- This topic has 28 replies, 10 voices, and was last updated 3 years, 9 months ago by
.
-
Posts
-
Does anyone think it's associated with DS staying in background processes even after its closed? If I want to re-start DS after a render, and don't want to wait 10+ minutes, I have to manually stop DS in Windows processes before I can open it again. Perhaps that's a separate issue.
In many cases even if a motherboard has multiple “physical” x16 slots chances are fairly good the secondary slot may only have x8 physical pins, and a tertiary slot may often only have x4. This was vert bad for proper SLI setups but for multi-GPU rendering it’s really not a problem.
The CPU doesn't have enough PCIE lanes in most cases to have two PCIEx16 slots with PCIEx16 bandwidth. That's why the slots are only wired for PCIE8X.
I don't believe that its actually bad for SLI gaming setups. In synthetic gaming benchmarks and tests such as Port Royal and DirectX 12 Raytracing feature test there is no measurable difference between systems that have two or more PCIEx16 slots that are actually full PCIEx16 bandwidth and systems where both slots are PCIEx8 bandwidth when using the same two cards.
For example, most of the top Port Royal and DirectX 12 Raytracing feature test results are done on systems where the motherboard is two PCIEx16 slots that act as PCIEX8 with two cards. Even when its a PCIE 3.0 system and PCIEx8 in both slots, there is no significant difference in performance when compared to a system with PCIE 4.0 and dual PCIEx16 slots with full x16 bandwidth.
The only “notable” slowdown will be when loading your scene into VRAM but even this is often hugely overblown and essentially a non-issue.
If you look at the Daz log files, you can calculate the loading time.
Loading Time Formula: (Total Render Time - Device Render Time) = loading time
The limiting factor is often the drive the assets are stored on and not the PCIE bandwidth of the cards. Even a PCIE 4.0 NVMe SSD is often the limiting factor when it comes to loading time and not the bandwidth of the slot the graphics card is in.
@Hamster "If you have a GPU with 8Gb and a GPU with 24 Gb that work together to render a scene, you will not be able to render a scene that requires more than 8Gb."
I thought it was the other way around: that the 8GB GPU was dropped and the scene is rendered only with the bigger GPU... what is the case?
The limiting factor is often the drive the assets are stored on and not the PCIE bandwidth of the cards. Even a PCIE 4.0 NVMe SSD is often the limiting factor when it comes to loading time and not the bandwidth of the slot the graphics card is in.
I tested that theory multiple times with DAZ and it has not been proven true. I tried a NVME drive vs a good old HDD and load times were about the same. If you look at IO data, disk access is very low in either cases. I think it's because DAZ isn't optimized. But my theory before I tested was that it would be a bit faster with a faster drive. At least not significantly. And I tried with large scenes.
I thought it was the other way around: that the 8GB GPU was dropped and the scene is rendered only with the bigger GPU… what is the case?
It will drop the GPU but keep going on the one with enough VRAM. That happened a lot when I had my Titan RTX with a 1080Ti. Sometimes scenes were just too large and Iray would just drop it but kept rendering on the Titan.
This has really gone off topic of the original post.
I tested that theory multiple times with DAZ and it has not been proven true. I tried a NVME drive vs a good old HDD and load times were about the same. If you look at IO data, disk access is very low in either cases. I think it’s because DAZ isn’t optimized. But my theory before I tested was that it would be a bit faster with a faster drive. At least not significantly. And I tried with large scenes.
I tested a scene with 8 different figures wearing different clothing, two sailing ships, fog, and a bunch of props. There was a difference in loading times, with all assets stored on the SSD, between a SATA-III SSD and a PCIE 4.0 NVMe SSD.
The render times and loading times didn't significantly change when I tried the same scene with the same GPUs on a Threadripper 3970X CPU system with PCIEx16 bandwidth for the slots the GPUs were in. Even with more PCIE lanes, four times the RAM, and more CPU cores, I couldn't get the scene to load any faster from the NVMe SSD than it did on the other system.
That made it seem that the SSD was the limiting factor in loading times. The only test to see if the SSD limits loading time is to test with full bandwidth PCIE 5.0 NVMe SSDs, in a system that supports them, and try the same scenes with other SSDs to see if there is a difference in loading times.
