New computer, new GPU, but CPU is still being used? Help!
Spectre576
Posts: 53
I just built a new custom computer with an i5 quad core, 16gb RAM, GTX 1050ti (4gb)...
I changed all the render settings to only use the GPU not CPU but every time I render something Daz uses about 50% of my CPU power and an additional 7GB of system memory.
The GPU is cranking away at 100% with all memory being utilized and it has cut my render times down to a speed that is extremely fast in comparison to my old system.
By the way when set to use GPU+CPU it uses over 90% of CPU so the GPU is certainly doing a lot of work now.
Now I know this is a total noob question but I am trying to understand what is happening here. I am assuming the scene size is simply requiring too much memory so it defaults certain processes to the CPU.
I have attached a copy of what the rendering box read when I started the last render and I see certain processes say GPU and others say "device 0". Can you all please help me understand what each of these lines mean, why this is happening on my system, and which aspects of my scene are taking up the most memory?
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
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
its not actually using the CPU to help render the image, the CPU has to work to 'power' the GPU. The harder the the GPU is being asked to work, the more usage the CPU will see also
... And it's an i5; rendering is going to keep it busy.
You have a decent gaming rig; rendering makes top-line gaming rigs cry like babies.
But it looks like you have a balance on what the CPU can deliver and GPU requires.
The image tells you the GPU card is being used.
The following lines are what RAM they're using; for comparison purposes, think MB when you see MiB, and GB when you see GiB; just ignore the i, unless you want an exact comparison.
The MB is more like a decimal value, whereas MiB is the value the way computers use memory (The binary is: 1, 2, 4, 8, 16, 32, 64, 168 etc; the units are GiB for example.); it is just unusual for consumers to see the notation like that; personally I think it is misleading to do what they do.
It's why reported memory can vary over what you thought was present.
... you have millions of bytes; dividing by 1000, or a 1024 to get kB or KiB (or their multiples).
And an example: 1000 MiB is 1048.576 MB.
It's labelling your GPU as Device 0 because computers start counting from 0 not from 1.
Iray will always uses one core on my system (i7 with gtx 970m). I would guess that with the faster 1080ti and the slower/older i5 iray needs the use 2 cores for the cpu to keep up with the gpu. Iray is rather inefficient with regard to cpu usage in GPU "only" mode.
Awesome, thank you everyone for helping to understand how this all works, it makes much more sense to me now. I am pretty new to 3D rendering and just really starting to understand the complexities of high end computing but it is all coming together.
As for my system, I agree with you nicstt that this would make a good gaming rig but because of what I am using it for I definitely see some upgrades in the future, such as a second graphics card.
patrickm576, I bought pretty much the exact same setup in December and I have some suggestions/observations.
Observation #1, I'm getting a 1080 as soon as it is financially responsible.
Observation #2, see Observation #1!
Ok in all seriousness, a lot of users here are underestimating the 1050ti, it's a surprisingly capable card. The ram limits how much you can drop in a scene and that in turn is going to drop you into CPU mode when you least expect it, you'd be surprised how quickly 1 fully dressed character in a room with some knick-nacks can chew up 4gb of vram because most of the items made by PA's are optimised for looks, not memory.
Also, I've found that many times I can overload my vram and get kicked to CPU rendering, remove items, and even if I'm under 2gb for the scene, it won't switch back to GPU rendering. In that case you have to save and restart Daz, and in extreme cases I've had to reboot completely.
The lifesavers for me have been -
http://www.daz3d.com/forums/discussion/137161/reduce-texture-sizes-easily-with-this-script/p1 This one is a MUST HAVE for 4gb cards.
http://www.daz3d.com/iray-memory-assistant Also a MUST HAVE so you know when (and how much) to use the above
Make sure CPU is unchecked in the advanced render options.
If all else fails and you have reduced the textures, don't forget to copy the contents of the temp folder BEFORE quitting Daz. Once you're back up you can just drop those files back into the temp folder, load the scene and render away (provided you're back below the cards limit).
Also, I've done a number of renders where it says the texture memory is 6+ and even 8+ GB yet the card still runs it in GPU mode, I have no clue what's going on there, I really wish there was more information/documentation about how the hardware processes Iray renders.
