[T0M0]

Today I did some testing with one of my previous works comparing V4 (4.0.0.12 - released 16.12.2016) and V3 (3.2.1.5 - released 13.7.2016),

Tested on configuration:

- Windows 7 x64 SP1
- dual Xeon E5-2680V2 (40 logical cores)
- ASUS Z9PE-D8 WS with 64GB RAM

Unfortunately I can't post the preview here, but the scene contains things like displacement, SSS materials, fur, ThinLens camera, ExtraSampling with Custom alpha, MultiLight enabled in Intensity mode and it was lit by emitters.
Just one regular scenes of my work.
Scene was rendered to SL 10 then parts selected by Custom Alpha were rendered using Extra Sampling to SL 13.


Commands used:
-nomxi (so the rendering process won't be interrupted by writing MXI file on disc)
-cpuid:100 (same "starting condition", so initial seed should be same)


Let's take a closer look on debug console (I won't paste here complete log just some informations):

Memory used before preprocessing [V3 vs V4]:

Memory used before preprocessing: 1394 Mb

- vs -

Memory used before preprocessing: 1426 Mb

Memory used after preprocessing [V3 vs V4]:

Memory used after preprocessing: 6472 Mb

- vs -

Memory used after preprocessing: 6505 Mb

Voxelization [V3]: voxelization time: 2m 35s

[02/February/2017 10:12:10] Start Voxelization
[02/February/2017 10:14:44] Maximum Memory used during voxelization: 10653 Mb
[02/February/2017 10:14:44] End Voxelization
[02/February/2017 10:14:44] Voxelization done.
[02/February/2017 10:14:44] Memory used after voxelization: 10513 Mb
[02/February/2017 10:14:45] Start Rendering

Voxelization [V4]: voxelization time: 2m 38s

[02/February/2017 12:38:29] Start Voxelization
[02/February/2017 12:41:07] End Voxelization
[02/February/2017 12:41:07] Maximum Memory used during voxelization: 10809 Mb
[02/February/2017 12:41:07] Voxelization done.
[02/February/2017 12:41:07] Memory used after voxelization: 10669 Mb
[02/February/2017 12:41:07] Start Rendering

Rendering process [V3]:

[02/February/2017 10:14:57] Extra Sampling: New (estimated) global target SL: 12.64
[02/February/2017 10:14:57] Benchmark of 338.000. Time: 12s. SL of 1.00
[02/February/2017 10:14:57] Time left: 1h09m41s
[02/February/2017 10:15:18] Extra Sampling: New (estimated) global target SL: 12.64
[02/February/2017 10:15:18] Benchmark of 315.312. Time: 33s. SL of 2.00
[02/February/2017 10:15:18] Time left: 1h18m50s
[02/February/2017 10:15:49] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:15:49] Benchmark of 311.144. Time: 1m04s. SL of 3.00
[02/February/2017 10:15:49] Time left: 1h17m48s
[02/February/2017 10:16:35] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:16:35] Benchmark of 310.669. Time: 1m50s. SL of 4.00
[02/February/2017 10:16:35] Time left: 1h12m48s
[02/February/2017 10:17:43] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:17:43] Benchmark of 312.145. Time: 2m58s. SL of 5.00
[02/February/2017 10:17:43] Time left: 1h14m45s
[02/February/2017 10:19:24] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:19:24] Benchmark of 314.137. Time: 4m39s. SL of 6.00
[02/February/2017 10:19:24] Time left: 1h16m48s
[02/February/2017 10:21:55] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:21:55] Benchmark of 316.477. Time: 7m09s. SL of 7.00
[02/February/2017 10:21:55] Time left: 1h08m18s
[02/February/2017 10:25:41] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:25:41] Benchmark of 317.007. Time: 10m56s. SL of 8.00
[02/February/2017 10:25:41] Time left: 1h10m38s
[02/February/2017 10:31:26] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:31:26] Benchmark of 315.923. Time: 16m41s. SL of 9.00
[02/February/2017 10:31:26] Time left: 1h03m30s
[02/February/2017 10:39:52] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 10:39:52] Benchmark of 317.631. Time: 25m07s. SL of 10.00
[02/February/2017 10:39:52] Time left: 50m27s
[02/February/2017 10:39:52] Extra Sampling: New global target SL: 12.65
[02/February/2017 10:51:33] Benchmark of 306.391. Time: 36m48s. SL of 10.84
[02/February/2017 10:51:33] Time left: 46m46s
[02/February/2017 11:09:08] Benchmark of 298.279. Time: 54m23s. SL of 11.73
[02/February/2017 11:09:08] Time left: 28m53s
[02/February/2017 11:35:39] Benchmark of 292.218. Time: 1h20m54s. SL of 12.65
[02/February/2017 11:35:40] Ending render process..
[02/February/2017 11:35:40] Render finished successfully

