Corona Alpha4 Benchmark scene

[NielsH]

I don't think you should compare just the CPU price and the performance, the price of the CPU is only part of the complete system and thus you should also factor in those costs when you really want to compare.

If you have a system which is 2000€ in total with a 400€ CPU which performs at 100% vs the same system with an 800€ CPU which performs at 140% then sure the cost of the CPU might be 200% and thus not cost effective. But the total cost of the system is only 120% making it a far better deal.

Besides that it'll never be a black and white problem considering that everyone has difference financial means, for one person 10% improvement for 1000€ is completely worth it while for someone else even 100% improvement for the same amount just isn't worth it.

[juang3d]

I understand that, my current 5820k setup costs exactly 807.39 + vat, it's a render node so it does not have a powerful GPU at all.

Anyways, I would be pretty happy having this same setup but instead having a standard HDD I would opt for SSD, a powerful GTX (not a Titan) and that's it, the Motherboard it's an Asus X99-A

So I doubt the improvement in rendering speed could be a lot greater, the same render node cost with the 5960x is 1319.78€, so 512€ more, so you get in the same machine 32% more performance for 63% more price (based on the benchmark results we saw here).

With that said I'll shut up already XD I just want to make numbers clear for people that want to acquire a machine, maybe someone could prefer more threads, more power even for more money because they can't afford a farm, or they don't want nodes, or because they need more workstation horsepower, and may be people that prefer to have two nodes working side by side and have a bit less workstation horsepower but a lot more render performance, it's up to everyone :D

In any case, no doubt that the new Haswell family it's outstanding!

Cheers!

[Jann]

Btw, the SHD's post basically shows nonetheless that 5960X is fantastic performer.

I'll add a bit to this fact.
We opened the bench scene in 3dsMax 2014 and Corona 7.1, and limited to 100 passes. (do mind possible variations from motherboard, ram, etc)
i7-4790 - 6.44 minutes
i7-3930k stock - 6.00 minutes
i7-3970x @4.2GHz - 5.18 minutes
two Xeon E5-2660 ES - 3.09 minutes
i7-5960x @4.4GHz - 3.10 minutes
two Xeon E5-2697v2 ES - 2.10 minutes

So yeah, apart from DDR4 price, the 5960x with good cooling is a fantastic cpu.

[juang3d]

The price is not so bad, 80€ for each 8Gb of Crucial Memory 2133 CL15, not the awesomest ram in the world, but it works perfectly, over the years I found that the difference made by RAM and Motherboard, when all they are from a medium-high family, is residual.

I'll try the same with the 5820k and see what time do it give me when I finish my current project.

If you have daily builds access, can you try the same with the 5960X but with the 19th October build?

Cheers!

[krimson]

Corona Renderer Alpha 4 benchmark scene
 Living room 100 passes
Intel(R) Core(TM) i7-5960X CPU @ 3.00GHz
Time: 0:2:17, Rays/s: 9,209,435

oc'd to 4,3 ghz

[Juraj]

Corona Renderer Alpha 4 benchmark scene
 Living room 100 passes
Intel(R) Core(TM) i7-5960X CPU @ 3.00GHz
Time: 0:2:17, Rays/s: 9,209,435

oc'd to 4,3 ghz

Krimson, what's your cooler ? NH-D15/CL/Custom loop ? Can you keep it 4,3+ GHz stable 24/7 rendering ? Thank you

[juang3d]

Juraj, apart from the answer you can receive from Krimson, I know you did not asked me, but I would like to give this info, keeping the 5960X 24/7 rendering at 4.3 should be easy, I'm rendering with the 5820K since I assembled it and it is at 4.4, it doesn't go over 45/46 celsius and it is completely stable.

The cooler I use is a cheap closed water cooling system, but it works perfectly, the "Cooler Master Seidon 120V", in my store just 33.88 + vat, I use this for all my nodes and workstations, and the temperature is kept very low.

http://www.pccomponentes.com/cooler_master_seidon_120v.html

I'm sure the 5960x should be as efficient as the 5820k in temperature if not more, so 4.3 should be easy for that CPU.

Hope this gives you some info.

Cheers!

