Having issues with nVidia 5080 and ToneX modeling engine

[bakerjt]

I've got a ticket in with support, and I am hopeful they will get to the bottom of it, though at this time it's still an open issue. I am not making this thread out of dissatisfaction with IKMM support, to be very clear, just trying to get a sense of if others are having a similar issue.

I upgraded my GPU from an nVidia 3080 12GB which worked really well with ToneX and made profiles relatively quickly with Advanced training. I ended up with an nVidia 5080 in my PC, and unfortunately now I cannot use the ToneX capture feature. It gets the audio portion just fine, but when I try to start training of any kind, it just hangs for a long time and then eventually tells me it could not start the modeling engine.

I am wondering if it is something related to the same problem that made, e.g. DaVinci resolve incompatible with these new cards until they released a beta that works - nVidia has discontinued 32-bit CUDA in their hardware, basically, following on moves they made starting with their Hopper microarchitecture some years back. Now, old videogames that used 32-bit PhysX no longer can calculate on it, and if an app isn't up to date to be able to use the most recent libraries, the app won't work. I don't know if that's what the case is here, but it is an nVidia card as recommended, and ToneX tries to turn on the modeling engine and fails. That is compared to when I had an AMD 9070 XT in it briefly, and it just used the CPU fallback automatically (which did work, albeit slowly, my 12900K processor took 5+ hours to do an Advanced training) - I can't seem to make it use the CPU fallback with the 5080 in my system, so right now, I can't make captures at all on my PC.

Anyone else with a new nVidia card struggling to use it in ToneX? Or, more helpfully, does anyone hav a new nVidia card from the 5000 series, and their ToneX captures work just fine? I'd like to know if I'm alone with the issue or if there are others experiencing it as well. If nobody can use the new 5000 series cards with ToneX modeling, that could be useful in pinpointing the issue.

[bakerjt]

I can't edit the original post but I meant to say "anyone have a 5000 series card and their ToneX Modeling Engine works just fine during the capture process?" - obviously, the captures themselves still work great, it is the process of MAKING captures that seems to be broken at least for my setup. If there are nVidia RTX 5000 series users out there who can try the ToneX modeling and see what happens, I'd appreciate it!

[pbwagner@sasktel.net]

Hello BakerJt,

Just wondering if you are getting any help. I am Brad that started the other thread on this. My card is a 4060.

Anyway, I am losing hope. This started for me in Feb. I have done everything support has asked me to do (updating drivers, windows, software reinstalls etc etc etc) and nothing. Last time I heard from them was April 2nd. Pretty disappointing especially because I have bought one of their pedals as well. I have quite a few amps in my studio that I was hoping to profile but stuck in the water here.

Anyway, just wondering how your experience is going?

Regards,
Brad.

[bakerjt]

Hm, a 4000 series card would still have the 32-bit CUDA hardware intact, maybe it is not just the 5000 series' inability to do that at play then. Well, hard to say based off of just two samples, really. They are still working on my issue, I've been receiving responses and suggestions to try different things - I need to do some more troubleshooting steps they want today and get back to them, just been very busy with other things but their last response to me was yesterday and I feel they have tried to stay on top of it behind the scenes.

I believe my next move will be trying to train with my Haswell 4770K-based PC with a RTX 2070 in it. I could then at least train pedal models which I would like to do, though I've been itching to train on my Mesa DC-3 and it will be a challenge to say the least to get it set up in the same area as that PC, haha, it's been taken over by my kiddos for Minecraft etc.! But if it works for training while my PC doesn't, maybe a useful data point as I continue to try to troubleshoot what's going on. I am eager to get back to being able to train with my main DAW PC again, hoping it won't be too much longer 'til we can figure something out.

[jezynski]

Hello! My system is Win10, and my computer is old: i7 4790m, 16GB RAM, NVIDIA GTX1660Ti. With the new NVIDIA drivers, it doesn't want to model at all and crashes in various situations. Currently, I'm using the old driver 528.02, and only with it does the advanced mode of the modeler work. It's strange that the engine is so sensitive to the driver—it doesn't release the memory of the graphics card and fusses about memory being borrowed by other programs. But for now, the old driver allows me to model in advanced mode with a time of 19 minutes. Perhaps the problem lies on the side of IK because, from what I see, the modeler's software files are from around 2022...

[bakerjt]

Well, support just told me that they have zero reports of this happening with any GPUs, suggesting that it has to be an issue with my system only. Where I'm at now is I have to disable my GPU in Device Manager in order to force ToneX to use my CPU and then it takes 4+ hours to do a training (they still turn out the same, though, it just takes longer). I'm glad I can still train an occasional model, but that certainly isn't the workflow I got used to with my old 3080 12GB, where I could make and train about 3-5 captures per hour depending.

Overall it's a bummer. Love ToneX software, love how it was working for me previously, my old captures still sound great and will be on recordings - I'm not trying to come down on anything here, just hoping we can get some progress as more data comes in, I remain willing to do whatever is necessary to troubleshoot.

If other users are experiencing this, please do start a ticket with IKMM and report that you too are having the problem so that we can hopefully get more data going on and they can figure it out.