Titan X compatability with older 2688 core Titan's when rendering with Iray.

I just finished speaking to a support rep at Nvidia, and he suggested that I ask here for advice.

I currently have 4x GTX Titan (2688 core) on an Asus Z9PE-D8 WS With dual Xeon E5-2687w, for rendering with Iray / Mental Ray.

I am wanting to replace the primary card with a Titan X, and keep the other older titans to help with GPU rendering.

The question is, can I use a Titan X and older 2688 core Titans at the same time when rendering with iray?

I have read that trying to use a an older Titan and a GTX 580 together in iray produces an error, so will using a Titan X and 3x Titan 2668 core together also produce an error in iray?

I am using the latest version of max 2016, which I believe already has the Maxwell Patch applied.

My O/S is Windows 7 pro x64.

I am aware that SLI will not work between the different chip sets.

I am also looking for any benchmarks that show how much faster the Titan X is compared to the older gen GPUs like the Titan (2688 core) 780ti etc in Iray. It will help me to decide if its really worth the upgrade.

Is it possible in one of Nvidia’s workshops/labs to put a Titan X and older Titan in the same system, to see if they can both be detected by 3ds max, and render together in Iray without errors?

Would be really helpful.

I am aware that using the older Titan’s with the Titan X will limit the Titan X memory to 6gb.

I plan to eventually replace all the old Titans with Titan X, but only if the benchmarks justify the cost. I don’t want to end up replacing all the Titans with Titan X only to find that an 11,000 iteration render on an interior architectural scene that takes 15 minute to render, is only one minute faster. It really wouldn’t justify spending £3500 on four new cards. So if there isn’t much improvement over the old Titans, I will only want to replace the primary card for connectivity reasons, but that’s only if its compatible with the older Titans in iray.

Does anyone know where I can find the information I am looking for?

Many Thanks.

Nice GPU ray tracing system!

Not sure to whom you spoke, but there are dedicated Iray and mental ray forums on the NVIDIA Advanced Rendering Center site.
There are “Forum” or “Community” links on the individual product pages here:
http://nvidia-arc.com/products.html
Direct link to the Iray forum:
https://forum.nvidia-arc.com/forumdisplay.php?181-Iray

Note that SLI only gets in the way for GPU ray tracing using CUDA. Always disable SLI for that or the OS potentially allocates surfaces multiple times needlessly.

In general mixing GPUs of different GPU architectures is a bad idea and it would need to be tested if you could get the Maxwell based Titan X (GM200) and the Kepler based Titans (GK110) to run under one display driver at all.
Since you cannot switch GeForce boards to TCC mode, that driver compatibility would be your main concern.
If that worked, you can select which GPU devices are enabled for CUDA operations in the NVIDIA control panel (default all) and Iray can select from the enabled devices, so selecting either the Titan X or the any combination of the three Titan boards as CUDA devices should be possible.

Disclaimer: Such configurations are untested and I don’t know of anyone who has a similar setup to say if that works.
Maybe the Berlin colleagues on the Iray forum have seen more such configurations.

Hi Detlef Roettger,

Thanks for your reply. I’ve been using Iray for production rendering for the last 3 years so know things like disabling SLI, PhysX set to CPU etc.

I have written the same question to the Iray Development team so waiting on the reply.

I think I may have to just upgrade all 4 Titans, but I am waiting on some benchmarks first comparing the Titan X with the older Titan 2688 core before I decide.

I found some worrying benchmark results on the German Tomshardware testing the Titan X against the Quadro M6000.

http://www.tomshardware.de/geforce-quadro-workstation-grafikkarte-gpu,testberichte-241759-3.html

The M6000 is 3x faster than the Titan X in the iray results, but this really doesn’t make sense.

The M6000 and Titan X have exactly the same estimated single precision performance at 7 Tflops.

The benchmarks for other renderers such as Blender and Octane are exactly as expected with the Titan X being slightly faster than the M6000 because of the higher clock rate.

So this means either there is a problem with the way TH Germany set up the iray benchmark, or Nvidia who develop Iray have intentionally crippled Maxwell Desktop GPU’s to force sales of Quadro cards, but I think this would be a seriously bad business move seeing as the majority of iray users from freelancers to small studios opt for Titans over high end Quadro’s because of budget, so Nvidia would risk loosing tens of millions in sales with such a dumb move. I hope its just a mistake with the iray test. The Titan X should definitely be faster than the M6000 in this test.

I can afford to buy 4x Titan X at around £3500 but no way can I afford to buy 4x M6000 at £16400 and I don’t think the majority of freelancers and small studio’s can, so I think if Nvidia have decided to cripple the iray performance in favour of Quadro cards, people will simply stop using iray over time, and it will be left to a niche number of users that can afford a quad M6000 set up and eventually loose development funding.

Maybe they didn’t apply the Maxwell Patch to 3ds max 2013, but that would not explain why the Quadro performs normally. It cant be a driver/hardware problem or the Blender/Octane/ratGPU and Luxmark results would be showing the same poor performance, so has to be software related.

The Iray forum registration never works, it always says there is something wrong with the password.

The passwords I have tried are all containing numbers and letter and over 8 characters long.

Did you try adding special characters like @!./ to the password?

This issue has been solved. The question was raised after a benchmark posted on Toms hardware DE that turned out to be a rotten egg. Plus there were a few people on forums posting that their Titan X cards do not work in Max 2015/2016 with the Maxwell patch applied, that didn’t even own the Titan X card yet, and only had it on order!

Anyway,

Magenius posted some benchmarks a while back, and as expected the TItan X is a little faster than the M6000 for Iray rendering because of the higher clock rates.

Link is below.

After all that, I decided to hold off on getting the Titan X’s. Replacing my 4x Titans with 4x Titan X, would mean a 46% speed up in Iray, but I will wait for the dual GPU version, not sure if it will be called the TitanX2 or GTX 990. That way I will not have to have 4 cards sandwiched together, and have better cooling with similar perforce as having 4 cards.

Although I am tempted still to get the Titan X’s. But if I do, it will be after the Computex and E3 shows, as I am sure Nvidia will drop prices with the new competition from AMD.