just one engine, all CPU based ray tracers scale pretty much the same way.
So if one CPU is twice as fast as another at cinebench, it will be twice as fast as it at vray, or Arnold, pretty much.
You should be able to look up cinebench scores for all xeons, just make sure they are all the same version, usually 15 nowadays.
 
Thanks.

Quixote, I've not found it easy to find cinebench 15 scores, those that I do have a "score", but I don't know what this means in real terms.......... I wish they just quoted the time taken to render the frame!!

Any pointers would be appreciated.

Cheers
Steve
 
Quixote, I've not found it easy to find cinebench 15 scores, those that I do have a "score", but I don't know what this means in real terms.......... I wish they just quoted the time taken to render the frame!!

I've asked before and nobody seems to know, other than the score is proportional to the performance in FPS.
 
Ha yeah, so no one knows! Which is what I've found......... it's amazing that in the google age such a seamingly simple thing (to enable one to make an informed choice) is not reported accurately.

FPS is about display I take it, which is not relevant to rendeirng with CPU surely.

Thanks for your input.

If anyone else can shed some light on it, it would be grately appreciated.

Cheers
Steve
 
I know its not a great solution, but I switched over to a renderer that also supported GPU and the gains were immense. Its worth learning something new sometimes for the increase in productivity.
 
Well firstly, you should ignore the Open GL score (This is viewport performance )& the CPU (single) scores if they are quoted.
You are just looking at the multi-CPU score. Which will be a number like 1073 in my case.

The number is simply a measure of how fast the CPU rendered the scene. It doesn't matter what the unit is (it may as well be arbitrary). What matters is that a CPU with a multi-CPU score twice that of another CPU will render twice as fast.
And this scaling will apply pretty much to all CPU-based Ray-Tracing engines. So a CPU with twice the multi-CPU score of another, will render a given VRay or Mental Ray scene approximately twice as fast as well.
It's simply a relative measure of performance.

So to see which is the best value, look at Cinebench points per £ spent.
But don't forget to factor in the cost of the whole machine and any software licenses too.
Sometimes it may seem that a more expensive CPU is worse value for money but if you need to build only 2 machines instead of 3, it can actually end up cheaper.

So in the end, Cinebench points per £ spent on the whole machine + software, is really the number you want to be looking at to determine value for money.

For really big scenes, with lots of assets ie meshes and textures, the time for distribution becomes significant. And here your network & I/O speed can be a factor. In these cases, fewer, faster machines have an advantage, apart from the space-saving and electricity saving aspect.

Hope that helps.
 
Last edited:
Back
Top Bottom