NVIDIA RTX for DAZ Studio
Cornelius
Posts: 131
Hello everyone! Sorry for the vague title but I have a question:
I am about to assemble a new PC to which I would like to add an Nvidia GeForce RTX 30 series graphics card.
I can't spend a lot of money, but at the same time I need a graphic component that gives me some satisfaction.
So, considering an (average) budget to spend, would a 3060 Ti or a 3070 be better?
For example, by IRAY rendering with DAZ Studio of a 4K image consisting of 2/3 human figures + environment, would you notice the difference a lot?
Thanks in advance to all who will reply!
Post edited by Cornelius on
Comments
You won't see much of a diffeence between the two. I would go with the cheapest one since they both have 8 gig of DDR. Heck, I would even look at the 3060 with 12 gig DDR, it might be a tad slower, but with the 12 gig of DDR you have less of a chance of the render dropping to the CPU. Keep in mind the most important factor of a video card for Iray is the amount of DDR it has so that you can fit the whole scene on the card and not have it drop to the CPU.
The benchmark thread can give you more concrete numbers on how well different cards perform, although I didn't see the RTX 3070 on there.
The 30 series is extremely weird when it comes to Iray rendering, as in every prior series, the higher model numbers have just been unambiguously better.
But, for various reasons, the non-Ti 3060 has ended up with a larger VRAM allocation than most of the models above it, with a generous 12 GBs rather than a middling 8 GBs.
Given how important VRAM is to Iray rendering (unless everything fits in VRAM, the GPU cannot be used), this makes the base 3060 (again, not the 3060 Ti) the budget champion of this generation, even beating last generation's 2080 Ti for the job (only being a few percent slower in Iray, but with an extra gig of VRAM, and using considerably less power to do it).
The 3060 Ti, 3070 and 3070 Ti might be worthy of consideration if something like gaming performance is a much larger consideration for you than Iray performance (although the 3060 is certainly not useless as a gaming card - you'll have to have middling settings on the latest games, but it can handle older stuff with ease), but otherwise go for the base 3060.
Yes, in theory, a 3070 would be about 60% faster than a 3060 for scenes that fit into 8 GBs of VRAM, but on any scenes that need 8 to 12 GBs of VRAM (and that's quite a lot, in my experience) the 3060 wins flat out, because the 3070 just can't render those at all.
I forgot where I put the numbers, I should have posted in the benchmarking thread. I saw a huge leap from 1000 series (non-RTX) to the 3000 series. My 3090s have 4-8x the performance of my Titan X Pascals (roughly a 1080 Ti).
2-3 figures plus an environment out of those options I would definitely go for the 12 GB 3060 for Iray rendering. Now say if gaming is most important to you, and Iray rendering not so much, then you might want to look at higher performing cards. There's workarounds to less memory like fiddling with textures manually or a product here like Scene Optimizer (of course, it's easier if you don't have to).
This thread may help you - https://www.daz3d.com/forums/discussion/560986/gpus-approaching-msrp-again-finally#latest
Just curious, because I can't really afford to go all the way with this these days. But would it be very difficult to get a higher level RTX 30xx GPU, and radically underclock it to keep the power use down? Does DS need an extremely stable GPU - which underclicking might affect negatively? By the way, I have no experience in under or over-clocking.
I have an unused new PSU, but it's only rated for 650 Watts, if I remember correctly (It's in storage).
I could get a cheap MB and CPU for now, and maybe some other day go wild on buying the parts I had in mind originally.
I'm tempted to get an RTX 3080 now, if possible for this purpose. I could use onboard video for everything but the rendering.
I haven't rendered in over a year because my current machine gets hot and I need it to remain alive for a hundred other daily activities. I'm probably walking into a hornet's nest here, aren't I? Thanks!
I went for the 3060 for the main reason that it had 12GB of vram, and was a lot cheaper at the time compared to the 3070, 3080 and so on.. And I can tell you I have nearly maxxed out the vram on it, having upto 10 G8 figures, with hair and clothing and the background scene, it came to nearly 90% of the vram being used..
