[Brian Buxton]

Random Control does a great job keeping users up to date with development.
The comparison with NextLimit is night and day.

[Niels van Oostenbrugge]

Tested the latest release of Maxwell Render within dual 4090 RTX

Tested with NVIDIA RTX 6000 too.

GPUs not recognized.

Requested an ETA for an update to support new RTX 4000, they don’t have actually any plan for an imminent update/release to support 4000 series.

So the latest viable solution for a gpus rendering workstation are 3090/3090ti or quadro counterpart

Hi Matteo,

NL just wrote me this last week, regarding “possibly” getting my RTX 4090 to work with MR (although they wrote about the 4070ti, which i don’t have). Have you tried this, and if so is this (rather unelegant fix) the solution to get fast gpu rendering using Ada Lovelace architecture?

Message from Fernando:
I've been asking my colleagues in case there was an easy way of making Maxwell 5 capable of using your 4070 Ti card. They told me about a harmless hack that could work.

If you want to try, please, follow these instructions:

- Go to this folder: C:\Program Files\Next Limit\Maxwell Render 5.2.1.49\extras-x64-v140
- You'll find several files (100). If you take a look at the last two figures of the name, you'll find that they repeat 10 times each with different endings that also repeat 10 times each These have to do with the GPU architecture. For example: 5a9b32231634eacfd1f41cbf9432c68480 and 5a9b32231634eacfd1f41cbf9432c68486 or 64b3ee06eddca2efe110fc8fa57c43fb80 and 64b3ee06eddca2efe110fc8fa57c43fb86 (check the 80 and 86 at the end of each of them.
- The idea is to duplicate the ten files that end up with 86 and replace the 86 with 89 like this: https://drive.google.com/file/d/12Z6hdN ... sp=sharing

After that, you can open the mxgpuprefs.exe application in the Maxwell Render 5 installation folder to see if your card is accepted.

[Niels van Oostenbrugge]

Random Control does a great job keeping users up to date with development.
The comparison with NextLimit is night and day.

I’ve tried Maverick and ooh man, i imported a model in the automotive preset file and within 45 minutes of texturing etc. i had a high resolution realtime rendering looking at least just as good as anything i’ve produced with MR. Rendering itself took a minute or so. What a joy this engine is (especially on a rtx 4090). However, it looks like it’s very much focussed on close-up small scale product rendering, don’t see much exterior/architecture examples around. But, biggest reason i’m not buying Maverick is the fact that it’s incredibly unstable. I imported several Rhino models using the converter and all models which were originally stepfiles crashed after ‘some actions’.

[Matteo Villa]

It worked.

Had to:

• clean instal of Maxwell Render
• reapply license
• modified the 10 files you suggested
• after an initial crash and restarting the Software now I can finally render with my dual 4090 setup.

And it’s fast. Really fast

I don’t know if they didn’t released an updated version within this workaround because of some instability issues.

But for now, testing within small project and preloaded scenes seems to perfectly work.


Ps: little detail to underline:

- its more stable with Nvidia Studio Driver
( avoid gaming one’s )
- gpus are both recognized as Ampere in the console
- scaling isn’t linear. Performance can be improved
- Gpu utilization it’s around 97/98% - never reach 100%

Hi Matteo,

NL just wrote me this last week, regarding “possibly” getting my RTX 4090 to work with MR (although they wrote about the 4070ti, which i don’t have). Have you tried this, and if so is this (rather unelegant fix) the solution to get fast gpu rendering using Ada Lovelace architecture?

Message from Fernando:
I've been asking my colleagues in case there was an easy way of making Maxwell 5 capable of using your 4070 Ti card. They told me about a harmless hack that could work.

If you want to try, please, follow these instructions:

- Go to this folder: C:\Program Files\Next Limit\Maxwell Render 5.2.1.49\extras-x64-v140
- You'll find several files (100). If you take a look at the last two figures of the name, you'll find that they repeat 10 times each with different endings that also repeat 10 times each These have to do with the GPU architecture. For example: 5a9b32231634eacfd1f41cbf9432c68480 and 5a9b32231634eacfd1f41cbf9432c68486 or 64b3ee06eddca2efe110fc8fa57c43fb80 and 64b3ee06eddca2efe110fc8fa57c43fb86 (check the 80 and 86 at the end of each of them.
- The idea is to duplicate the ten files that end up with 86 and replace the 86 with 89 like this: https://drive.google.com/file/d/12Z6hdN ... sp=sharing

After that, you can open the mxgpuprefs.exe application in the Maxwell Render 5 installation folder to see if your card is accepted.

[Niels van Oostenbrugge]

Hi Matteo,
That’s good news!
I haven’t tried yet, but probably will. In the meantime i’ve switched to Twinmotion and i’m liking it! The fact that NL doesn’t/hasn’t released an update, V5x/V6, with this rather simple fix to me just underlines again that development has stopped. Again, too bad there’s no communication regarding this (except the note, still, that Ada Lovelace is not supported).

