Windows 10, CUDA, and Iray
Subtropic Pixel
Posts: 2,378
Hello:
I am trying to understand the current issues with Windows 10 and multiple CUDA GPUs. Here's what I know and what I don't know. Can anybody help me fill in the blanks?
1. I know that people are having trouble with Windows 10 and multiple Titan X GPUs, and that renders will run slower with 2 GPUs than with a single.
Does anybody know if the problem has also been reported for people using other CUDA cards, such as GTX 970, 980, or 980 TI?
2. Is this problem well-known in the industry?
I ask because I cannot find any other information about it outside of the DAZ forums. It's not even in "tenforums", which is often a good go-to for me.
3. Is this problem known to Microsoft? To Nvidia?
I ask for the same reason as in 2 above. I can't find anything about it, so I don't think we can assume that Microsoft and Nvidia know about it, or have acknowledged that it is a real issue.
4. Do we know all the Nvidia driver version numbers that have experienced the problem? Is it still a problem with the latest driver version and what version is that?
Just trying to understand the problem and who knows about it.
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
It is a driver issue, and you will have to keep an eye on the "Release notes" for the drivers. (Also look at the beta-drivers, as they usually handle these specific "issues", months in advance.)
Though, windows may ALSO attempt to help with resolution, there is no way to tell when they fix it. It will just come with an update, which you card-drivers may tell you... "Issue #xxxxx: Requires MS-Udate KB#xxxxxxx" Which would normally come in any auto-update from windows.
If you wish to expadite the issue, just send Nvidia an email, or join the forums and mention it in the bugs area, etc... Even if it has been said before... Volume speaks louder than individual personal requests. (The more that mention it, the higher a priority they tend to value it as... Blog it, and they will look at it for sure. xD)
Titan X is slower than Titan Z. Because it only has 3072 cuda-cores not the 5760 of Titan Z... I think the complaint about speed is more about people thinking that the X is somehow better than Z, but it is only faster at some GAMES, not anything "Real world", due to core-clocks.
As for Windows 10... Give it a year, and 90% of the bugs will be subverted, and some actually fixed. I believe they are placing a "version freeze" on windows, as an employee who does teck-work has told me. Like they did with PS3 and X-Box360, and PS4 and X-BoxOne. (I believe MS and Apple, now have an unspoken agreement to "stall releases", for the gain of "version stability advancement". Plus, the whole computer-boom and technology-boom has been essentially dead for the past two years. Like they had any real choice. They have to work to make sales now, actually work, not just re-paint OS's with new colors and regurgitated ideas that didn't work before.)
3072 cores is going to be much faster than older cards with fewer cores and blow the socks off of CPU rendering...no matter what.
But, it is a known issue that has had 1 recent update (geared mostly toward gaming, because Iray doesn't need SLI). And it is par for the course, because every Windows upgrade, in the past 20 yrs has had a 3 to 6 months of 'roll out' performance issues.
Lots of good conjecture in your post, but I'm seeking answers to my questions above, not just general guessing.
The problem that I am aware of happens with two Titan X's and not in SLI, as noted by a PA and another user elsewhere in the forums here. It's straightforward: Under Windows 10, two Titan X cards render Iray slower than a single. A lot slower. This is not a gaming issue, it is with iray renders within DS 4.8 on Windows 10 and with two Titan X cards. Searches for more info have yielded nothing but a bunch of Windows 7, Windows 8 issues that have long-since been fixed. I am looking for information.
I have not heard if Titan Z is also impacted. Do you know if it is?
And to reiterate my question in OP: Do we know if GTX 970, 980, and/or 980 TI are impacted? Where can I go to read about this?
And to reiterate my other questions in OP: Do we know if Nvidia and Microsoft ACTUALLY know about this? I can't find info, so I'm not so sure they really DO know. Does anybody know where can I go to read about this?
We can assume that Nvidia does know and care. Microsoft...who knows.
The recent driver update indicates that Nvidia IS working on Windows 10 related performance issues. As to this specific one, with Studio's implementation of Iray, they are probably aware of it, but you won't find anything specific mentioned. Other big name implementations of Iray will be suffering from it, too. So try looking there if there are any more specifics. And gaming issues, always have and always will, be addressed first.
