Looking for Recommendations for Wokada Dual PC Stream Setup.

[kirkyeehee]

So I'm in a situation right now where I am actively streaming using W Okada, as well as recording and streaming at the same time with OBS and one thing I'm noticing is when I hit intense games like Final Fantasy 7 Rebirth, Monster Hunter Wild, (the benchmark) And other games require more resources, I'm noticing the hardware I currently have cannot keep up with everything and it's causing games to stutter and play bad, as well as causing issues with W Okada breaking up.

Ideally, I would have my V Bridger and V tube studio running on the same PC that would have OBS and W okada on it as well. so my main PC could be dedicated to the games.

(Current build:
Ryzen 9 7900x
Nvidia RTX 4080
64GB DDR5 Ram
This is what I plan to dedicate to gaming)

I plan on doing a dual PC setup for streaming and getting a new high end 5000 series is out of the question, same for a high end 4000 series card. (4080 and above is out of my current budget) If I have to buy other parts as well (and I do) then it has to be mid range at best. So maybe someone with experience could guide me on what might be a good middle ground. For a CPU I already have a Ryzen 7 5800 X from my previous build, so I'm looking to build around that.

I currently have the chunk set to 128 so I can stream within OBS's limits and I've recently dropped the Extra from max (131072) down to 16384 to reduce the CPU usage from around 50$ to 10% (which has helped some with the voice cutting out, but not completely for those demanding AAA titles) Ideally, I'd like to be able to play games without having to set everything to the lowest settings and hope that it's enough. Between having to run OBS on my computer and W Okada, it takes a lot of resources away from the game and everything suffers for it.

Any input would be appreciated

[Kuuko-fokkusugaru]

In your case, and also being vtuber, I can understand pretty well your situation. Definitely, vtubing app + games + RVC is a really high cost to keep up with everything.

If you already have a CPU laying around, I'd recommend getting the cheapest motherboard available with 16 GB of RAM and a nvidia 3060/4060. I would leave this little setup solely for RVC and do the rest of tasks in your main computer. You probably could also run vtube studio on this little (not so little really lol) machine but depending on how heavy the model is, it could interfere with RVC performance. You would run RVC as server instead of client and connect your mic to the little pc while running a cable from this to the main computer as if the little one were a mic. It's, after all, the main purpose of how RVC should work. You would run RVC through start_https.bat instead of the regular http one so you can access to RVC UI through your main computer just using a browser.

All you need to keep in mind is that, the less powerful the GPU, the less tasks you can do with that one but ideally you want to isolate RVC from any other heavy task. This also goes for OBS. If you want to use OBS + RVC in the same computer, it also depends on how many layers you got in OBS, if they are heavy, etc (you know that layer that makes emotes bounce around screen when viewers sends them? That one is a hell of bad optimized stuff that slows down everything Idk why lol).

I hope that helps. Feel free to ask more questions if you still have some doubts or if you want a different configuration setup. Also, if you want, drop your channel URL here so I can check and drop a follow :P

[kirkyeehee]

Would there be any extra latency introduced like this or would it be as low as what I can do with my current setup? (My worry is if there is extra latency added by running RVC as a server)

I also don't think I have anything that would be considered heavy as far as layers go in OBS, but because I do stream on both Youtube and Twitch, it does mean I have to double up on some things so there are a lot of layers all the same. Most of them being your typical overlays of chat for both platforms, pop ups and notifications when someone follows, gifts the sub, and all that stuff, and of course there's the Game itself when I'm streaming that.

The "heaviest" thing by far is recording the high quality vod for YouTube afterward that has the music removed and is at a much higher bit rate so it doesn't look like complete crap.

