PDA

View Full Version : Titans or less?



JBT27
03-08-2014, 08:58 AM
We're coming up on getting new machines, and debating hard about the Nvidia Titan GPUs, or getting a lesser one. It's not about gaming, at all - Lightwave, maybe Modo, and the Adobe CC apps, notably After Effects. My personal view is to go for them, because of the 6Gb for starters, because of wanting enough overhead to handle 4K shots which even in our sector (heritage) can't be far around the corner.

But we're not big on hardware and would appreciate any thoughts and experiences with this.

Thanks.

Julian.

dwburman
03-08-2014, 10:50 AM
I don't think the Titan would be much benefit to LightWave directly except for being able to load more/larger textures into the card for display purposes (i.e. not rendering). However, the stronger GPU would most definitely benefit things like TurbulenceFD or Octane that can be used within LightWave. I imagine that the Titan would also be very beneficial for After Effects and any other software that specifically uses the GPU (3D Coat comes to mind), but you probably already know that. :)

That's about the extent of my knowledge on the subject.

JBT27
03-08-2014, 12:18 PM
Thanks. That is my level of thinking too, and I came up with those same justifications. There is a 4Gb card which is no mean beast, but several hundreds of pounds less (GTX 770) - I saw a review which said that several 'workstation features' were switched on in the Titan which are not active in the 4Gb GTX 770 - the 770 is, as I say, 436 cheaper - we're looking to order from PCSpecialist, in the UK. So it begs the question, given I am coping (barely) with an old 1Gb GeForce 9800 GT, would the 4Gb do it? Probably. But for a bit of future proofing, do I get the 6Gb Titan? It's conceivable that the 436 saved now, by getting the 770, could go to a 6Gb card when and if they become more common, and should I need one. We're speccing the machines with 6 core i7 and 32Gb RAM anyway, but I am using AE heavily, and may well want to use LW tools that rely on OpenGL. The usual tough teeth clenching decisions :)

Julian.

AbnRanger
03-08-2014, 04:11 PM
Thanks. That is my level of thinking too, and I came up with those same justifications. There is a 4Gb card which is no mean beast, but several hundreds of pounds less (GTX 770) - I saw a review which said that several 'workstation features' were switched on in the Titan which are not active in the 4Gb GTX 770 - the 770 is, as I say, 436 cheaper - we're looking to order from PCSpecialist, in the UK. So it begs the question, given I am coping (barely) with an old 1Gb GeForce 9800 GT, would the 4Gb do it? Probably. But for a bit of future proofing, do I get the 6Gb Titan? It's conceivable that the 436 saved now, by getting the 770, could go to a 6Gb card when and if they become more common, and should I need one. We're speccing the machines with 6 core i7 and 32Gb RAM anyway, but I am using AE heavily, and may well want to use LW tools that rely on OpenGL. The usual tough teeth clenching decisions :)

Julian.I bought a GTX 670 4GB several months ago, and found that in general tumbling/orbiting around the object in the viewport, it was somewhat more lively than the GTX 470 I was upgrading from. But, after selling my 470, I noticed a major problem in 3D Coat, using that card. When I would turn on Wireframe in the Voxel room, anything over 1-2mill poly's would bring the 670 to it's knees. The FPS would just drop through the floor. I had no such issue with the 470. I asked Andrew about it and he put me in contact with his NVidia rep, and the guy acknowledged it was a known problem, and that NVidia was taken aback somewhat from the reduced CUDA performance in that line, and that would be addressed in the next architecture (Titan and 780).

The 670 is practically a GTX 680 with fewer CUDA cores. And the 770 is basically a GTX 680 with some modest enhancements. So, it's rather deceptive to put a 700 series label on it, when in fact it's still a Kepler Card. The 780 and the Titan are Maxwell architecture...and they bumped the Memory Bus back up to 384. The Kepler line, for some odd reason went backwards to a memory bus of 256. For this reason, I sold my GTX 670 and bought a GTX 580 3GB on eBay. It has no problems whatsoever with the wireframe in 3D Coat.

In terms of CUDA performance, it remains 2nd only to the Titan...still. The Fermi cards were very strong CUDA cards. So, as was mentioned previously, if you plan to use Octane Render, or any other GPU-based renderer...or TurbulenceFD, then it might be best to save some serious cash and just buy a GTX 580 3GB card on eBay, amazon, or some other auction type of site. I'm happy with mine, and it rocks using Thea's Presto GPU rendering engine (similar to Octane) or in 3D Coat.