Additionally, there are some PAs who set their default SubD to 3 - 4, or have 30+ 4k textures on a set of armor etc. This is great for closeups using 6-12gb cards, but you're probably never going to render a ladybug crawliing across a knights shoulderpad anytime soon (now I want to render a ladybug crawling across a knights shoulderpad...
) and won't notice the difference with the full figure in frame, even in a 4k render. Tinker with the surfaces tab for a while and you'll learn how to save a lot of memory, there's dozens of ways to do it, it just depends on what you;re trying to render. (tip: the first thing I look at is SubD, if I'm not zoomed in on it the item, I set it to one or sometimes 0 if it's far enough away)
Bottom line, if you start enjoying this as much as I have, you'll probably do the same as me and start planning to do a BIG UPGRADE later this year to a 1080. There's even a rumor that a 1080ti with 10gb might be out before xmas.
Hope this helps, good luck! :)
One last thing, unless something has changed recently/changes soon, you're only going to be using vram from one card, the vram can't be shared. So 1 8gb card is WAY better than 2 4gb-6gb cards.
But the CUDA counts of cards do stack. So if your render fits on both cards, you can get extremely fast render times with 2 cards. Maybe two 970's don't have the memory of one 1080, but they will render faster if both cards are running! And this is where a 1080 starts to drop in value. Because the 1070 has the same amount of memory as the 1080, the only advantage is CUDA count. But as I said the CUDA counts stack, so you could buy two 1070's and your render times would destroy anybody who's s running one 1080. And with the extortionist pricing scheme Nvidia has, you could basically buy two 1070's for the cost of one 1080. Unless you just have cash to burn, the 1080 is not worth it.
And that is my plan. What I will do is buy one 1070 at a time. I have a 970 right now. So I could pair the 970 with the 1070. Obviously, the 970 will drop out if its vram is exceeded, but for those scenes it runs, the render will fly. Then once I can afford a second 1070 (or who knows, maybe another card will be out by then,) then the real fun begins.
Hi Brugan,
Thanks for the input. The GPU is definitely working when rendering. the log file I attached (which I don't see so will attach again) shows around 2gb if I'm reading correctly using the conversion a previous commenter helped me with. Correct me if i'm wrong though. If that is correct though it should only be using the card for actual rendering. Like i said it is very fast compared to using just a cpu alone.
Also as another person said the cpu may be helping with the process though it still all fits on the GPU VRAM. Quite frankly it doesn't slow my computer down one bit while rendering I just wanted to make sure the GPU was being fully utilized. I agree though an even better card in the future would be nice.
I downloaded that script so thanks for the advice. Can you briefly expand on what the SubD levels affect?
Thanks
Also, I agree that the 1050ti is a pretty good card for what i'm using it for, it would simply be nice if it had more memory.
Good question, I was under the impression that Daz itself can do that but maybe I was wrong
While CUDA counts do stack, there are other limitations one can quickly run into (power, PCI-E lanes, etc.) that don't necessarily indicate that two lower speed cards are better than one higher one. In the case of buying one 1080 or two 1070s, I would tend to recommend the 1080. Why? It consumes less power, has the same amount of memory, and only takes up one slot. Also, current prices put 2 1070s as a bit more than a single 1080. Yes, you will have more CUDA cores with two 1070s. But unless you have a very high-end CPU and MB (with all the extra slots and PCI-E lanes needed) going beyond a two-card setup isn't very simple. So buying a single 1080 still leaves you with room to improve (with a second 1080) whereas if you already are running two 1070s, the options to improve are considerably less available.
Obviously, the next choice is to go with a single Titan X (pascal), though that gets even MORE expensive......
The best option is to buy the best single card you can afford, so that if you find you STILL need more speed, you can always add a second card after saving a bit more money.
Someone with more experience should explain the SubD levels, basically from what I have read it affects mesh smoothing by increasing the polygon count through some mathematical magic I don't pretend to fully understand. What I can say for sure is that anything above a setting of 1 can drastically increase memory usage (relative to a 4gb card), and unless you're zooming in for detail, most items will look fine with a setting of 0.