And my read on it...no matter what the cards are, two or more CUDA cards, in or not in SLI suffer performance issues under Windows 10. The Titans, being the fastest will be the most noticeable. There are some mentions of performance issues on the various review sites for the prerelease Win10 version, that some were saying should be fixed by release time...and evidently aren't.
Thank you for your insight.
Do you have a reasonable confidence that we'll actually find out when "this problem" is fixed? Like many, my main workstation is still on Windows 8.1. I am pretty much ready to move ahead with Windows 10 except for this problem. How will people like me know? Without having to set up a test machine with 2 GPUs, I mean!
I'd say, keep an eye on gaming sites...or watch the driver release notes on Nvidia. My guess...November or December for a good 'fix' and major performance boosts.
I am experiencing crashes with Win 10 build 10525 in Iray with large renders. Renders get to about 95% then hard crash and restart. This started with the build 10525 update. I am running a Gigabyte Geforce GTX 960 w/4GB, AMD 8350 Black Box w/32GB RAM.
The issues I've heard about have to do with performance only, not crashes. And only in systems with more than one GPU card. Crashes might be due to any number of other issues, and your setup with only a single GTX 960 doesn't apply to this particular situation. You might want to confirm your Geforce driver. You also have an AMD 8350 CPU. I don't even consider AMD processors anymore, so I have zero idea what that even is, except that you would no doubt have a different chipset, and thusly all different drivers to support your hardware under Windows 10. Sorry, your issue isn't the same as the one I raised in this thread.
More cores do not equal more speed. You have two cars with identical engines; a car with one engine, and one with a hundred (imagine how that looks in your head).
The 100 engine car will not go any faster than the single engine. But with the right tuning the single could outpace the hundy, and vice versa.
My point here is cores doesn't matter in terms of how fast, it comes down to clock speeds and if your system can handle cross talking all that heavy duty rendering between the cpu, ram, gpu, and vram.
My GTX950 can push out a render pretty quick, and I use my old 650 to supplement it. Two older high-speed cards can out outpace to heavier monsters because my system is pretty mid-range. Try getting cards in this range just for rendering, get two 950s and try them out.
NVidia should look into making a CUDA rendering card just for 3d artists. With Daz and poser making rendering more accessible to the masses it would actually net them a ton of money. And they could prolly sell them for cheap.
Also, daz should replace the dynamic clothing system with a physx system. Would go insanely faster.
*scratches head* There was a lot wrong with that post...
No...1000 'slower' cores will be faster with a render than 300 'fast' ones. Of course, when talking about relatively the same number of cores the faster clock speed will win out. But it is still the total number of cores that is the primary determing factor for render speed. Then I'd guess whether or not the motherboard actually supports 2 or more cards at full 'speed' is the next factor...with core clock being third or fourth.
They do. It's called a VCA, and one can be yours for the paltry sum of $50,000.
nVidia, bless their hearts, aren't all that interested in the hobby 3D market. Their main push in 3D scene rendering is toward graphics and effects houses who invest in their own equipment, and also supporting cloud rendering with a few million dollars in nVidia hardware. Outside of 3D scene rendering, nVidia probably makes more money in allied fields like medical imaging, where they can sell equipment and customized programming in the millions of dollars.
CUDA cores are worker drones that each are asked to do a few calculations of math. When the core is done with the calculation, it sends the result back to the GPU. The more bees in a beehive, the more honey. It's true that some CUDA-based cards are faster than others with the same number of cores, but here, the whole idea behind CUDA is that it supports the multi-threading capability of parallel processing.
If you look at a program like Z-CPU while rendering, you'll see that the bus is only mildly taxed during processing, so bus speed is of marginal importance compared to other applications, like gaming. More important is the speed of memory reads/writes on the GPU itself. Slower RAM can impact the speed of renders.
That's starting to get into 'specifics'...
But, yeah...memory speed of the card is another player. Really, there are a ton of things to consider, with core clock being somewhat down the list. And some of them may only make a very small difference that it may not be all that noticeable...a few seconds isn't a make or break kind of thing.
True enough!
Zis is Kaos. We don't do specifics here!
I don't recognize this thread anymore!