While the 3060ti and the 3070 and 3070ti are faster, they only have 8GB of vram which can be problematic with scene complexity.. The other problem right now is how limited the range of high vram nvidia cards, you have the 3060, the 3080, 3080ti, 3090 and 3090ti the problem there is, is that the 3080 and up are really expensive..
Ironically right now a 16GB Quadro RTX A4000 is cheaper than a 3080 or 3080ti..
The thing about the RTX 3060 12GB is that it has lower power usage than the higher end 30 series cards, so you may be able to fit it into your 650w PSU envelope fairly easily depending on your CPU situation.
And it's a lot cheaper to boot, so you could 'save' that extra money to give the rest of your system some TLC (more ram, etc.), or set it aside for a system upgrade down the road when it becomes time to retire that system.
If you are concerned about heat, the 3 fan versions should be overkill for the 170W envelope, so it should run fairly cool to begin with. More fans good!
So I'd imagine that you could get back to rendering fairly confidently with a RTX 3060 12GB if you were so inclined, without dramatically impacting your pocketbook. Also, with the 40 series cards coming towards the tail end of this year, you could set aside the money you saved not buying a bigger card for one of those cards once they become generally available, i.e. not immediately grabbed by scalpers and then sold at much higher prices.
Also, if your system has an extra PCIe-16 slot, you could combine that RTX 3060 with a second card down the road. I do multi GPU rendering, and it's fairly easy peasy with Daz Studio.
I was using the integrated AMD GPU to drive the Daz non-Iray viewport modes and system desktop for a while as well, keeping my Nvidia card completely set aside for rendering, which worked pretty well until the card eventually failed. The rest of the system still works pretty well, and the 3060 that's supposed to be on the way will eventually replace the dead card.
My quick google search shows these cards needing about 200 watts or less (170ish?), maybe a bit more if it's some uber OC kinda 3060 card, not sure.
If you are just rendering casually anyways, the RTX 3060 at MSRP has a lot of bang for buck. It's slightly less than half as fast as say a 3080 Ti, sure, but it's faster than the 20 series or earlier cards, and for less than $500 MSRP - some MSRP for less than $400 if you can find one for that cheap, probably soon enough as vendors try to clear out their existing stock to make room for the 40 series cards...
Not sure how RTX 3060s react to undervolting, there's probably some info about that via Google maybe. Here's a Reddit thread where a guy talked about dropping his 3060 from 170w to around 140w:
https://www.reddit.com/r/nvidia/comments/ndk38m/rtx_3060_undervolting_is_it_possible/
I was undervolting my Nvidia GPU for rendering fairly routinely in my other system until it died. The way it eventually failed, well I don't think it was an undervolting issue (card just stopped working completely) but I was rendering fairly regularly with that card and other people had similar failure issues with that particular card even during 'normal' operating parameters (i.e. they were not undervolting) so yeah...
In any case, hope this helps!
Sorry, you've all talked about VRAM which is so important, but I forgot to add a little detail:
I don't need the PC I'm preparing to game, but to render.
The graphics card must guarantee me a fairly high load job (considering the CUDA Cores of a graphics card), on a 3060 Ti they are 4864 while on the 3060 (base) they are 3584.
Obviously I rely on the general discourse related to DAZ's Iray rendering engine as almost everyone says that the most important thing for DAZ Studio is the number of CUDA Cores of a graphics card.
Who is "almost everyone"?
If the scene doesn't fit the GPU's VRAM, one might as well use the integrated GPU or some $30 generic one to drive the monitors and save hundreds of dollars or euros.
