1x RTX 4090 or 2x RTX 3090 and hardware questions
charliecharles
Posts: 6
Hi everyone, this question probably has been asked a million times and first I got say I'm already decided on getting the RTX 4090, but have a few questions to make sure I'm not overlooking something.
I'm currently rendering on a RTX 3080 Ti and most of the time it works just fine, but lately it started to annoy me that some of my scenes are not fitting in the vram. I know I can optimize the textures but I started feeling soon I will really need more VRAM. Aside from VRAM, from what I read 3080 Ti and 3090 have almost the same rendering performance.
Like I mentioned at the start, I'm set on getting the 4090. Just want to ask if there's any benefit in getting 2 RTX 3090 instead. I know it can use NVLink for Memory Pooling but I've read it's glitch in DAZ, is that true? And I don't see myself needing 48gb vram soon (can't always tell for sure but not likely).
Then what I want to ask is: Before I go ahead and buy it, am I missing/not considering any other possible reason for having 2x RTX 3090 over 1x RTX 4090? (I will keep the 3080 Ti together for smaller scenes since I have the psu)
Then something else that maybe I should make another post: I was more enthusiastic about hardware decades ago, I stopped following the market and got back for a little while when Ryzen was launched, read a thing here and there and then stopped following completely by Ryzen 3rd gen. Although from time to time I do keep up with some stuff, I don't know in-depth the new cpu technologies, so yeah, bottomline is I'm outdated about CPU and motherboards.
Which leads me to another question: DAZ itself is awfully slow (not about rendering), don't get me wrong I love it, I'm sure there are already few things that can be done to improve the experience a little bit (yet another question to look up later), but one can see it's a badly optimized single-core software.
I'm currenly using the Ryzen 3700X, suppose I upgrade to a i9-13900K, will I see noticeable performance gains, like more responsivess? Not talking about numbers or other software but actual improvement in DAZ experience itself.
Thank you!
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
Well, two 3090s should be a little faster than a single 4090, so speed would be primary the reason. But you would need enough space and a lot more power to run them. One thing with Iray is that it scales pretty well with multiple GPUs, not 100%, but pretty high.
Daz Studio's big problem is that it is largely single threaded, so throwing tons of CPU cores at it doesn't really do anything. Supposedly Daz Studio 5 will be releasing **soon**, but has no date and Daz is super tight lipped about it. They claimed it was coming soon over 2 years ago, so who knows when it will actually come out. Hopefully DS5 can take advantage of more cores and thus run better on many CPUs. However, we don't even know if this is the case or not, again because of how tight lipped Daz is. We know nothing about DS5.
What we do know is single threading is more important for DS4, a CPU that has better single threading will provide the best uplift. Even so, I don't know how much better DS can truly run. It feels to me like it has so many different software bottlenecks that hardware can only help so much. So I wouldn't put too much priority on it at this time if money is any factor. A 3700x is still a fine modern CPU. A 13900k may indeed run DS better because it is faster single threaded, but how much better is the key question. I don't know if it is worth building a whole new machine for. I'd consider waiting a bit longer for the 15000 series or Ryzen 8000 because some crazy things are in the pipeline for future CPUs (like AI cores could be coming which could be capable of exponentially speeding up some tasks.)
If loading stuff is an issue, check out Turbo Loader, as reducing your morph counts on a figure goes a long way to improving load times. This is also something that Daz 4.21 improved, the recent 4.21 releases helped loading a lot, which further proves that this is a software bottleneck rather than hardware. If you don't have 4.21, get the BETA instead of updating the main DS because it may also break some items you have.
I also discovered that some geographs can cause scenes to take much longer to load. I don't know why. DS is a very old program now. It may get updates and changes, but its base is pretty old.
From memory my 3090 used to do about 320w while rendering, my 4090 does about 290. Dual 3090s need serious cooling.
While using the same 4090 I swapped from a 10700k with ddr4 that is almost identicle in performance to your cpu to a 13700k with ddr5.
The most noticeable difference is the time it takes for a render to begin. The scene in the benchmark thread took 6 seconds or so to begin rendering on the old system, the new one takes less than a second. If you're doing a lot of single figure test renders that may only take 20 seconds its annoying if the pre render phase seems to take nearly as long. For cpu heavy stuff like ultrascenery the new system is much more pleasant.
For general interface responsiveness, loading, saving and posing you'd be hard pressed to tell the difference unless you had both running side by side. You still need to be mindful of content used and settings.
If you're also a gamer the 10700k was a noticeable bottleneck for a 4090 and the main reason I wanted to upgrade.
Thank you for the reply outrider42, it was insightful! I do have the psu and space for running two 3090 or 4090 + 3080 Ti if needed (not sure how the temps would be with two monster inside the case on air). Didn't know two 3090 would actually be faster than a single 4090 though, I somehow missed that. Remember reading somewhere a 4090 was twice as fast as a 3090 Ti, I'm probably not remembering correct. Better read the benchmark thread again.