Anyway, thanks for the feedback!

Best regards,
Niels

[Matteo Villa]

It’s a “Simple” fix. But not perfect.

I tested the benchwell benchmark and you can see my latest score posted online.

I’m not far from a build with 3x 3090ti + 3090
while using only 2 RTX 4090.

But it’s a workaround since

• you still loose some performance on the fly
• the 2x 4090 doesn’t hit the 100% usage ( hitting 97-98 %) and sometimes they drop the score drastically.
• Plus rtx4000 are seen by Maxwell as ampere generation.
• Maxwell strangely see only 23 of 24 VRAM

It’s usable yes but they need to optimize and check the code to support the new tensor/Cuda core.

Sometimes the render start with an higher benchmark score, but suddenly the estimated time to finish the render jump from 1 minute to 15/20 minute.

Imagine something similar happen while rendering a project with a strictly delivery timeline.

That’s probably a bug in the code they’ve to resolve to fully support ADA LOVELACE.

But like you underlined the question is if they still want to release a fully optimized/updated software.

When it’s work, that little hack is a game changer for who’s using RTX 4000 series.

[Forester]

Niels, you are correct that there are few examples of architecture for Maverick. I've done quite a lot of them for my clients. I've also acquired quite a few outdoor scenes made with Maverick.

Here's one quick example of an architectural scene. Maybe the image size here is too small to make out the spice jars inside the alcove on the right. But, this simple scene has well over a hundred objects in it - all textured with *sbsars, rather than with conventional shaders or texture maps. The original larger-size render required just about one hour to render in Maverick. It required two further minutes to render in Maxwell, and I think I got about the same quality.



Here is a Maxwell version with slightly different lighting, camera focal point, fewer objects, and maybe a very slight difference in the camera position.



So, I like Maverick for tons of things. But, the hitch is that any purely-GPU based rendering engine is going to slow-down a LOT when the scene contains hundreds of objects. I use both Maverick and Maxwell regularly, because when it comes to a complex architectural scene with lots and lots of objects, CPU rendering with a full deck of memory sticks is better than GPU rendering. This seems to be true, even when you're sporting one of the big nVidia cards. I believe that most professionals know this - that is, that purely GPU rendering engines lose their competative advantage over engines that use CPU's or can use both, like Maxwell, when it comes to scenes with hundreds of objects.

[Forester]

That being said, I did this architecture work in Maverick first, because I could just use the *.sbsars and work out a reasonable set of texture files before proceeding with anything else. I also made some renders in Maxwell. Notice that the Maxwell shaders can be made to be almost identical to the Substance Designer *.sbsars. The client ended up choosing a mixture of the Maxwell and Maverick renders for her project. She doesn't notice any difference in the final renders.

[Mark Bell]

Interesting observation between the two renderers - GPU and CPU with only a couple of minutes difference in render time. With all the hype on GPU you'd expect it to be much quicker. I recall a video Next Limit made a number of years ago filmed at Siggraph on their work with GPU that highlighted similar obstacles you note about limitations on render speed with larger or more complex scenes.

Acknowledging the two images are different mood lighting, I find I prefer the bottom image due to it showing more of the scene through the wider angle and brighter illumination (it explains the space better), but prefer the top image due it being a tad closer to eye level, and more convincing (to me). When I zoom in closer in the top image, there are areas which appear to show noise whereas the bottom image appears clean throughout. Did you use the Maxwell Denoiser?

[Forester]

Hello Mark.

I thought everyone was aware that GPU bcomes more limited the greater the number of hi rez objects in the scene.

The "noise" in the renders probably is because I reduced these from their original large size - using Topaz Studio. And, "No," I didn't make use of the Maxwell denoiser. Wasn't wanting anyone to look at these extremely closely. Just making the point that there is very little difference between Maverick and Maxwell, both in terms of speed in the presence of a large number of objects, in terms of shader results, and in terms of overall appearance/quality.

I feel very comfortable using both. I prefer Maxwell whren I know there are going to be a lot of high rez objects with full PBR texture sets. I like Maverick for quick block-outs, and for experimenting with textures. This preference may be more suitable for those of us that a fully devoted to Substance Designer. I've pretty much ditched Photoshop and Coredraw and any other material maker because I finally hit a skill level where I can do almost any thing faster in Substance Designer. So, Maverick is good for me for my initial start. However, when I start hanging it due to scene complexity, its time to let it go. I have a lot more confidence in Maxwell for the final shots for any big scene.

[Mark Bell]

Hi Forester,

I thought everyone was aware that GPU becomes more limited the greater the number of hi rez objects in the scene. - probably lost somewhere in the back of my mind. Reading your comment made me aware of this (again). The interesting thing I noticed from what you posted was that there was little difference in render time between GPU vs CPU renders and that, to me, the lower Maxwell image looked better/clearer. This challenges the perception that Maxwell is slow and that GPU rendering is the only way to go.....so thanks for the post. This time I learnt something

[Forester]

Thumbs Up !