I'm also not 100% sure what you mean by "running a cable from one Pc to the other." I have absolutely no familiarity with a 2PC setup for something like this so I'm not sure if there's a special kind of cable, or if it's just a standard usb cable, or maybe even a network cable, or something else. I do not have a physical media mixer like a lot seem to. Too rich for my blood and this desk... Yea no. No room on it. (I need a new and bigger desk, this one is tiny)

Testing out the benchmark tool for Monster Hunter Wilds made me realize if I want to stream Triple A games I absolutely need a 2 PC setup or a 5090, and the 2PC setup is probably cheaper considering a 5090 is going for $3000 and above at retail price here in Canada. (before taxes)

When testing that I was getting around 8FPS in cutscenes with frame gen on, and the gameplay part of the benchmark was at least showing 40 to 60. Compare that to when I tested it out without any of the extra stuff I was hitting my max frame rate of my monitor with the settings maxed out. Also with frame generation on because at this point in time it seems like all Triple A games need that stuff to run at their best. Without frame generation, it broke everything. The stream was freezing up for several minutes at a time, the PC would freeze up, and my vram was just maxing out, causing everything to die until I closed out of the benchmark.

I should be able to play it on lower settings obviously, but you don't put big money into a system like this just play it on low settings right? I put a lot of work into getting this gaming computer And I'd like to get the most out of it if I could. Either way, you've given me a lot to think on and if there's anything else you can offer or suggest I am all ears and treat me as if I'm a 5 year old that knows nothing about any of this because, yeah I'm still learning. I've only been streaming for just a touch over 2 months between other work, so it's been finding time as I go to figure this stuff out. I appreciate you taking the time out to help. Chances are I will be slow to respond because of other commitments. (a stream when I wake up)

and finally... here is my channel on YouTube. The stuff with the voice model is all the recent stuff. You can ignore the past 11 years of stuff as myself lol. If this isn't allowed for some reason, please delete it. I'm only posting because it was asked for.: https://www.youtube.com/@Kirkyeehee/videos

[Kuuko-fokkusugaru]

I am not sure if is allowed or not, but I am pretty much the only one giving support to this software commenting here and there the whole time. The software has been without much development for the past months so I highly doubt Okada may care much lol.

Ok, so about latency, here is the thing. When you use server mode, and this is something that you can already try by yourself, using WASAPI mode for the input and output, the latency is smaller than when using the client mode. That's cool, right? :D that's because it access to the drivers in a more direct way. This can be used even if you haven't a secondary computer, but sometimes you could find that certain software may not be able to access certain inputs and outputs because RVC is like taking complete exclusive mode not leaving access to others apps (this doesn't happens always so feel free to test). About using a cable, it's easy! So your motherboard have a mic input 3.5mm jack connector, and your little one would have also output 3.5mm connectors. All you would have to do is to buy a cheap and simple male to male 3.5mm jack cable to connect one computer to the other so the audio of the little one goes to the main one as a microphone while you would connect your REAL mic to the little one through whatever method you use already, USB, 3.5mm jack, etc... There are also ways to stream the audio from one computer to another with third party software over a local network but that could possible add a tiny bit of extra latency and it feels like an extra hassle to setup when just a simple cable would do the job.

I have been thinking, and no matter what we look for, we both know that you may have to spend some money, but that's not new, after all. If you want to take advantage of your GPU and get the most out of games, there is a setup that is not that pricey and extremely good for vtubing. So you clearly need your main computer to squeeze every possible frame for which a 4080 is great. But then the rest of the software can be quite heavy, especially Vtube Studio and, to some degree, OBS. You could aim for a second machine with an APU. An APU is not that expensive really and it includes an integrated mid range GPU and a CPU. The integrated GPU may not be top performant for gaming at max quality but they are very good and should be more than enough to run something like vtube studio and OBS, while you would run RVC on a 3060 (if going for a cheap option). A 4060 would probably more suitable if you want to run RVC + OBS + VS but I am not sure how much it can hold all of that. I will guess it should do it fine. If not, you could move OBS to the main computer and stream VS and RVC to the main computer. Yes, you can stream vtube studio from one computer to another (or at least you should be able to). For example, when I stream and I also lack performance, I use my vtubing software using the integrated GPU on my i7 14700K. This GPU is quite weak, but it's fine to run Warudo (I use 3D model) at 1080p. Then I leave the rest of the performance on my 3080Ti for games and OBS. That said, I don't usually play extremely high end games and I rarely use RVC as I usually do it using a TTS (I am not confident enough to speak on mic). That means that, if you pic a CPU with integrated graphics and also a mid range GPU, you can divide all the tasks between 3 different GPUS, the integrated weak one (if doesn't work for vtubing, it would for OBS at least), the little one for RVC (and possibly vtubing if necessary), and the main big one for the latest games. Consider it a way of saving some money by not having to get a big GPU. I am solely naming this solution because I can't confirm whether a 4060 would be enough to run RVC, vtube studio, and OBS all on itself alone (maybe is?). You would need a capture card though. But you don't need to go crazy on this because you won't capture using the card but OBS. There are ways to send the image from one computer to the other through your network by using something called "NDi". There may be other ways too but I didn't research enough. And, like mentioned before, there are cheap cards that you can get even from AliExpress for like 20$ to simply transmit the HDMI signal from one computer to the other. Those devices usually have one input and two outputs, one in form of USB to capture the video in the secondary machine and another one in a form of HDMI to return the signal to the monitor. But here we you would need to consider several things. Not all of these cards, especially cheaper ones, may support variable refresh rate, high refresh rates, display port, etc...