Yeah, money is a factor, my priority is the GPU and you very well described my question
Perfectly put. Basically doing a cpu upgrade in a build for DAZ studio is most likely a waste of money then. I don't have high hopes of multi thread in DS5 or none at all. I do have the 4.21 BETA, not the most recent release probably, I did some tests and for me it didn't improve scene loading by much, it's actually worse because the rendering was slower than the loading gains. Small differences in both though from what I remember but mayeb I should run tests again to confirm. Will remove the current BETA and try the latest build later.
Thank you!
Oh, so you had both that's a great input, thank you! I get what you mean about the pre render phase taking too long. Some times I already face that while doing said quick tests at smaller resolutions, very annoying indeed.
I'm not a gamer unfortunately, I still could benefit from the extra CPU power in other software but the 3700X does the job so far. I mess with blender sometimes (which is blazing fast) but mostly with DAZ because I just know it better than I know blender.
I guess it's fair to say that if it was limited about DS, the gains are minimum and won't justify the cost of moving to a newer platform then? (New high end motherboard and ddr5 ram)
It is close. In the benchmark, my 3090 is hitting around 16.7 iterations per second, while 4090s (on the same version) are hitting 28.5. So two 3090s should get you to 30 to 31 or maybe 32 since they don't scale perfectly. Two 3090s should be around 3 to 4 iterations faster than one 4090, at least in this scene. That will vary in other scenes, but the two 3090s should always come out faster.
However, it is important to note the 4000 series is not being fully utilized by Iray, and the Iray devs have acknowledged this. It is probably a big reason why the 4090 only uses 290 Watts when rendering Iray. I always thought it was fishy when people started saying the 4090 was using like 280 Watts to render. The dev team has already posted updates claiming to fix this, but Daz Studio has not yet updated to any of them. According to Richard, Daz may be waiting to update because of how 4.20 broke some things, so they are waiting for a "major update for Studio" before including a new Iray. I rather think that is silly, the thing they broke have already been broken, so I don't believe there is a point in holding back. Especially given that Iray claims to improve and even fix a number of things...perhaps some things that it broke previously have been fixed.
At any rate, the 4090 and 4000 series as a whole will get faster...eventually. The dev team also said basically all GPUs see performance boosts in the new Iray, but Lovelace sees the most.
It could be any number of things slowing down your load times. I have found that some OOT hairs with their custom OOT shaders murder my load times. The way you save things can cause them to load slow, like if you create some custom shapes or things, but do not properly save them as new morphs or props, this can hurt loading as well. I have characters saved as scene subsets, and some load in a couple seconds while some take at least a minute. So products like Turbo Loader will help the Genesis model load faster, but it can't help with these other things that can also kill load times.
If you render indoor scenes, sometimes 4.21 can actually render faster with Guided Sampling enabled. But sometimes it doesn't and can take way longer. I haven't figured out why. But again, the new Iray 2023 which Daz Studio doesn't have promises to improve this.
This is the direct quote from the Iray Dev Team from June 7. I will also note that they have already updated Iray a further 4 times and are now shipping 2023.0.4. We just have to wait on Daz to put it in. Or if you have Nvidia Omniverse, you can try it out now, as it has the most up to date Iray at all times. On that note, the USD format that Omniverse uses is something that may be coming to DS5, that is the one thing we have seen posted.
Iray 2023.0.0 final
..was just released.
Major improvements over the beta:
Performance, performance, performance. :)
We mostly got back all the perf loss we introduced in the beta when running the new caustic sampler 2.0 (compared to cs 1.0 in 2022.1.X), plus we optimized general performance on the Ada Lovelace GPU architecture, and Iray is now faster when using rather complicated MDL materials (i.e. MDL JIT path is triggered).
Then a lot of performance optimizations all over the place, e.g. when using fibers/hair, rounded corners or cutouts.
In addition there is also one additional new major feature over the beta: Support for the new Mie phase function for volumes, introduced along with MDL 1.8.X.
On top we have dozens of important bug fixes, as usual.
The new release will also be available soon within Omniverse, of course, via the RTX Accurate rendering mode.
The choice of buying a new 4090 is good enough, plus your 3080ti. You don't need two 3090s with Nvlink though it'll be appr. 15 - 20% faster than a single 4090. We've had many experiment and tests with such a rig. Tiny performance + with the link and there's issues with mem pooling when handling textures and ray tracing. Seemingly Daz stopped optimization and bug fixing on Nvlink Peer options starting from a certain DS version , as well as Nvidia, esp. on the consumer graphics card starting from 40 series.
