GTX 980 Ti vs Titan X for Iray
Aala
Posts: 140
So, it's a simple question for anyone who has experience with any of the cards. Are the 6GB extra memory on the Titan worth it for the money?
Where's the limit for the Ti's 6GB when it comes to number of figures/props on a scene?
Your help in the decision would be greatly appreciated. :)
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
That's exactly what I want to know! Titan X is number 2 and GTX 980 Ti is number 3 top of the world as of Sept 2015. http://www.futuremark.com/hardware/gpu But does it matter for iRay?
Well, when it comes to speed, they're almost the same, 2816 vs 3072 CUDA cores. But the Titan X has double the memory, which is vital for Iray scenes. I'm just not sure how much it's worth it because I don't have expereince with GPU rendereing. So far, I've done all my experiments on the CPU.
I got a Titan X because it has 12gb of Vram, my reasoning was that in the future even if a better card comes out and 12gb ram becomes standard I can still use the Titan for rendering and not be limited by ram. While you can't use multiple GPUs for gaming if they are not the same model, you can use them for rendering but memory is capped to the lowest GPU Vram. Memory is important if you are going to use a lot of high definition textures which take up a lot of ram. If you use compression and don't have that many high resolution textures you can still do big scenes with 6gb.
Iray performance wise should be similer with an edge going to titan x. Cuda cores is what is mainly used for rendering, Titan X has 3072 and the 980Ti has 2816, the clock speed on both are the same.
I just installed a Titan X last week. Speeds up Iray renders tremendously. I highly recommend it. The only drawback is it runs incredibly hot. Make sure your case is well vented. Right now I'm rendering with my side door off. If I close the case the Titan X eventually throttels down and runs slower because it gets too hot. I'm going to have to build a custom vented door. It's worth it though. Saves hours on each render.
Yeah? Hmm... seems like I might need an aditional cooling option in that case. Or I better look for a good case for the PC.
Yeah, I'm thinking along the same lines too. The RAM might come very handy in the future.
I've been debating back and forth. It comes down to how often that extra 6 GB of memory is going to matter.
And I have no freakin' idea.
I might be able to afford two 980Tis! Twice as fast! YES! Eeeexcept there's the potential memory bottleneck. ... my head hurts
I have the Titan X only because of the extra memory. As far as speed in Iray the Titan X is only slightly better speed wise. My Tital X's are overclocked. They are EVGA. I have scenes that exceed 6GB and are pushing close to 12GB. You will be surprise how fast this can happen with a complex scene and multiple figures. I believe I paid about $200-$300 over what the 980 overclocked with 6GB would have cost. The Titan X's were recently on sale at Amazon for under $1,000 for the EVGA. I picked up two more which I have yet to install. If you do not have complex scenes with multiple characters the 980 is your best bet but the 6GB version not the 4GB version. Both the Titan X and 980's are excellent cards.
I have an EVGA 980Ti. It drives my 2560 by 1440 monitor as well as doing Iray rendering in Daz Studio. Here are two fairly large scenes I've done with it, to show you the amount of memory those scenes took (according to GPU-Z).
Wooden Weapon Welder Wanted (rendered 2000 by 1600) = 3610 MB
I Dare You to Release That Arrow! (rendered 2000 by 1500) = 4716 MB
Take a look at the software that came with the card and create a custom fan profile. My 980ti ran at 47% fan speed and 82 degrees C out of the box. After I set up the fan profile, the fan now runs at 65% and the temperature is down to 66 degrees when the GPU is fully loaded (and I still don't hear the fan).
I run out of memory occasionally with the ti; it isn't just about all mats etc that are used, but the output render size too; i've not tested since the ti, but on the 970 (when I used it as my render card) I've taken it's memory usage from just over 2GB, to 3.8GB just by upping the resolution.
If you need large (very) you'll use more memory.
Overclocking the card won't matter for Iray. Memory and CUDA cores are the only specs of importance for Iray render speed. The design temps on the TitanX and 980TI are 90C. My 780 runs right up to the 80C setpoint w/ 50% fan speed and stays there. I would have expected similar from the 980TI and TitanX.
The attached scene needed 2880 MB at 2000x1125. When I started the render, I had several other apps and windows open, (GPU using 655MB before render). If you can use onboard video or a cheaper card, it might make or break the GPU render. Also, close everything but Studio.
