Dual E5 or i7

[Sekhar_R]

Hi,

I am planning to build a new workstation. Since Corona has changed the game play, I am more inclined on CPU than GPU(due to its limitations and cost). so the question for the Onda and Keymaster.

Which would be ideal CPU for Corona setup

Dual E5 - 2680 V3 2.5ghz or i7 5960X 3ghz

I am also going to add 2 GTX 980 and 64gb ram.

thanks

[M.S.Referee]

No doubt, Xeon CPU is still the best choice. If you got budget for Xeon, just take them!!

[pokoy]

Have a look in the benchmark thread here and compare Corona's results with official benchmarks (for example whether a speed ratio between two certain CPUs is comparable to the ratio from the same CPUs from a dedicated benchmark site). I found that fore some reason dual Xeon setups do not scale linearly and that a simple GHZ vs time rule is not reliable. A single high end i7 might be a better investment in the end when compared to a super expensive dual Xeon rig.

[Juraj]

Hope some mod will take this to HW section.

This choice depends purely on your needs and expectations and of course...budget.

Good workstation with 5960X will come you at around 3000 euros including high-end GPU.
Good workstation with 2p 2680v3 will come you at around 6000 if you include high-end GPU.

5960X offers higher overclock for both single and all cores, up to about 4.4 Ghz per average chip under best cooling. This means you can reach about 40perc. more performance for both single and multi-threaded tasks.

2680v3 offers only mild turbocore feature and about 4-5perc. BLK overclock if you use Asus workstation motherboards. That in reality gives you 2.9-3.0 Ghz for multithreaded rendering, and 3.3-3.4 for singlethreaded workstation tasks. More than enough.

That concludes to fact that 5960X makes slightly better workstation horse, but 3.3+ Ghz for Haswell chip is well beyond any penalization that you could even notice compared to running super-clocked i7.
In multithreaded tasks, 5960X is about 80-90perc. as powerful as single 2680v3 in default turbocore, and roughly 20perc. faster when super-clocked to maximum potential.

To take out: High end dual Xeon (2680V3 + and higher) will be twice faster than 5960X, and cost twice more. That makes it very easy choice...it depends how much money you're willing to spend.

And:

Have a look in the benchmark thread here and compare Corona's results with official benchmarks (for example whether a speed ratio between two certain CPUs is comparable to the ratio from the same CPUs from a dedicated benchmark site).

No. The benchmark is faulty, and based on old Corona core. It basically ignores the performance of dual CPU setup.

I found that fore some reason dual Xeon setups do not scale linearly and that a simple GHZ

Than you found wrong. I run quite few dual-xeons...and yes, they scale almost linearly. Until you reach the too many cores of top models, and then certain applications including Corona have problems utilizing all the cores, but this is software issue and not HW.

[pokoy]

Than you found wrong. I run quite few dual-xeons...and yes, they scale almost linearly. Until you reach the too many cores of top models, and then certain applications including Corona have problems utilizing all the cores, but this is software issue and not HW.

Comparing my old dual Xeon PC vs the new one where speed ration on the paper is almost 1:2, Corona's speed up was not as good, it's somewhere around 1:1,7 - not really linear. Other renderers and programs where indeed linear when compared on these machines. What's odd is that the ratio was 1:2,5 with old daily builds and decreased continuously. Mind you, I have not checked in the last months so the numbers I state here may be absolutely wrong for the final 1.0 release. I guess you were more thorough then so stick to Juraj's research, OP.

[Juraj]

At the moment I am using latest daily build before 1.00, because I need to finish current project next week and then I'll install 1.0. But I don't expect the performance scaling to differ:

Without overclocking 4930k which we use most as workstation in my office, the performance of 2680v2 is exactly tripple. It's very easy to measure by amount of Ray/s and it scales very good.

Now as few people reported, once you go over 40 cores, there are issues which can currently be solved (for both Vray and Corona) by running multiple spawners, i.e distributed rendering at same machine.
Not very comfortable solution...but it is mostly by Autodesk's fault which ignores second cores group and each group is around 40 if I remember correctly.

The scaling you report seems only slightly off (85perc.), I don't think that is what really makes noticeable difference.

It all comes to budget and comfort. If you want the best for you buck, yes, i7 is better deal. But it's not super big difference as some people still think (there is no such thing as "super-expensive" Xeon, remember, 64GB of DDR, PSU,etc.. also cost something, and you only buy it once, and that doesn't include software licences, etc..), it can actually be very close.

[Sekhar_R]

Thanks a ton for all you Masters and Juraj specially for the detailed explanation. I am on my 7years old Asus laptop (i7 2720QM) with GTX 460M + an old Dell E5450 3ghz- It still manages to get some decent render times compared to Vray 3. You can understand my agony. At work place, they have i7 4930K (not overclocked), just factory setup. How the heck, you overclock. may be I need to check some YouTube research. Never thought about that.

I am saving for a new workstation for quite some time now. My theory of comparing these two CPUs are based on the CPU Benchmark,which is More cores = Fast Corona Render. Considering 48cores compared to 16cores. Budget is still a concern. Now learning from the discussion, not really sure, if Corona can actually make use of all 48 cores.

Thank you guys for the info. May be by the time I am ready to spend on a workstation, new updates will occur. My current setup is complete obsolete now.

[PROH]

Hi Sekhar_R. AFAIK the limit is 64 cores. If you got more cores than that, you'll need to split them into to groups, and use them as if it was 2 machines. The Scanline renderer however is limited to 32 cores.