CPU performance discussion...

[Njen]

Does Corona see more benefit from more cores, or faster threads? For example, a dual Intel Xeon E5-2630 v4 setup has 40 threads at 2.2 GHz for a theoretical maximum of 88GHz, while the Intel i9-7940X has 28 threads at 3.1 GHz for a theoretical maximum if 86.8 GHz. Attempting to put aside the fact that one is a Xeon, and the other is a 'prosumer' CPU, which of the two would have an advantage when rendering with Corona?

[PROH]

Hi. The simple answer is: both! Not sure your maximum GHz is calculated correct, but according to the Corona benchmarks for the mentioned CPUs, the i9 seems to be faster than the 2 x Xeon.

Besides rendering, 3dsMax still has a lot of single threaded functions that benefits from a faster CPU.

[Juraj]

You also have to account for AVX turbo offset that Intel architecture has in Broadwell/Haswell/Skylake. The v3/v4 Xeons have it fixed but for i9 you can set it yourself in UEFI with sufficient cooling. This can lower the performance of Xeons a bit compared to i9 in Corona as they will run at lower clock.
In short renders, precomputation will account for more significiant part of total rendering time than in long renders, and in that case, faster (in clocks) CPU will sometimes go through all the parts faster, the difference is not big but it's there. In my experience the various precomp parts vary strongly in their usage of cores.

Outside of that, the difference will be very minor.

The Xeons are lot more efficient for nodes, my 2698v4 nodes (x2 setup) have 4500 Cinebench score with 270W total consumption. My newest i9 7980XE overclocked to 4.6 Ghz (only temporarily) reached same Cinebench score with almost 700W :- ).
I didn't bother to run Corona benchmark.

[nikoLAB]

Hi to all,

My toughs are about buying Amd Treadripper 1950x.
How Corona works with that Amd processor? And do i have some better deal for about 2800E as I presume that would be a price for the treadripper configuration? 
I'm asking because I don't know does corona have some instructions that they don't work with Amd processors.
   

[lenogre]

http://www.comptoir-hardware.com/articles/cpu-mobo-ram/34813-test-amd-x399-threadripper-1950x.html?start=12

[PROH]

@nikoLAB: Search for Threadripper on Corona benchmark site, and see the result from other users. I think it's working very well with Corona :)

[nikoLAB]

Thank you so much for a prompt answer!
So, as I think, probably a best buy right now.

[cecofuli]

Speaking about WATT, this is my experience:

UPS: 25Watt
Monitor 27 inch: 80 Watt
PC in normal usage = 185 Watt (I7 970 six cores,  with a little OC, 4 HHD, 1SSD, 24GB RAM)
PC in rendering = 280W
Average = 320 W total

So, in 8 hours of work, my PC cost me: 320W * 8 hours = 2.5KWh => 2.5 KWh * 0.25 euro = 0.625 euro/day
Speaking about only PC=                     280W * 8 hours = 2.25KWh => 2.25 KWh * 0.25 euro = 0.562 euro/day

Of course, CoronaBench= 4:30mi => 4.5
So, my efficiency is  4.5/0.562 = 8
It will be interesting so know your efficiency

In the picture: UPS+Monitor+PC in rendering status (100%load)

[alexyork]

You also have to account for AVX turbo offset that Intel architecture has in Broadwell/Haswell/Skylake. The v3/v4 Xeons have it fixed but for i9 you can set it yourself in UEFI with sufficient cooling. This can lower the performance of Xeons a bit compared to i9 in Corona as they will run at lower clock.
In short renders, precomputation will account for more significiant part of total rendering time than in long renders, and in that case, faster (in clocks) CPU will sometimes go through all the parts faster, the difference is not big but it's there. In my experience the various precomp parts vary strongly in their usage of cores.

Outside of that, the difference will be very minor.

The Xeons are lot more efficient for nodes, my 2698v4 nodes (x2 setup) have 4500 Cinebench score with 270W total consumption. My newest i9 7980XE overclocked to 4.6 Ghz (only temporarily) reached same Cinebench score with almost 700W :- ).
I didn't bother to run Corona benchmark.

Juraj,

How are you finding that overclocked i9 up against your xeon setup, in terms of speed/reliability etc.? Bang-for-buck it looks pretty compelling against the xeons. I'm ignoring Wattage energy cost because that's not relevant to us (fortunately).

[Juraj]

I am having second delid process done, I am not satisfied with the temperatures. The chip is good apparently, stable low voltage, but to keep it above 40x multiplier, some cores reach 100C. Of course, this is also when the chip consumes 450W alone :- )
This second delid will feature direct die replacement for IHS cover (risky and quite nasty stuff, like old Athlon processor for those who remember how coolers used to be mounted on :- ), a toy from german overclocker Derba8er and probably custom loop ( EK system ). Imho this is a lot of effort...