FWIW, I've done some quick tests which lead me to conclude that there's more to GPU memory usage than resolution or number of figures. When I render a single clothed figure with HRDI lighting I get 2816 MB at 750x1000. Rendering the same figure at 3000x4500 uses 2990. I suspect that texture usage and compression works a a big factor.
Thanks a lot for the feedback guys! Helps my decision tremendously, which is why I think I'm going with the Titan.
Just one more thing, does RAM matter for Iray much? I'm gonna be rendering on Iray exclusively, and I might cut back my planned specs for a PC from 32 Gb to 16 Gb to make the Titan hurt my pocket less. I can always upgrade the RAM, not so much the GPU.
Memory always matters. If you can do both that would be good. However, if you cannot I would go with the Titan first and add the memory latter for your main. Just make sure you setup paging at about twice your memory preferably on an ssd drive.
Memory always matters. If you can do both that would be good. However, if you cannot I would go with the Titan first and add the memory latter for your main. Just make sure you setup paging at about twice your memory preferably on an ssd drive.
Yeah, I think I'm leaning toward a Titan.
At first I was thinking 'well, I don't do high resolution/big images with a lot of stuff in them ANYWAY...' and then I reminded myself that the REASON I don't make those images is because the last time I tried it took 30+ hours to render.
And, well, it'd be nice to make big stuff.
Yeah, i know, which is why I'm saying the Titan first too. RAM is easily upgradable, while the Titan will be usable even if I someday decide to get a better card in the future (hearing of 32 Gb VRAM cards next year!).
I'm just wondering what exactly DS does with the RAM. I'm on a laptop currently, and I haven't had any issues with its 8 GB's of RAM so far, but still leaves me to wonder...
I'll still use it to make 1080p stuff to be honest, doubt I'll go higher than that in the near future, but then again, 1440p might be a realistic target too.
I've been intrigued by the idea of making large art print works, which are like 9k x 13.5k.
That's a 208MB PNG. Eep.
you will be back at 30 hour render times
Then I'll need THREE Titans!
And then recovery from what my wife would do to me...
Out of curiosity, I rendered one figure with hair at 7070x10000 and it took 3756MB according to GPUz; (wouldn't go above 10k).
1273x1800 (same aspect ration) was 2190MB. BTW, no clothes, or any background with just a 50MB HDRI for lighting. It rendered to 81% convergence in: Total Rendering Time: 1 minutes 31.89 seconds; the higher resolution was on the 9th iteration at that stage, and hadn't decided the percentage (still on 0). That was just using the 980ti. The .png size was 47MB.
EDIT
Just realised I can change the maximum resolution if I wish. :) I can't see a reason for changing the minimum from 1, unles it is to increase it /nod
so daz iray has a resolution cap? Intersting. Not like it would impact me, i tend to say below 6K.
I rendered an image yesterday to 98% convergence (3630 iterations) at 10,000 x 4000. It was one character with clothes and hair inside a set. Light was scene only with mainly emmissive light from the ceiling and two spotlights for fill on the character. It took about five hours. (It didn't really need that long. It looked good after three and a half. I just let it go to see how long it would take to get to 98%.)
My system is a dual 3.10GHz Xeon with 128 GB Ram. I've managed to get my old Quadro K5000 and my new Titan X graphics cards to both work in the system at the same time, so the K5000 helps a little with the render too.
Here is the iteration breakdown from the render log.
CUDA device 0 (GeForce GTX TITAN X): 2521 iterations, 35.490s init, 19447.547s render
CUDA device 1 (Quadro K5000): 656 iterations, 37.648s init, 19412.704s render
CPU (30 threads): 453 iterations, 21.832s init, 19586.143s render
As you can see the Titan X does about 2.5 times the work of both the K5000 and the CPU together. I wouldn't want to live without it.
Lars: It has a maximum, but you can unlimit it like most parameters. What's annoying, though, is if you save and reopen the file, the limit comes back. So you have to change resolution every time.
(That's a bug that should probably be reported, but it's ... rather inconsequential so I haven't bothered)
Any idea how much memory it used? Interesting stats, the Ti should render just as fast I suppose.
Does anyone know how much memory a fairly average scene with about 4 clothed characters in it needs? I might not be able to get the Titan afterall because it's out of stock where I live, and might have to settle for the Ti. :/
thanks for the info.
Not exactly sure how to read this but it looks like it took 1720 MB on each graphics card to render my simple scene. Here's the relevant part of the log:
Rendering with 3 device(s):
CUDA device 0 (GeForce GTX TITAN X)
CUDA device 1 (Quadro K5000)
CPU (30 threads)
Rendering...
