Computer Spec
Patroklos
Posts: 533
in The Commons
Hello, I have an Apple mini M1 8gb ram. Its a lovely machine but it renders pretty slowly which is frustrating. Also the memory is not large enough, think I would go for 32gb.
What sort of PC with Nvidia card do I need to see a maybe 10 fold increase in rendering speed?
Thanks for any help offered.
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
Depends on budget, but any Nvidia 30xx or 40xx GPU with at least 12 gb VRAM will render vastly faster than what you have now
I'm in the same boat. Long time Mac person now looking to get a PC that will render animation at blazing speed. My trusty old Mac with a GTX 980 is just not cutting it. I know nothing about PC's though. I assume we're talking a tower, not a laptop, and with room for more than one GPU?
I'm also looking into the Boost cloud rendering service.
Personally, I consider a Nvida graphics card with 12GB of memory in a PC with 32GB of system memory and 2TB of storage to be the minimum. More video memory would be better if you can afford it. A PC with the above specs will be much faster than what you are using now.
If your budget will run to it, I would suggest a tower PC with two GPUs. A relatively modest one to drive the screen and a much bigger and more powerful one, as specified above, for your Iray renders.
Cheers,
Alex.
I think the common 'consumer' models already have an integrated GPU that can be used to drive the monitors if one wishes to do that.
Personally, I would never choose a motherboard with integrated GPU, but that's just me with my own ideas about how a computer should be built.
One needs a nVidia RTX GPU with 12GB's of VRAM, minimum 32GB's of RAM, a second storage drive (in additon to one holding the OS). 500GB's to 1TB is enough for the OS-drive, but the second drive should be something like 4TB's (external USB drives are cheap)
One should check that the PSU (power supply unit) is big enough, in my computer 750Watts is good for RTX 3060 12GB, i7 and 15 storage drives
The rest isn't that significant, unless one goes with some cheap all in one package with non-standard components.
I found 1 M1 benchmark in the benchmark thread. It got 0.315 iterations per second. It will be pathetically easy to beat 10x that. Almost any modern Nvidia GPU will be an order of magnitude faster. My 3090 is about, oh, 60 times faster. Sixty. It is an absurd number. The person who ran the M1 on the bench took 1.5 HOURS, but my 3090 does it in 1.5 MINUTES, and the 3090 is last generation now, the 4090 is even faster. That M1 bench is a bit old, so you can test it for yourself if you have around 1.5 hours to spare.
https://www.daz3d.com/forums/discussion/comment/7210726/#Comment_7210726
It is very important to note that the VRAM requirements depend entirely on you. How complex are your scenes? VRAM is important, and finding out you bought a card that doesn't have enough VRAM for your goals sucks. However, that doesn't mean you need run out and get a 4090, either. The typical RAM usage you see can help a little. In general you can expect RAM use to be around 2, 3 and even up to 5 times the VRAM use, but the actual number can vary wildly. So if you are typically using less than 24GB of RAM per scene, then a 12GB card is a very safe bet to suit your needs. You might even be fine with 8, but again, it would suck to be wrong. 12 would also give you a little room to grow as well. Because when you start rendering faster, you might start wanting to do more in your scenes, too. It can be a vicious cycle.
As for speed, anybody wanting a direct comparison can just run a benchmark from the thread linked in my sig and see for themselves. As I already said, any modern Nvidia GPU is going to be 10X+ faster with ease. The numbers can vary between versions of Daz, so try to compare like for like when possible, and the thread is a bit of a mess, but that's how it is. It is still by far the best such asset specific to Iray you will find on the net. If you do run the bench, please post your results as instructed in the first post as it can help others.
Then it is a matter of finding the card with the best combo of speed and memory that fits what you want to do. At this point you pretty much should stick to the 3000 and 4000 series, though coming from Apple anything is going to be much faster. But the last 2 generations are such big improvements than what came before that is hard to recommend anything older (for anyone thinking of building with used parts). With the exception of the top end 4090 and their overall fantastic power draw, the 4000 series is kind of a let down IMO. However it does offer more 12 and 16GB models. The 4080 Super (and non Super) looks to be skippable, as the 4070ti Super is extremely close and offers the same 16GB for less money. The 4060ti 16GB version is the cheapest 16GB model. The 3060 12GB is the cheapest 12GB model. Now keep in mind both of these have versions with smaller VRAM capacities, pay attention to what you are getting if you do buy one. The regular 4070 is a fair bit faster than the 3060, and the 4070 is the cheapest 12GB card in the 4000 line.