Rendering process [V4]:

[02/February/2017 12:41:21] Extra Sampling: New (estimated) global target SL: 12.64
[02/February/2017 12:41:21] Benchmark of 312.963. Time: 13s. SL of 1.00
[02/February/2017 12:41:21] Time left: 2h16m40s
[02/February/2017 12:41:43] Extra Sampling: New (estimated) global target SL: 12.64
[02/February/2017 12:41:43] Benchmark of 297.548. Time: 35s. SL of 2.00
[02/February/2017 12:41:43] Time left: 1h25m22s
[02/February/2017 12:42:16] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:42:16] Benchmark of 292.975. Time: 1m08s. SL of 3.00
[02/February/2017 12:42:16] Time left: 1h22m05s
[02/February/2017 12:43:04] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:43:04] Benchmark of 294.669. Time: 1m56s. SL of 4.00
[02/February/2017 12:43:04] Time left: 1h22m39s
[02/February/2017 12:44:18] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:44:18] Benchmark of 292.482. Time: 3m10s. SL of 5.00
[02/February/2017 12:44:18] Time left: 1h14m27s
[02/February/2017 12:46:06] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:46:06] Benchmark of 294.142. Time: 4m58s. SL of 6.00
[02/February/2017 12:46:06] Time left: 1h18m33s
[02/February/2017 12:48:44] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:48:44] Benchmark of 297.759. Time: 7m36s. SL of 7.00
[02/February/2017 12:48:44] Time left: 1h12m04s
[02/February/2017 12:52:40] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:52:40] Benchmark of 300.527. Time: 11m32s. SL of 8.00
[02/February/2017 12:52:40] Time left: 1h09m17s
[02/February/2017 12:58:34] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 12:58:34] Benchmark of 302.338. Time: 17m26s. SL of 9.00
[02/February/2017 12:58:34] Time left: 1h05m01s
[02/February/2017 13:07:49] Extra Sampling: New (estimated) global target SL: 12.65
[02/February/2017 13:07:49] Benchmark of 298.988. Time: 26m41s. SL of 10.00
[02/February/2017 13:07:49] Time left: 1h13m16s
[02/February/2017 13:07:49] Extra Sampling: New global target SL: 12.65
[02/February/2017 13:16:45] Benchmark of 278.029. Time: 35m37s. SL of 10.52
[02/February/2017 13:16:45] Time left: 1h01m53s
[02/February/2017 13:22:41] Benchmark of 271.373. Time: 41m33s. SL of 10.84
[02/February/2017 13:22:41] Time left: 52m43s
[02/February/2017 13:30:38] Benchmark of 270.440. Time: 49m30s. SL of 11.26
[02/February/2017 13:30:38] Time left: 43m00s
[02/February/2017 13:41:18] Benchmark of 269.614. Time: 1h10s. SL of 11.73
[02/February/2017 13:41:18] Time left: 29m34s
[02/February/2017 13:52:25] Benchmark of 269.826. Time: 1h11m17s. SL of 12.14
[02/February/2017 13:52:25] Time left: 18m27s
[02/February/2017 14:00:27] Benchmark of 269.619. Time: 1h19m19s. SL of 12.40
[02/February/2017 14:00:27] Time left: 9m11s
[02/February/2017 14:08:54] Benchmark of 269.360. Time: 1h27m46s. SL of 12.65
[02/February/2017 14:08:54] Ending render process..
[02/February/2017 14:08:54] Render finished successfully

Sumarry [V3 vs V4]:

Total time: V4 was slower by 6m 52s than V3.