As for CPU, you don't have to be in a hurry to upgrade your Ryzen. As for DS, GPU is the king, RAM is the queen, well CPU is more or less a bishop.
As for pre-render time, that really depends... maily the complexity of your scene. The more VRAM it's gonna consume, the longer it's gonna take for pre-render. Single card is faster than dual or multiple cards. High level SubD mesh and HD texutres consume much more time. If you have complex SBH or fur in the scene, generating high level of line tesellation consumes much more time... so on and so forth. In general, all these 'preparation' has to be done in pre-render phase before your scene is fed into GPU.
Its up to you depending on if you want faster renders with the two RTX 3090 cards. I use both an RTX 3090 Kingpin and an RTX A6000 and I can tell you from experience you'll be dealing with a lot of heat. My advice is that if you're going to go with 2x 3090 cards, get a case that can support two 360mm radiators for water-cooled 3090's. I have a Lian Li O11 Dynamic XL Full Tower Case with the radiator mounted in the top(exhaust) + 3 bottom intakes + 3 side-front intakes + 1 rear exhaust. The A6000 gets up to 80-85c since its a blower card, but my 3090 stays between 45-58c depending on which VRMs are being utilized. Needless to say, rendering will dump a lot of heat into the room so an AC unit is a must. In the winter, my room will get up to 80+F while rendering if I don't have the AC on. You'll most likely want to get 128Gb RAM with room to upgrade, so that means a workstation/professional grade motherboard.
Imo, if you don't need the increased render speed, a single 4090 will do and will be much cheaper since you won't need as much RAM + huge case + professional MB.
Yes but that temperature comes from running your dual cards for rendering at the same time. I have an RTX A6000 + 2 Quadro RTX 6000. While rendering with 3 cards (full load), A6000's GPU temp is around 86, Hot Spot 92. If rendering only with A6000, GPU 70, Hot Spot 76.
Thanks everyone for your input, you people are awesome! So as I thought the gains in updating CPU in DAZ are not worth it at least not for now, definitely crossing that off my list at the moment. This year I will upgrade the GPU and if the new upcoming CPU are a true breakthrough in all my other activities I will upgrade to ddr5 next year.
The best way to improve DAZ experience is like you guys mentioned through 3rd-party tools and good practices of managing/optimizing the textures before it becomes too much to handle at once.
This is quite interesting, it's rare but sometimes I do create things with geometry or use DForms and save them just like that. Will look into how it can be improved.
Now I have all the information I need (and I was indeed overlooking stuff) regarding DAZ and rendering times to help in my decision! Guess it was worth asking, thank you all again! :)
It's a nice to have, not a necessity and the upgrade cost is pretty steep. As you've said yourself tools and workflow are the way forward. Some bits of DS are just clunky, an upgrade to slightly less clunky is still pretty clunky.
One more note that could be a big deal. A 4080 Super is coming, which will be between a 4080 and 4090. Some rumors speculated it might sport 20gb, but it looks more likely to have 16gb like the regular 4080. There looks to be at least one more Super, a 4070 Super and possibly and an absurdly named 4070ti Super (that's a mouthful). What makes the 4070ti Super interesting is that it may pack 16gb of VRAM and perform almost like the original 4080.
But the juicy part is that Nvidia might decide to release these at more competitive prices. The 4080 Super could be closer to $1000. Obviously any 4070/ti Super will be less than that, possibly in the $800 range. Getting near 4080 level performance for $800 would be quite a nice change from the $1200 the 4080 was released at.
However, we do not have a time frame. I would expect it to be sooner rather than later, as the 4080 Super would be the fastest GPU China could buy once the 4090 exports are banned.
If these happen, then that might change your outlook a bit. Unless you are really in for the 24gb of VRAM that the 4090 has, a 4080 Super with 16 or possibly 20 might be tempting if the price is right. Sadly 4090s have actually gone up in price a bit recently due to the China export ban. It looks like there is a run on Chinese buyers trying to scoop up 4090s before this ban takes effect in November. So until November the 4090 is going to be both harder to find in stock and cost more than it already does. Once this buying spree cools down the 4090s should drop back down again, and might even drop below MSRP since China would no longer be a market for them.
On top of all this, the 5000 series will likely release in 2025. It was originally looking to come out in the usual 2 year cadence, which would be late 2024, but all indicators are that it is being pushed back to focus on AI markets. But this could change, too. If the AI market takes a sudden drop that might push Nvidia to switch gears and get the 5000 series out faster. For what is worth, early rumors are saying the 5090 may get 32gb of VRAM, and the 5080 might get 24 or 20. The 5090 getting 32gb would be a by product of the AI market, which wants more VRAM like we do. The professional series always doubles the VRAM of the gaming series, so this would mean the pro GPU would sport 64gb. That is why we would get 32gb on a gaming card, as they use the same chips and board designs.