Export Material and Updating Time - Hardware relationship

[GIOLETS]

My question is related to the phases :
- Export materials
- mesh creations
- Updating time
that occurs every time the "send scene / restart render" is pressed.

I wish to learn how PC hardware influence the "export material time" and "updating time" phases.
During rendering test, specially having very complex scene, these take a lot of time everyday, so I think is important (almost as the rendering time) to know what PC parts is possible to upgrade in order to decrease this time.

I have made some test on various machine of my office, using always the same scene and, I'm pretty sure that the export time is not related to the CPU type, the GPU type and also if the card is directly mounted on MB or connected with a riser.
Some data:

Machine 1 (i7-975 4 cores 3.6GHz, 32GB DDR3, 2xGTX1070 + 2XGTX 980ti + 1XGTX 780, SSD, Asus A6T7 WS supercomputer, PCI 2.0, all GPU with riser cables):
export material time 39 sec
mesh creation 5 sec
Updating time 20 sec
total time 65 sec

Machine 2 (i7-4790 4 cores 4 GHz, 32GB DDR3, 2xGTX 780 1x TITAN X, SSD, AsRock Z97 Extreme 9, PCI 2.0, All GPU on MB):
export material time 30 sec
mesh creation 3.5 sec
Updating time 20 sec
total time 54 sec

Machine 3 (i7-2600 4 cores 3.8GHz, 32GB DDR3, 2xGTX 1070, SSD, Asus Sabertooth P67, PCI 2.0, All GPU with riser cables):
export material time 33 sec
mesh creation 4 sec
Updating time 17 sec
total time 54 sec

Machine 4 (i7-5820 6 cores 3.6GHz, 32GB DDR4, 2xGTX 980ti, SSD, Asus X99 E WS,PCI 3.0, All GPU on MB):
export material time 26 sec
mesh creation 3.5 sec
Updating time 16 sec
total time 46 sec

Machine 5 (i7-4930k 6 cores 4.2GHz, 32GB DDR3, GTX 770 + 780ti, SSD, Asus P9 ,PCI 3.0, 1 GPU on MB 1 on riser cable):
export material time 32 sec
mesh creation 2 sec
Updating time 13 sec
total time 47 sec


The Machine 1 have a 6 years old MB, the Machine 4 have less than 1 year, so my attention is focused on the MB or the RAM , but perhaps someone can better teach me about the hardware parts influenced by these process.


Thanks a lot.

[GIOLETS]

I've found this 12/2016 topic (also without answer) that asked the same things.

viewtopic.php?f=9&t=58318

The export material and updating phase takes really a relevant quantity of time every day, for heavvy scenes could take 2-3 minutes ... so it's really important I think to understand which hardware parts are involved, expecially in case of PC upgrade.

[Iceman9]

I would be like to see some guidance on this from Otoy.

[GIOLETS]

Some useful tips here, from Cinema4d plugin forum (thanks Aoktar):

viewtopic.php?f=30&t=59614

But again I think Otoy should give us more and better focused info on this...

[glimpse]

Hi, Guys.

I'm not a dev & might not have deep insight of how things work by design, but I can easily give You some ideas about performance & where to look for planing new build.

Based on previous posts with similar questions, some tests comunity already made & some personal observation I could say few things:

* don't look for more than 4cores right now if You're not planing yo work on other tasks while rendering. Take the fastest last gen quad core, preferably unlocked, add good AIO or big chunk of metal & OC a bit. Why? It seems that OctaneRender does not take advantage of more than 4 cores, so their speed is more important than having extra two or even twice but running them slower.

* RAM does matter, since Your image while it's rendering sits in it & from new (v3) version it seems that CPU has more to do during rendering. HOWEVER, paying twice more for 10-20% faster RAM modukes makes little to no sense, since their influence is not that big. Stick with value pieces from good brand & don't risk stability for questionable soeed gains.

* risers do not matter(if they are good quality) what matter is their speed. Avoid using cheaper x1, x4 cables since new version feel their impact more than older.

If You take all this into perspective..& go through Your results, I'm pretty sure You'll see evidense of that reflected. Put CPUs based on their architecture in a line, discard extra cores & take a look at their operating frequencies. Keep in mind that jums from one to other architecture lead different gains, pluss add a fact that multi core & single core boost differs (as some of workloads do not require all four cores this matters as well).

Hope that helps. Feel free to leave more questions, notes & I'll try to reply to those. Again, a lot of this was from observation & discussions here in the past, so I'm not claiming this is 100% accurate, but it explains a lot of things & gives a possibility to predict performance of system for stages You mentioned.

[GIOLETS]

Thanks for your really important reply Glimpse.

following your note I've just made a simple test overclocking a bit my Machine 1, so here the results:


Machine 1at std 3.3GHz (i7-975 4 cores 3.6GHz, 32GB DDR3, 2xGTX1070 + 2XGTX 980ti + 1XGTX 780, SSD, Asus A6T7 WS supercomputer, PCI 2.0, all GPU with riser cables):
export material time 39 sec
mesh creation 5 sec
Updating time 20 sec
total time 65 sec

Machine 1OC at 3.9GHz (i7-975 4 cores 3.6GHz, 32GB DDR3, 2xGTX1070 + 2XGTX 980ti + 1XGTX 780, SSD, Asus A6T7 WS supercomputer, PCI 2.0, all GPU with riser cables):
export material time 34 sec
mesh creation 5 sec
Updating time 17 sec
total time 56 sec

So at first sight, with a clock encrement of about 17% we have almost the same gain of speed (+14%).
Really interesting and useful, thanks a lot for your help Glimpse!

[glimpse]

Thanks for your really important reply Glimpse.

following your note I've just made a simple test overclocking a bit my Machine 1, so here the results:

...

So at first sight, with a clock encrement of about 17% we have almost the same gain of speed (+14%).
Really interesting and useful, thanks a lot for your help Glimpse!

yes that's right, when You increase speed of Your CPU, all of those things You mentioned (apart rendering itself) will be faster. But it's worth to know that You from OC You do not get the same % increase in performance - that's just the way is is with any benchmark

again, to sum up: 4 cores, new architecture, maybe a bit of OC for additional stretch - all You have to look for good performance.

hope that helps ;)

Sinc.
tom