The 3060 12GB is relatively cheap in general, and a decent starter card for this. It also roughly 20 times or more faster than the M1, and it is considered low end.
I'll try to keep this breif and bulleted. My assumption is that you'll end up bying a pre-made machine, but there is a joy in building your own PC these days. "Lego for adults" was the term that came up a lot during lockdown when people new this area 'ascended' to PC self-build-ownership for the first time.
You must know the motherboard and power supply of the machine you are buying. This is made difficult due to the shop/manufacturer main page rarely telling you this. Why is knowing this important? Sometimes you cannot replace the power supply on big brand (Dell/HP) machines. Even if you can with small brand machines, you need to know what power supply it has as as some point in the future you will want to change the graphics card it comes with.
*The power supply must come with with an 8 pin power cable. This is needed if you want to replace whatever graphics card the machine comes with. Power hungry cards needing more than 220Wats will need 2 of these power cables.
* Powewer supply efficiency: If you are using it for many hours per week, 'Gold' class power efficiency is recommended a) to save money over the next 10 years and b) its an general indication that the power supply is decent to begin with. A 'bronze' (or much worse still '80+ base' spec) class supply will potentially vary more in quality (they can still be fine).
* 4 memory slots on the motherboard: Whatever memry size you decide to go for (a topic for another post), you'll inevitebly want more in 2 to 5 years time. If you have extra spare memory sockets on your mothrboard, you wont have to thow out the ram the machine comes with, you can just add more. Again, this can be difficult to find out when looking at a machines initial product advertisement page. But you should make the effort to dig deeper.
* Platform decision - perhaps your biggest decision: There are 3 to chose from currently with Intel expected to bring out a new replacement motherboard socket in the next 6 months. Intel LGA 1700 (eg for i5-12400), AMD AM4 (eg for Ryzen 5600x), AM5 (eg for Ryzen 7600x). You'll need to make this choice when buying a machine. However I think that this would need to be discussed in a separate post.
Thanks
I recommend you find a used 3090 for best price to performance. Mainly for the 24GB VRAM. Having that extra VRAM is all the difference. Nothing else matters if your scenes run out of VRAM and forces the render to fallback on the CPU. Spending 2.5x more for a 4090 is not going to be worth the price which has the same VRAM but a little better performance. I use a 3090 and work with massive scenes, 3-4 G9 characters (subD 3), 4K resolution output and texture compression threshold to 8192 and it renders these scenes insanely fast. When I used to be on a 11GB VRAM GPU, I was constantly failing to render the scene on my GPU and it was very frustrating to work with.
So in short, prioritize an NVIDIA GPU with high VRAM and build around that.
RTX 3060 (12 GB)
Interesting about VRAM thanks
I love my 3090. And it more than half paid for itself.
However... People brand new to PC ownership tend not to dive in right at the deep end of the 2nd used market. And a 3090 is the deep end. Odds are that its an ex mining. And of those odds are that its memory, and more specifically the memory heat pads, got badly abused during that time.
Additionally, new to PC people tend to buy pre-builts (or if building their own, use only new parts). And of the types of pre-builts that don't already come with a gfx card will never have a power supply capable of powering up a 3090. Never. You might say, sell the card it comes with. Problem: warranties. [and again the likelyhood of a new-to-pc person dealing with the 2nd hand market again]
I compared the GTX 1070 8GB, RTX 2070 Super 8GB, RTX 4060 Ti 16GB, and RTX 4080 Super 16GB at this link:
https://www.daz3d.com/forums/discussion/674576/super-fun-happy-computer-upgrade-time#latest
How much complex geometry (incl. SubD) and large texture sizes will determine what you can run on the card. Learning to optimize your scene and using PA tools available at the DAZ store will help fit more scene elements in less VRAM.
First figure out your budget then determine what you can buy for that using the recommendations mentioned here and in the thread I listed.
Sites like https://gpu.userbenchmark.com/ can give a sense of the performance difference between Nvidia cards. It is aimed a gaming performance but I have found it useful to see "roughly" what kind of rendering performance can be achieved since I know what my current card can do.
https://gpu.userbenchmark.com/Compare/Nvidia-RTX-4080-S-Super-vs-Nvidia-RTX-4060-Ti/4156vs4149
If all you are doing is DAZ rendering then all you need is a CPU, motherboard, and enough system RAM to support the amount of VRAM on your video card because it's all about the Nvidia baby. I demonstrate this point in the link I referenced at the beginning of this comment.
Thanks very much to all responders.