1h 20m 54s vs 1h 27m 46s

Benchmark [V3 vs V4]:

292.218. vs 269.360.

The final MXI files have approx. same size 405MB ( V4 file is larger by 3kB).

Was there something changed under the hood in Maxwell V4 CPU version ?
There weren't announced anything new.
Is there plan to make some optimization so the V4 will be at least fast as V3 ?

(I don't plan to switch to V4 if it's slower than previous version )

[luis.hijarrubia]

Sumarry [V3 vs V4]:

Total time: V4 was slower by 6m 52s than V3.

1h 20m 54s vs 1h 27m 46s

Benchmark [V3 vs V4]:

292.218. vs 269.360.

The final MXI files have approx. same size 405MB ( V4 file is larger by 3kB).

Was there something changed under the hood in Maxwell V4 CPU version ?
There weren't announced anything new regarding to CPU version.
Is there plan to make some optimization so the V4 will be at least fast as V3 ?

(I don't plan to switch to V4 if it's slower than previous version )

No big changes on CPU from latest 3 to 4. The thing is not our code itself, but the compiling process. We changed the compiler, and we have to optimized to the maximum again as maxwell 3 was. Initial maxwell 4 was about 30% slower, and now it's like 10% slower. We are still working on it and hope to have it at least as fast as maxwell 3.

[feynman]

In a nutshell, that means that for production work, V4 is not usable still (CPU rendering too slow or GPU rendering missing essential functions) and one must either remain with V3 or change renderer.

[luis.hijarrubia]

In a nutshell, that means that for production work, V4 is not usable still (CPU rendering too slow or GPU rendering missing essential functions) and one must either remain with V3 or change renderer.

Well, the cpu is just 8% slower than in V3, although we are not happy with that, it's not like not usable for production.

[T0M0]

That's true Luis.
Compiler was changed due CUDA/GPU engine ?
And when can we expect new early build ?

[luis.hijarrubia]

That's true Luis.
Compiler was changed due CUDA/GPU engine ?
And when can we expect new early build ?

No, most of CUDA code (the one running on gpu) is compiled with it's own CUDA compiler. The change is natural, new versions, dropping 32 bit...
I hope to have a new early build this week.

[feynman]

In a nutshell, that means that for production work, V4 is not usable still (CPU rendering too slow or GPU rendering missing essential functions) and one must either remain with V3 or change renderer.

Well, the cpu is just 8% slower than in V3, although we are not happy with that, it's not like not usable for production.

Thanks, Luis - I hear what you say and appreciate your efforts.

Mind you, an 8% reduction in speed is not what many design studios were looking for... One must hope that V4 GPU rendering becomes fully functional in a short period of time...

[jwiede]

In a nutshell, that means that for production work, V4 is not usable still (CPU rendering too slow or GPU rendering missing essential functions) and one must either remain with V3 or change renderer.

Well, the cpu is just 8% slower than in V3, although we are not happy with that, it's not like not usable for production.

Luis, with all due respect, an 8% CPU performance LOSS with minimal benefits over V3 is a bigger deal than you're admitting, esp. almost four months after "release". If GPU render were, as NL had advertised, at full parity with CPU render, the situation could be different, but it is not.

NL is asking existing customers to pay a MUCH higher price just to receive a SLOWER CPU renderer, and an incomplete, buggy GPU renderer in return -- this is a key factor why so many customers are so upset with NL over the MaxwellRender licensing/pricing changes. The progress over the last few months also makes it quite clear that GPU render parity with CPU render is not going to arrive in any sort of near- to mid-term timeframe.

[F. Tella]

Hi,

We are testing right now a new early build that solves the problem with speed. We'll keep you updated.

Best,
Fernando