I've said it in the other thread, but I am little bit disillusioned. I might suggest to stay with dual-xeons in the end. The single-thread performance benefit is absolutely non-detectable and the multi-threaded score still sort of pales in comparison to top tier dual-xeons models.

The Wattage does matter, not because of the electricity cost, but because you need to cool it off, and that's a lot of heat. It's not possible to do it silently (water loops aren't silent contrary to mythical public opinion).

[alexyork]

I am having second delid process done, I am not satisfied with the temperatures. The chip is good apparently, stable low voltage, but to keep it above 40x multiplier, some cores reach 100C. Of course, this is also when the chip consumes 450W alone :- )
This second delid will feature direct die replacement for IHS cover (risky and quite nasty stuff, like old Athlon processor for those who remember how coolers used to be mounted on :- ), a toy from german overclocker Derba8er and probably custom loop ( EK system ). Imho this is a lot of effort...

I've said it in the other thread, but I am little bit disillusioned. I might suggest to stay with dual-xeons in the end. The single-thread performance benefit is absolutely non-detectable and the multi-threaded score still sort of pales in comparison to top tier dual-xeons models.

The Wattage does matter, not because of the electricity cost, but because you need to cool it off, and that's a lot of heat. It's not possible to do it silently (water loops aren't silent contrary to mythical public opinion).

Yeh these are my fears.. but our supplier claims max 55C CPU temp at 4.6 full load... so... that's pretty damned good. Specially chosen binned chip, of course. But we'd be guaranteed that as part of the order... so it's kind of interesting.

How do you mean about the single core performance not being worth it? It's 4.5ghz V around 2.6Ghz for those top-end (realistic $$$) Xeons. That's more or less double the performance, so things like general PC use/displacement etc. should be way better no? Is that not your finding?

[Juraj]

Well, to each point:

1) It's not 4.5 vs 2.6. Almost every (Non-ES) Xeon has Turbo clock of 3.6, some OEM up to 3.8 Ghz for 1-4 cores. So it's 4.5 vs 3.6. That's barely noticeable.
    Also, where do you notice the benefit of faster single-thread ? In software responsiveness? Not at all, if software hangs or feel sluggish, it's because it's sluggish software to begin with.
    Photoshop takes up to a minute to save large file, despite SSD being able to transfer those packets under a second. Majority of workstation workflow won't benefit in any noticeable way.
    3.6 vs 4.5 is placebo, but I'll let everyone believe what they want.

2.) Prebinned chips are expensive (does SiliconLottery even sell the top-tier right now ?), I got my chip for 1370 euros, otherwise I wouldn't bother. But still, I doubt 55C at 4.6. Possible but seem lot better then pretty much every delidded review I've read, which had it at 70-80C under 360mm radiator.

[Njen]

So you're saying a top of the line i9 doesn't really offer that much benefit over a comparable Xeon?

I currently have a dual Xeon E5-2630 v4 setup with 40 threads at 2.2 GHz, and am now starting to spec out a new computer to render on and was considering a water cooled i9 7960x, because if I go on the numbers alone on the Corona benchmark site, that gives me roughly a 25% - 35% decrease in render time. But then again, I am not taking energy efficiency into account.

[Juraj]

It doesn't. It does holds its weight if you're willing to do some or all of following: Selected binned chips (From e-shops like SiliconLottery,etc..), delid&relid, delid&direct-die, and solid cooling.

But the single-core benefit is massively overstated (outside of computer games). 3dsMax, Adobe apps,etc.. don't benefit much at all from this.. Stuff will not feel more responsive. Single-core modifiers will instead only take 8 vs 10 second. But the stuff that already takes under second just feel like it could have lower latency, that won't improve. It won't be "snappier". My old i7 2600 from 7 years ago had the same snappiness like all the following workstations. Clean Windows installation will help you more than that..

And by comparison to Xeons, I do mostly mean the superstar chips like 2696v4,etc. , ie: 5000 Cinebench multithreaded score and more.

[alexyork]

It doesn't. It does holds its weight if you're willing to do some or all of following: Selected binned chips (From e-shops like SiliconLottery,etc..), delid&relid, delid&direct-die, and solid cooling.

But the single-core benefit is massively overstated (outside of computer games). 3dsMax, Adobe apps,etc.. don't benefit much at all from this.. Stuff will not feel more responsive. Single-core modifiers will instead only take 8 vs 10 second. But the stuff that already takes under second just feel like it could have lower latency, that won't improve. It won't be "snappier". My old i7 2600 from 7 years ago had the same snappiness like all the following workstations. Clean Windows installation will help you more than that..

And by comparison to Xeons, I do mostly mean the superstar chips like 2696v4,etc. , ie: 5000 Cinebench multithreaded score and more.

But if you're saying that is that there is either a marginal single-core benefit, or none at all, then what you're getting is the same performance as a dual xeon system for less than half the cost. So pound-for-pound that's a huge benefit.