One last tip, when you're going to use that reduction script, switch out of Iray before you run it or the process will take several minutes instead of several seconds. :)
Also, it only reduces the items you have selected, not the whole scene. Took me a while to figure that out lol,
SubD divides the geometry; it increases four times as dividing in half both ways horizontal and vertical has that effect.
As an example: HD is 720 (Full HD is 1080); quad HD is 1440. 1280 x 720 for HD; 2560 x 1440 for Quad HD. Each axis is doubled making it four times the number of pixels.
So each SubD increases vertices by four times, although geometry doesn't have a huge impact on RAM useage; decent sized (4096x4096) textures take up a lot more room. If you're struggling to fit a scene on your card, either use characters and other assets that share textures; reduce the size of textures, or render in sections.
To some extent; I was interested to read Mec4D's comments on how making sure the cards she supplied had enough processing power to supply their needs. They were older generation cards though.
We're short (short doesn't mean there are none) on tests on how the 10 series cards perform in multiple setups. They may be the same as older versions, but we don't know that yet; we need more data.
You're free to disagree, but while PCI-E lane considerations for speed aren't an issue with Iray, the NUMBER of available lanes still is. And GPUs aren't the only things using PCI-E lanes.....that's a common misconception. M.2 drives, SATA controllers, even onboard sound and network devices can 'occupy' PCI-E lanes, reducing the number available. And you can get to a point where you don't have enough. Most GPU cards will NOT work with less than 4 PCI-E lanes......so if enough other devices are using lanes, you can run out of them, and a GPU will drop out....or something else will.
Also, power consumption is a major issue when you go above two cards. Even with pascal-based cards (the most power-efficient ones to date) putting in a third 1070 would push you to about 450 watts of consumption JUST for the GPUs (3 x 150W TDP) and you have the additional case cooling fans that are inevitable with having mutliple cards tightly packed (most MBs don't provide a lot of space when you get to 3+ cards.......
There's a lot more to choosing than just money and CUDA. And based on all reports I've seen, it's always better to leave some room for expansion.......
Currently, 1070 cards are running an average of about $420 (lowest price was $370 for new) vs the 1080 which is averaging around $630 (with the lowest $580 for new). So when you recommend two 1070 to start, instead of a single 1080, you are actually costing more, and limiting further expansion more severely. While they will spend more in the long run (when they get the second card) they'll also be better equipped to determine if they really needed it for that price.
Also, the more GPU cards you run, the more the CPU has to do with keeping them sync'ed.....and the PCI-E speed starts to come into play again. And every update that comes off that card has to be transferred as data to the CPU and to main memory. Northbridge and Southbridge come into play, as does any drive you are using for temp space. It's a lot more complex of an interaction than you think.
PCIe 3.0 is 1GB/s per lane. Twice as fast as PCIe 2.0. So you don't need as many lanes if your motherboard supports PCEe 3.0. In fact, it pretty much impossible to see a difference between x8 and x16 PCIe 3.0. For those that want to know, video cards will continue to work down to 1 lane thought it's possible some of them require at least x4 (I do know x8 is required for SLI). And while the BIOS can re-allocate lanes, you can't run out of lanes to the point where a device will no longer work. If this happens to you, get a better motherboard.
In non-realtime rendering such as iRay, PCIe lanes will matter mostly on startup and even then I'm not convinced there's that much of a difference. During rendering, there is certainly back and forth between the CPU and GPU, but again, saturating x16 or even x8 is nearly impossible.
Here's a link to gaming benchmarks between x16 and x8 PCEe 3.0
http://www.gamersnexus.net/guides/2488-pci-e-3-x8-vs-x16-performance-impact-on-gpus
I agree I just bought a 1070 for a great price and it is my 3rd 1070. I use the 1070 with my 6gb 780's and my render times are really low. for a single charater with clothes and basic lightning It happen under a 1 min. I would stick with 1070's for bang for buck and wait till new cards with more vram come out or try to snag a titan x.
FYI, http://www.daz3d.com/iray-memory-assistant on sale in fast grab :)