While 8GB's was enough a year ago, the rapidly increasing trend with new products is more textures/maps with higher resolution images and unfortunately, badly optimized UV's - It doesn't take much for the textures alone to use over 3GB's of VRAM and if one for some reason is using high SubD's for the geometry, the available VRAM (~5GB's on 8GB GPU) is quickly eaten
This will depend on the kind of renders you will routinely be doing. If you render a lot of assets (characters, etc.) together, this can fill up VRAM fairly quickly, which may require workarounds to get the scenes to fit inside of a card's available VRAM.
It won't matter how many CUDA cores you have if the scene doesn't fit in the VRAM and goes CPU only. This is why a lot of us recommend either the RTX 3060 12GB these days, or the 3080 Ti. I am currently making do with 8GB of VRAM (same as the 3060 Ti) and I hit the VRAM wall often enough that the extra 4 GB of VRAM is worth it to me. Of course, if you can afford a 3090... yeah the extra VRAM those have would be very nice...
I had an 11GB 1080 Ti for a bit, but that card eventually died. The extra 3GB of VRAM (over a 1080 8 GB) was rather helpful for the rendering I was doing, and I am hitting the VRAM wall fairly often now with just 8 GB...
40 series cards are supposedly a huge jump in performance over 30 series cards, but we won't see those for a bit yet, tail end of this year. The rumor mill is saying close to 2x performance over 30 series, but 450-600W TDPs for the higher end cards. If this interests you, this is where the strategy of 'tiding yourself over' with a RTX 3060 12 GB isn't a bad strategy. Hopefully you could still use the 3060 card alongside a 40 series card for multi-GPU rendering when the time comes, but until someone has a 40 series card to test multi-GPU rendering with an older card alonside a 40 series card...
This is why spending over $1000 on a card right now is a mixed bag. 40 series isn't too far away, and yeah that rumored bump in performance... but each usage case is different so that's your call.
However, if you are mainly rendering single characters and not too complex scenes, yeah you may be able to get by with 8 GB of VRAM fairly easily, so the extra CUDA core horsepower of the 3060 Ti cards will help in that instance.
It all comes down to your usage case and what you plan to do, and how much time you are willing to spend to 'optimize' scenes to fit inside of your available VRAM as required. I manage with 8 GB, but end up rendering scenes in two passes often enough to find this annoying.
Some people manage with 6 GB or less, it's certainly possible, again it depends on what kind of renders you are doing and how VRAM intensive the assets you plan to use are.
Of course, if you are not using Iray, this discussion is less relevant, but you seem to be focused on Iray rendering so...
And about the rendering speed... I have an RTX 2070 Super 8GB (waiting to be replaced with RTX 3060 12GB) and usually the rendering (95% conversion) takes 10-20 minutes.
The RTX 3060 12GB is about as fast as the 2080 Ti, which was the fastest previous generation card and a bit faster that my current RTX 2070 Super.
Does it matter if the renders are done 10-20% faster?... Not unless one is rendering animations with thousand(s) of images.
Does it matter if the rendering drops to CPU due to having not enough VRAM?... Definitely, as rendering on CPU will take 10-20+ times longer.
For reference:
The RTX 2080 Ti has an OctaneBench score of 345
The RTX 3060 has an OctaneBench score of 285
The RTX 3060 Ti has an OctaneBench score of 368
The RTX 3080 Ti has an OctaneBench score of 644
The RTX 3090 has an OctaneBench score of 653
https://render.otoy.com/octanebench/results.php
OctaneBench can give you a ballpark idea of performance in Daz Studio as well, but of course the Iray Hardware Benchmarking thread elsewhere in this forum section is a bit more relevant.
Very good. Based on these valuable statements I believe I will opt for the 3060, considering the 12 GB VRAM.
For me personally the leap in quality would be big when I consider my current build.
Here is which PC I work with now:
CPU: Intel Xeon E3-1231 v3 (Q2'14)
GPU: Nvidia Quadro M2000 (4GB and 768 CUDA)
ROM: Samsung SSD 860 EVO (SATA)
RAM: 8GB DDR3
This is my current setup which still allows me to render simple scenes with IRAY.
I will strictly listen to your advice. Thank you all!!