Using device 0 to build data for host
CPU (30 threads): Scene processed in 19.223s
CPU (30 threads): Allocated 916 MB for frame buffer
CUDA device 0 (GeForce GTX TITAN X): Scene processed in 19.804s
CUDA device 1 (Quadro K5000): Scene processed in 19.864s
CUDA device 0 (GeForce GTX TITAN X): Allocated 916 MB for frame buffer
CUDA device 1 (Quadro K5000): Allocated 916 MB for frame buffer
CUDA device 1 (Quadro K5000): Allocated 804 MB of work space (1024k active samples)
CUDA device 0 (GeForce GTX TITAN X): Allocated 804 MB of work space (1024k active samples)
Yeah, seems so. Thanks for the help!
Hey all,
I'm trying to create my first build; but I'm completely out of date on the new hardware, can you help me understand --
1.) Which card provides more benefit for rendering times:
- Two (2) 980 ti Hybrids now in SLI, or
- One (1) Titan X Hybrid? [With plans to add a 2nd in a year or two in SLI later].
2.) Which motherboard & processor would you recommend (upper-end, future-proofing), if you were constructing a build right now (and should number of cores utilization be a considered factor for the motherboard)?
- i7-5820K + x99 motherboard (6-8 cores, with no integrated 3.1), or
- i7-6700K + Z170 motherboard (4-cores, with integrated 3.1).
3.) And if x99, due to current cores use, then the:
- ASUS Sabertooth x99 TUF, or
- ASUS x99 A (is Broadwell E, or integrated 3.1 an issue here?), or
- Some other/better x99 board?
4.) If I wanted the QUIETEST system (I have hyper-sensitivity to noise), yet coolest (air/liquid, doesn't matter) Case, Power, and Fan/Water system, which would you recommend for quiet operation? And why (or any other tips)..? I don't mind a big tower, in fact, I prefer it -- and I'm not a gamer, so LED lights, etc, are not important to me. I will use the system for Photoshop, Lightroom, Illustrator, Daz3D, possibly more 3D in the future, and big monitors.
Thank you!
1. Two 980ti will give you a faster render speed then one titan x (speed or memory.. your decision)
2. - i7-5820K + x99 motherboard (6-8 cores, with no integrated 3.1),
The reason to go with x99 mainboard for a GPU render workstation is PCI 3.0 support with 40 lanes..(up to 4 GPU cards with pci 3.0 (x8) plus ssd ... CPU's for the X99 chipset dont have grapic on board... so always count one PCI slot for a monitor card in addition to the render dedicated 980ti's or titans....
The i7 5820k has only 28 PCI lanes... so you cant run 2 GPU cards with full x16 speed! -> if you dont need more then 4 cores the XEON E5-1620 V3 3.5GHz offers enough speed AND 40 PCI lanes for 300USD. the xeon is the better option for a GPU render workstation with 3 - 4 GPU cards.
I7Skylake has only 16 PCI lanes and 20 from the chipset on board for ssd)... but it is a very fast CPU and offers integrated grapic.. the maximum speed are 2 GPU cards with PCI 3.0 (x8) because of this.
3.) ... Gigabyte offers the cheapest boards with 4 SLI (x8) slots... they do this because they limit m2 ssd bandwith (but still fast enough).. if you want it all.. you have to pay 400usd up.
4.) Go with a fast xeon and x99 and dont overclock... use a very quite CPU cooler such as the Noctua NHD-14.. and a spacy case with optimized airflow Corsair Carbide Air 540 as example.... if you dont overclock your GPU cards you wont need additional cooling that way...
Edit: And make sure that you pick quite gtx 980ti's .> MSI GeForce GTX 980 Ti Lightning 6GB attention this card takes 3 PCI slots size!
(Yes; I am going to pay dearly -- but worth it, as my 7-year-old computer is totally done).
I want to keep it between $3,000~$3,500 -- do you see any weakness in this build?:
http://pcpartpicker.com/p/7G8wMp
Thanks for your advice, I will utilize x99 & since I am using smaller scenes, I will prefer Rendering Speed over Memory (regarding 980 ti vs. Titan X). Of course, other drives, like scratch drive, and 3 terabyte storage, I'll rob from my current computer. Some things are totally flexible, such as the Cooler, and the Case Tower, changing to Quieter Fans, and the Power Supply. Any advice on those matters, can be helpful. Usually, I have a lot of storage to plug in. But with USB-3 this is certainly less important these days, I suppose.
Regards,