I tested a scene with 8 different figures wearing different clothing, two sailing ships, fog, and a bunch of props. There was a difference in loading times, with all assets stored on the SSD, between a SATA-III SSD and a PCIE 4.0 NVMe SSD.
I don't know how you were able to see a real difference. I tried with a huge scene, never could see anything significant between having everything (DAZ lib and scene) on a NVME drive vs HDD. If you look at IO metrics while loading scenes in DAZ and at the beginning of a render, IO will burst here and there but most of the time it's flat down to zero and DAZ is processing some shit.
I'm currently hosting my DAZ lib on an HDD and my scenes are on an SSD but in an enclosure so... going through USB 3. It's not any slower. And I have scenes with multiple characters and stuff that can use 18 GB of VRAM sometimes.
Unless things changed recently. My tests were executed over a year ago. So maybe it's different with new versions... but I doubt it.
I don’t know how you were able to see a real difference.
I actually calculated the difference in the Daz log files. The SATA SSD consistently had a longer loading time.
Loading Time Formula: (Total Render Time – Device Render Time) = loading timeI thought it was the other way around: that the 8GB GPU was dropped and the scene is rendered only with the bigger GPU… what is the case?
If you ask DS to use both GPUs, it will respect your preferences. It will build a scene that can run identically on both GPUs and it will be based on the smallest memory. This behavior makes sense, indeed. If your GPUs have a similar power and if the scene is small enough to fit the smallest memory, you will get a speed-up.
If your scene is too large to fit on this smallest memory, there is a fall back and DS will drop the smallest GPU. But I do not know if the fall back will be on the CPU or on the biggest GPU (which would be smarter).I actually calculated the difference in the Daz log files. The SATA SSD consistently had a longer loading time.
Loading Time Formula: (Total Render Time – Device Render Time) = loading timeI did that. Never seen a significant difference in times, ever.
@Hamster Thanks for input! From what I remember, it falls back to the bigger GPU and not the CPU...
I'm planning to buy a 4090 with 24gb when they are available. I'm thinking about keeping the two 2080 I have, but that would of course be nuts if I never could use the the full 24gb when needed.
(With permission...) Neither Daz 5, nor version 9, will improve. Iray, as I have always said, and I maintain, it is an outdated, old render engine that even Nvidia has abandoned. (A crude copy of his Mental-Ray, which Autodesk used even in 3ds Max, or Maya...).
As long as Daz doesn't abandon Iray, and proposes more advanced rendering engines, you're going straight to the slaughterhouse: EVERYONE.
So, excuse me, use Arnold or V-ray... (You can RenderMan from Pixar, or Maxwell from Next Limit (creators of Realflow...)...but...)
Not Octane, not RedShift, not that stuff..., licensed under "login"...
If you want good results, in optimal time and a "very decent..." finish, Iray is NOT your thing: That's why it's free...iRay may be old and outdated but as someone who had to suffer the pain of 5-6 hours of CPU rendering of a 1024x768 render back in the days of 3Delight I much enjoy the identical scene rendering in less than 2 minutes upconverted to iRay shaders at the same render size. I've also used DAZ Studio for over 12 years and like @InverseKinematics I'd rather skip having to learn a completely new software system from scratch which would take me another decade plus to learn as well as I know DAZ Studio. At least when DAZ added iRay the learning curve wasn't too hard for me.
If there's ever a new version of DAZ to be released, it should have the ability to plug-in alternate rendering engines, instead of relying on just what they supply.
It would be helpful to us but also in their own best interests, as it might prevent users from dropping DAZ completely and moving on to other softwares that don't rely on the iRay bottleneck.
Has there ever been a speed comparison done between Cycles in Blender and Iray in Studio using the same scene?
Its difficult to do the same exact scene in both unless you use very basic objects. Even some of the assets, such as vehicles and buildings, have different surface settings and appearance when both a Daz and Blender version are available. So you can get a rough estimation, but not an exact number.
I think Blender cycles is faster than Iray based on my limited use of blender, but I couldn't give a definite answer.
- You must be logged in to reply to this topic.