Increase your RAM to minimum 32GB's too.
That's an oversimplification which relies on the assumption that in nearly every previous circumstance, higher models with more CUDA cores have also come with at least equal VRAM, if not more.
The reality is that VRAM is completely make-or-break for a card's performance (if the scene doesn't fit, the GPU cannot be used), and is probably the genuinely most important thing for a card, at least until you get above any range you're likely to need to use.
Now, even that's not a completely hard and fast rule, but this highly unusual circumstance of a lower CUDA count model with more VRAM totally turns the old "CUDA is best" assumption on its head. As is, enough stuff can fall into even the 11-12 GB range that, considering purely Iray performance, I would personally say the 12 GB 3060 has the edge over the 11GB 2080 Ti, even despite being a little slower (5-10%). Of course, if gaming is any kind of consideration, the 2080 Ti kicks the 3060's butt, but that's a completely different complication.
As is, the 3060 coming with a large VRAM allocation on an otherwise mid-range card (albeit a mid-range card that can go toe-to-toe with last generation's flagship) has created an absolute budget workhorse. It's a shame that supply of the 30 series cards has been comparatively limited/expensive, because this card alone should have been a renaissance for Daz Studio users. An RTX card with 12 GB of VRAM at what was supposed to be a $329 MSRP is ridiculous compared to anything else anywhere near that price point previously.
Certainly. But at the moment I can't afford it.
I'm saving money to buy the graphics card. Here are some components I currently own:
CPU: Intel Core i7-10700F
RAM: 16GB (2x8) DDR4 2666MHz
ROM: Samsung 970 EVO Plus
You just saved enough to get 32GB's of RAM by choosing the 3060 12GB instead of 3060Ti or 3070
Yes, take the 3060 with the 12 GB!
Rumors are, that in order to take full advantage of the VRAM on a graphics card, you need 3x as much computer RAM. Although I don't know why. Anybody?
So, if true, that would be 36GB for a 12GB VRAM card. I'm using an RTX-3060 but get by just fine with 32GB for my typical scenes which include up to about four M4 or Gn or G2 figures and just a backdrop or a simple set, without complex lighting systems. And a 3060 would probably work with just 8GB of computer RAM but I'm not sure how the limitation would present itself.
Because all the texture files have to be decompressed (out of JPG/PNG/etc format*) into effectively BMP format in system RAM so that the CPU can build the render to send to the GPU, but are then (mostly) recompressed into the VRAM using Iray's own compression algorithm, so less VRAM is used than the system RAM needed for initialisation. Also, things like displacement maps have to be used for initalisation, but aren't needed on the GPU, as by then it's baked into the geometry sent to the card.
* Because it all gets decompressed and recompressed into another format, the original on-disk format is irrelevant to how much VRAM is eventually used, although a lossy format will retain any compression artefacts that were introduced.
And, to be honest, I have found that DS can sometimes eat four times the VRAM in system RAM, depending on scene composition (although possibly some of this can be handled by page file). As is, I built my RTX 3060 based system with 64 GBs of system RAM to have decent overhead, as well as hopefully a bit of future proofing for when Nvidia finally bother to do a 16GB card (the 16 GB 3070 Ti was teased so often, but is now apparently completely dead in order for Nvidia to focus on the 40 series).
But yes, I'd definitely get at least 32GBs of system RAM for the 3060. I'd missed that it was being built as a 16 GB system. 32 GB should usually be enough to get out most of what the 3060 can do, and the cheaper card should free up some budget compared to the 3070.
Wonderful explanation!
Thanks tj_1ca9500b,
These are good points about the 3060, which I've tried following in these threads. But I still can't justify the 3584 CUDAs, even if I were to double that with another RTX 3060 or similar later.