I have had several machines custom built way back, but more recently have fallen for Apple stuff.
Plenty to think about
DAZ Studio runs on nVidia Iray technology, but nobody here is talking about CUDAs.
My understanding is that in addition to what has been already mentioned, the number of CUDAs in a nVidia GPU determines how quickly the render will be finished. If you don't have enough VRAM to hold the scene, you will have problems (to be sure), but if you don't have enough CUDAs, your render might take more time than you want to spend waiting for it to finish.
I always do a comparision of GPUs by VRAM and CUDAs when determining which card to buy. To this end, a comparision test with render times given a certain number of CUDAs is invaluable. As always, the more of everything is better, especially if you are not in the habit (or financial position) to buy new stuff to accommodate the latest power-hungry requirements of the newest software.
The CUDA cores cannot be compared between generations
As for the rendering speed... Depends on what one is doing, stills or animations.
This was rendered on RTX 3060 12GB in 5 minutes and 5 seconds without any optimisation (the whole town is there unhidden)
I agree with those who said it depends on what are you rendering; size of the scene, how many characters, texture quality, resolution, etc. I'm gonna add that if you have money to burn and if you go for the RTX 3090 (highly recommended) you should also invest in RAM, at least 64 GB. I've had scenes swallowing +40 GB of RAM. I used to have 32GB then 24GB and couldn't render some of my scenes. Having a high ram also makes the user interface (loading characters, posing and whatnot) smoother and it doesn't stutter as much.
The benchmark thread is the best way to compare GPUs, as CUDA cores change each generation. Normally, the cores do get a little better with each series, but not always. The 3000 series changed how CUDA was used, hard to explain here, but they basically split the compute cluster and counted that as a CUDA core, which meant that the CUDA counts on 3000 series were much higher than initially expected. On one level it could be correct to count them that way with how the processes work. At any rate, while core counts were drastically higher than the 2000 series, the actually performance was not in line with that. Clock speeds are also a big factor in how fast the CUDA cores actually run. The professional GPUs are often slower than gaming versions for this reason, even though pro cards often have more CUDA cores, they are clocked much slower for maximum stability and efficiency.
But CUDA cores are but one part of Iray, the ray tracing cores are huge factor in why newer cards have become so much faster. If we didn't have RT cores the renders would be a lot slower. RT core impact can vary, however. They are not as straightforward as CUDA is. If you render in pure CUDA like with the 1000 series and prior, the performance was extremely predictable. If you made a scene that rendered twice as fast with one card versus another, like say the 1080ti and 1070, then you could bank on pretty much EVERY scene behaving the same way, the 1080ti will always be twice as fast.
But ray tracing changes that. The performance gap between RT and non RT widens as the scene geometry increases. What this means is that the more complex you scene gets, the bigger a difference the RT cores make. If anybody remembers the "RTX Show Me the Power" thread, that was a benchmark scene that gave a taste of the potential of RT in Iray. In the benchmark scene we have been using, the 2080ti was around double the speed of a the 1080ti. And that sounds pretty good for one generation, right? Well in this scene using strand hair. In this scene the 1080ti took 130 seconds, however the 2080ti was able to do it in just 29 seconds. That is a much wider gap in performance compared to the benchmark thread we use.
That means you can expect the 2080ti to be at least 2 times faster than the 1080ti, but it can get even faster in comparison as the scene gets more complex. Imagine a scene with multiple strand hairs, and complex geometry everywhere, and you can easily achieve a situation like what Nvidia advertised back when they launched Turing and RTX. I don't know if people recall, but Nvidia claimed a 6x increase in ray tracing with Turing. You can find people in other places who claimed they could see even 9x speed increases over the fastest Pascal with Turing.
Ray tracing performance has expanded each generation. The gains are not as massive as that first jump to RTX with Turing, but they are gaining faster compared to the CUDA performance improvements. So while Ampere was not as huge a leap as Turing in RT, it was still a big jump and the 3000 cards pretty stomp Turing across the board. And that makes comparing CUDA counts even less important as time goes on. The only way to find performance is to properly test them. I do wish there were more benchmark scenes that could help show these differences, but making a benchmark is difficult given the nature of Daz content. Anything in a benchmark must be available for free forever. It cannot be a limited time freebie, and they need to be things that people can get easily. The scene also must be able to fit on smaller VRAM cards so they can participate in the test. In the absence of professional benchmarks we have no choice but to do this with a crowd source method. It would be interesting to see a Furmark style scene that has bunches of strand hair.