As you can see there are just too many ways of doing this. The best way to know which one you should use is to determine if your main computer can handle games + extra things others than RVC because RVC would go to the second machine. If you can run everything fine and your only issue is RVC, then you would just use the second machine as a mic for RVC. But if you run into other performance issues due OBS or vtube studio then you may need to start considering adding that extra weight to the secondary machine.

Sorry for the long text but it's hard to explain all this so I just tried to put different scenarios. You can keep asking if you want or we could talk on Discord if you prefer that. Just bear in mind that, while I know about hardware, I do not know of all the performance of each. I can tell by helping people here with RVC, that some people got a really big GPU usage using RVC on certain GPUs so I am not sure how much juice you can take out of a 3060 or 4060 at time of combining RVC with extra stuff, hence the suggestion of maybe getting a CPU with integrated GPU but maybe is not even necessary at all. Your 4080 and my 3080Ti both have a really high amount of CUDA cores so this help greatly with RVC. I barely use 10% of GPU for RVC so I have the rest for games. Like you also mentioned, capping fps to 60 or lowering some settings also helps. I don't play fortnite at max settings because it gives some tactical disadvantages so I can run the game at like 180fps but to be able to use RVC without a huge delay or stutter, I have to cap it at 120fps (and this is without streaming lol).

[kirkyeehee]

I've done tests without using the RVC model on a private stream set up as if I was actually doing a stream and out of everything the RVC takes up the least out of it all. Unfortunately, when Steph needs to be sacrificed, typically V RAM, that's the first thing to suffer. My voice will start stuttering before the game starts giving out or my Vtube model starts giving out. I was actually shocked at just how little V RAM W Okada takes, maxing out at around 1.5 GB. Depending on the extra number selected, it can take up to 50% of my CPU. (Typicaly when I choose 131072) Still with everything though (games included) it never typically pushes past 80% Same with overall GPU usage.

Yeah it's always a limitation of V Ram that hits me the hardest, which is crazy when you have 16 GB of Vram lol. (If I could only download more like you can with normal ram :P) OBS while recording and streaming at the same time takes about 20-30% of my CPU and around the same for GPU. (so around 4-6GB of Vram when I checked. Which was perfectly fine until I decided to stream a AAA game.

That's where it gets really bad when you get right down to a comma to get a new PC means I have to put money into it but I can't readily test this stuff without doing so. I'll never know if the specs you suggested would work well without actually just purchasing it and testing and if it turned out it didn't for the way I would prefer to do it, then at the end of the day that's money out of my pocket.

Now that I think about it, I wonder if just getting something like a 4060 and putting it into my current PC would be enough (and selecting that just for RVC and maybe OBS, but likely just RVC). Obviously, I'd need to get a little bigger power supply because the 4080 is quite power hungry as it is and I don't think my 850 watt psu would be enough to handle a 4080 and a 4060, but overall that's cheaper for me compared to having to basically build a new machine by getting new ram, new motherboard because my old one died that's what forced the update as it was, a new case, a new cooler, a new PSU, and even a new desk because my current desk isn't big enough for 2 PC's on it. I mean, the CPU usage is still being pulled by everything in this scenario, but it would at least help with the bigger part of running out of Vram with the 4080.