I'm thinking of render time. None of my ideas would require large scenes. All are simple & even fit in my GTX 1060 with 6 GB! I've tried some of them, loaded but not fully rendered. Any more than that and my machine will go boom:-)
My goal is render speed and image quality. You folks here are in a different world from mine. I'm just a story-teller with no need for huge scenes. Plus there's another issue about time - I'm old and have no interest in planning for 'next year' - life's too short. Surprises come. Life is to enjoy now.
Thanks for the link to undervolting the 3060! So the higher the power needs of a card, the more underclocking can make a noticeable difference. Therefore, that delicious 3080 Ti may be my best target!
The question remaining is, does DS/Iray care about underclocking & the possible instability?
Thank you again for the info, the 'boost.' And, MB, CPU and system RAM prices aren't quite as high as I'd feared, so this is beginning to look doable.
The render quality does not change, no matter what you use.
The benchmark thread can give you an idea, how the different GPU's compare against each other.
https://www.daz3d.com/forums/discussion/341041/daz-studio-iray-rendering-hardware-benchmarking/p1
The 3060 results are starting around here;
https://www.daz3d.com/forums/discussion/comment/6830466/#Comment_6830466
The 3060 does 8.1-8.3 iterations per second. There were no 1060's, but 1050Ti does one iteration per second and 1070 does 2.5 iterations per second, so in theory the 3060 should be about four times faster than your current card and the 3060Ti with 10.5 iterations per second would be about five times faster than your current one.
Some cards where I've messed with their clocks have immediately crashed Iray, even with small changes. Other times, it's been extremely insensitive to much larger changes. I suspect that underclocking is less likely to result in an instability, but my advice is still mostly that you can't rely on being able to mess with clock speeds to bring power into limits.
I'm also inclined to say that a lot of 30 series cards are noted to need *more* than their recommended PSUs to remain stable. But if you're already buying a $1000 card, then you're better off buying a new, properly rated PSU and selling on your old one (decent brands hold their value well, so if it is a respectable brand you probably won't have lost much on it).
See my followup post here discussing 8 (and 6) GB of VRAM options:
https://www.daz3d.com/forums/discussion/comment/7425771/#Comment_7425771
As I noted, I'm making do with 8GB of VRAM now, and for less intensive scenes you can usually make do, plus there's Scene optimizer and rendering in multiple passes, etc. if needed.
As I noted in the comment, in this instance the 3060 Ti is indeed around 20% faster (see my post here about OctaneBench scores)
https://www.daz3d.com/forums/discussion/comment/7425851/#Comment_7425851
OctaneBench is a decent benchmark to give you a 'ballpark' idea r.e. rendering performance, but of course the Daz Iray Hardware Benchmarking thread has more relevant results, albiet in 'stereo instructions' format.
So if you don't mind making do with less VRAM, yeah 3060 Ti. The 3080 Ti is a rather impressive card if you can afford it, and it has 12GB of VRAM to boot!
As for undervolting, as I noted before, I was able to do an undervolt on my 1080 Ti, but I don't have my 3060 yet (being shipped atm). I'm not really planning on undervolting said card in my current usage case, but I may do so once I add it to the system that has the dead 1080 Ti.
The thing about undervolting is that you can tell pretty quickly if your undervolt is too aggressive or not. The render will simply not complete, at which point you bump up the frequency or voltage and try again. Once you find your 'sweet spot' you should be good to go. I was using an overclocking interface for my 1080 Ti, and the only issue I had was that occasionally the 'saved' settings wouldn't 'take' and I'd be back to 'default' performance. As my goal was to keep card temps down, I could spot this fairly quickly if the temps climbed over 80c... My target was 70c, as I was hoping to stretch out the life of the card, but the card failed eventually anyways. As I noted, the card in question had several other reported failures in Newegg reviews and such, at 'stock' settings, so I don't think my undervolt was responsible.