[krimson]

Hello Juraj,

A noctua NH-D15.  You recommended it to me on cgarchitect.  :D
It runs fine on 4,3 (all cores) and the hottest it gets until now is 60°C.  The longest time it rendered until now is about 1 hour.
Also i only have intake (2x140) in front of the case and extraction on the back (1x140mm) + 1x140mm intake on the bottom.
The case is very quiet.

The rest of the components of workstation is as follows:
mobo: asus x99-e ws
mem: ddr4 crucial ballistix sport 32gb
nvidia gtx 980
2 ssd's samsung pro
case fractal design define xl
cpu passmark is 20403 and cinebench 15 cpu is 1673.

By the end of this week my renderslave (dual xeon e5-2640 v3 with asus mobo + 64 gb) will arrive.  Will post the benchmarks aswell.

I will keep you informed regarding stability.  I believe it also has got to do with the asus motherboard which has better temperaturemanagement then other motherboards.

kind regards.

[Juraj]

Oh :- D Sorry, don't keep much track of where and what I say :- ). Congrats to workstation ! Impressive machine.

Well, nice to hear, I thought NH-D15 would hit some ceiling around 4.3-4.4Ghz with the temperatures. Seeing as the average with custom water loop is now 4.6Ghz (average for past 20 units sold by overclockers UK),
I think your temperatures quite nice. I will begin building one probably next week, just debating whether to go for full custom loop or stay with Noctua classic.

I presume you don't plan to use 4-way SLI (not even physically possible I guess with Noctua), so what was the reason for choosing WS compared to simply spending such money (or less) on Deluxe or ROG ? {but it is the nicest out of them aesthetically, that has to be given :- ) beautiful board

I see you also went with quite low Xeons, what's the all-core turbo frequency on those ?

[krimson]

Hello Juraj,

As for the xeons i know...  First i was planning on buying 2680 v2's.  But then for budget reasons i had to go to 2650 v2's.  With the v3 range coming out the xeon e5-26xx v3 range has shifted and the 2640 v3 has the same (or even better) cpu passmarks then 2650 v2, and is even a bit cheaper and is more "future proof cpu".

- The 2640 v3 has 8 cores and a clockbase of 2.6 ghz turbo 3.4 ghz. (so it's shifted up compared to v2 range)
- As for the motherboard: Asus states that it uses more endurable components in its ws boards, that was the only reason.  It's not that much more expensive (only 50 euros more then deluxe)
- As for the 4-way sli: I'm not planning on getting more gtx's.  But i did put a psu of 1200w in just in case i would put one in extra.  You never know if renderers will combine gpu and cpu in the future.

Good luck on building you new workstation.

kind regards,
Wim Clissen

[Juraj]

No, that's One-core maximum turbo-frequency, the all-core I think is 2.8, that's a big difference. But the Asus board gives you option to try 5-10perc. of overclock on them, so maybe, you could potentially get up to 3.0 Ghz during rendering.

[krimson]

hello again Juraj,

Do you think 0.2 ghz will make allot of difference.
I'm thinking about product garanty on the xeon cpu.
Anyway, the passmark is a bit higher then the old 2650 v2's, the cores and clockspeeds are the same, and it's a bit less expensive.

kind regards

[tomislavn]

hello again Juraj,

Do you think 0.2 ghz will make allot of difference.
I'm thinking about product garanty on the xeon cpu.
Anyway, the passmark is a bit higher then the old 2650 v2's, the cores and clockspeeds are the same, and it's a bit less expensive.

kind regards

From my personal experience, 0.2Ghz higher "All core" makes for around 3-5% decrease in render time - on average. This was on last generation CPUs though.

[Juraj]

You're not breaking any warranty, the Asus overclock will probably only readjust BLCK strap or some other trick because the multiplier is locked. Even with such modest overclock (it depends on how lucky you are with the pieces you get), you're not really rising the overall TDP much, it's more than fine.
Yes it does make difference,  can be another 4-7perc. free performance. (obviously the percentage depend on relative increase, for i7 at 3.5 to 3.7 is nothing, but at 2.6 to 2.8 it's more noticeable. )

[krimson]

ok thanks for the info juraj
you've been a great help to configure my workstation and renderslave.

kind regards
wim clissen