It's also a huge shame that they don't seem to be working on this anymore because I would have all of the voice changers I've used, this one has by far been the best. I haven't gotten as clear results with any other as I have this one. Some of that Could come down to how much I worked on refining the voice model I'm currently using, but even before switching it around like that and doing a few different revisions of the same voice, it felt above and beyond other programs I used for the same purpose. I've actually gotten a lot of compliment on how clear the voice is compared to most that they hear. It's actually been enough to fool some people which is not the intent because I go out of my way to let them know this situation that there is a guy behind the model.

Apologies it took so long to respond. I got up and went into a 9 hour stream then had supper, followed by scooping out the yard at 1am lol. Been a busy day.

[Kuuko-fokkusugaru]

Nothing to apology about!
And yeah, an extra gpu could be a great solution. It's just that sometimes isn't that feasible for everyone. Sometimes is hard to fit two of them even if you have space, or you may prefer for your 4080 to have more room to breath. Bu definitely is a good solution.
Some people also prefer to run RVC through Google Colab, so you are running it remotely in Google servers. I tested and it works pretty well, but after 5 minutes it stops working because it's a paid service. No idea how expensive or cheap it is, but in the long run it may not be the best.

Oh, I heard your model and it did sound pretty well. I wish I knew how to train one myself. The one I use now sounds very good and it never glitches or sounds distorted. It's the only one I have found that is this good (paid). The rest always have similar issues where the voice distorts and sounds robotic. I'd love to make my own solely so I can have a good original voice but no idea how to even start.

Good luck with your new setup and let me know if you get that new GPU. Thankfully there are a lot of smaller thin models for the 4060 which would do wonders in your system since it wouldn't take much space and they would be suffice for RVC. Regards!

[kirkyeehee]

I looked into running rvc through Google Collab and when I tried to follow the steps that I found online, it doesn't work anymore There are things that are needed when it's building that are out of date and it can't make or buy those steps anymore. So that isn't even a viable option at this point.

Making a voice model is stupidly easy actually. I was shocked by what little effort it takes, it more takes time than anything. The longest part of it all was creating 100 voice lines that areupward of 10 seconds in length. That took a long time to write out and go through to make sure there was no distortions in them. Other than that, it was just a matter of running the data through an rvc beta setup I found and just let it run overnight to train. There are tons of guides online for it that are still accurate thankfully and they're super easy to follow.

I also have a solution if I were to have just a Secondary gpu attached to my PC, I would get one of those extending cables and mount it outside of the case. With the 4080 those cables stick out far enough that I can't put my side panel on anyway because it bends the power cable in a way that could damage the already finicky cable. So getting or making a bracket for it outside of the case would solve the extra breathing problem. I would have no choice but to do that anyway because I actually have a network card in the 3rd slot and I wouldn't have enough room in the case to fit another card anyway.

I also found that with W Okada specifically, I do not use noise suppression or echo reduction. I've had much better results turning that stuff off. The distortions I get are considerably lessened since I did that and to combat it picking up on breathing as distortions all I do is set the microphone about six inches away from me and off to the side of me so it doesn't pick up on me breathing as easily. I've had it upwards of one foot away and it still picks up just fine.

[Kuuko-fokkusugaru]

That's great. You can also increase the S.Threshold so it doesn't pics low volume noises like breathing. And yeah, with noise suppression activated, there are some details that can be missed. My model is great, it picks breathing noises and converts them just fine but that's it, it's not "my model" itself and just made by someone else who lacks the personality that you can give to one of your own. Also, without noise suppression, it picks my keystrokes so it sounds like I am making funny noises with my mouth XD... If you know of any specific link that helped you the most when creating your model, I'd be very happy so I have a starting point. If not, that's fine. I can research on my own. It's just better if I get a source that may be reliable seeing that you got good results on your own.

The GPU idea, if you don't mind moving it a bit out with a cable, sounds really great. In that way you avoid clutter, get better airflow, and get the job done.