If you are planning on a 3080 Ti though, yeah I'd spring for a higher wattage PSU to go with it just because. You MAY be able to make do with a 650w, but that's cutting things a bit fine depending on how often the card decides to draw extra wattage. Power supplies aren't that expensive to begin with, unless you want to buy say a 1600w one... I'm seeing a few of the 1600w ones in the $150-200 range, and most of the 750w supplies are less than $100. I'm even seeing some 800+ watt power supplies in the $100 range over at Newegg. I'm sure there are some decent deals on Amazon.
If you plan to shell out over $1000 on a 3080 Ti, yeah spending an extra $100 on a power supply seems trivial, but of course you could try to make do with the 650w power supply first and see how that goes. Depending on what else is going on power usage wise in your system, i.e. how much power your CPU, etc. want, it might be possible, but it sounds a bit tight to me. Instability can result if you don't have enough power overhead though, keep that in mind.
Hope this helps!
Very helpful information here regarding GPUs.. I too am building a new computer and just so happened to go with the RTX 3060 picking it up for 600$ a few months ago, which was a great price compared to others at the time. I almost sent it back thinking the 3070 Ti the way to go, but after reading here glad I didn't. The price point being a huge deal for me considering my limited budget--I just didn't want to spend $1200 - $1500 (as much as the entire build) for a GPU and poured my budget money into extra RAM (128) & the PSU-PSU being almost paid off. I did read it is better to buy ram all in one kit (all the same ram batch--ram also was hard to find and out of stock a lot). I will never skimp on the PSU quality or voltage again...I had one computer rendered completely unusable and unfixable due to a lower end PSU.....and, with my current build, I initially went cheaper (Gold rated) + lower watts---it died after a year--thank goodness it didn't fry my mobo. Long story short I went with a massive 1300 watt platinum rated PSU this time around. I figured this leaves room for extra watts under load and, a bit of future proof if I should ever save up for the 4000 series--which I am sure are PSU voltage hogs. The 3060 minimum requirement is 600 watts. I am hoping for once I can start doing some complex landscaping scenes without thrashing & crashes .. The only bummer I am dealing with now is the desire to dump windows 10/11(another memory hog), but I am guessing a linux version of DAZ studio is a long ways off, if at all...sigh
-By render quality, I just meant I don't ever want to go back to 3DL, and especially not Filament in its current state.
-The Benchmark main chart: Yes. I check it often for inspiration:-)
But here's something I didn't notice until now because my Firefox doesn't show section "3. Benchmark Results", which is just under the main chart. The data is missing. Only the headers for each card show. Am I alone in this? Just checked Chromium - It has the same problem.
So I saved the page to edit it in Kompozer. Cleaning it up revealed the missing data. Could this be correct?:
-RTX 3060Ti: Its 10.648 iteration rate took 2 minutes 53.70 seconds
-RTX 3080: Its 12.062 iteration rate took 2 minutes 38.53 seconds
The 3060 Ti has 4864 CUDAs. The 3080 has 8960 CUDAs. How can a difference of 4096 CUDAs between the two cards result in only 15.17 seconds difference in render times?? I'm sorry if I've messed up something here. My mind is a bit swirling from a day of research. Not used to that!
If those statistics above are correct, tho, I'll go for the 3060 Ti, maybe two! ... Both underclocked :-)
A very quick study of underclocking today did give me the impression that it's not as delicate a change as overclocking can be. I could try it by working down in steps. And from re-evaluating my thoughts about building a renderer on slim margins, I agree with you that I had better treat this with respect and not cut corners too sharply. It's typical for me to be stingy and regret it later down the line. I appreciate your insight!
Great info!
So the OctaneBench scores say that the 3080 Ti renders a bit less than twice as fast as the 3060 Ti. So with some layering of 3DL for backgrounds, & more use of billboards and HDRIs, I'm thinking this is looking better than I imagined! Maybe even, animation in the future?
And thank you for describing your experiences with undervolting. It helps support what I studied earlier today. This looks good! Much to sift through.
But really, if the GPU problem is turning around, now is the time for people to jump while we have the chance. I'll dig into my storage and see what I purchased two years ago before everything crashed, and pick up from there. Cheers!