I'll wait until the next full line comes out, before I consider upgrading.

gordonrobb
03-08-2014, 06:02 PM
There are too many pros and cons, and you are unlikely to get a consistent view here. I have twin Titans because I use Octane and needed more than 3 Gig. The 770 will be good, you could get 2, but I'm not sure what programmes (other than Octane) allow this to just double the performance. The 780s are good, especially the 780 Ti (basically the same as a Titan Black, but with only 4Gig).

spherical
03-08-2014, 08:12 PM
Render performance and viewport performance are two different things. The 6xx and 7xx series viewport performance was crippled in order to sell more Quadro cards, where handling huge datasets in the viewport is necessary. The 5xx series do better on the tumbling aspect but the 6xx and 7xx series do better on the CUDA performance, which is rendering.

akademus
03-08-2014, 10:34 PM
It depend what yo need it for. 780ti would be a great option if you are after GPU rendering. Regarding memory, mine has 3Gb on it (780) and so far I haven't had any memory issues. I tried it with really large scenes and it never ran out of memory.

gordonrobb
03-09-2014, 03:09 AM
The size of the scene isn't really what matters. You can have a huge number of polys in a scene and never approach 3 GB. What matters is the textures and their sizes.

gordonrobb
03-09-2014, 03:12 AM
Also, if you use the card for your UI (display connected to it) you won't have all 3 GB available to use.

JBT27
03-09-2014, 10:13 AM
I'll have two HD displays connected to it, same as now with my old 9800 GT running them. I'll mull it over awhile yet. Plenty of people extoling the Titan, and yet plenty more saying it's overkill, but admitting they would never need one - I plan on doing quite a lot with Element3D for AE, and indeed use AE a lot. But then the big thing with AE is RAM.

I appreciate all the comments - certainly helping to home in on the choice.

Julian.

UnCommonGrafx
03-09-2014, 11:43 AM
Titan.

Yes.

The 6gbs is a bit of future-proofing and the speed is excellent. When I had my display hooked up to it, it was awkwardly fast. Can't wait to have three in a box with one as the display.

They are over-hyped, sure. But so are most champion-caliber entities.

JBT27
03-09-2014, 12:50 PM
I can't deny that I was leaning towards one at the start. There is plenty of sense being offered here, and I am still weighing it up, but like those fictional detectives (and possibly real ones) I have a hunch it is the one that I want and need, despite the over-hyping. I also keep a machine for some time, as you can see from my signature - the leap between that one and this new one will be just plain silly :)

Julian.

UnCommonGrafx
03-09-2014, 01:42 PM
I also keep a machine for some time, as you can see from my signature - the leap between that one and this new one will be just plain silly :)

Julian.

I do, too. EVGA offered a ten year warranty, too, so I feel pretty good about the investment. I don't overclock so no big deal on making it that long, I figure.

spherical
03-09-2014, 09:27 PM
I also keep a machine for some time, as you can see from my signature - the leap between that one and this new one will be just plain silly

Hear that. I ran a dual Athlon XP (modded to be MP) / 512MB RAM motherboard and an AGP 4X bus 6800 GT, right up until late last year, when I could finally afford to build a new workstation. Going from 32-bit non-SSE2 chips to 4.5 GHz 64-bit 12 thread on 32 GB RAM and a GTX670 4GB card was like a dream. LightWave worked great on that box. I see lots of worrisome posts about "Is this or that good for running LightWave?" Hell yeah. If that machine could tumble models and render stuff, anything you can buy today will surely handle it. Might've been slow by today's standards (some F9's I could go make a sandwich... and eat it) but it worked just fine. I take care of my stuff and it takes care of me.

JBT27
03-11-2014, 01:49 PM
Hear that. I ran a dual Athlon XP (modded to be MP) / 512MB RAM motherboard and an AGP 4X bus 6800 GT, right up until late last year, when I could finally afford to build a new workstation. Going from 32-bit non-SSE2 chips to 4.5 GHz 64-bit 12 thread on 32 GB RAM and a GTX670 4GB card was like a dream. LightWave worked great on that box. I see lots of worrisome posts about "Is this or that good for running LightWave?" Hell yeah. If that machine could tumble models and render stuff, anything you can buy today will surely handle it. Might've been slow by today's standards (some F9's I could go make a sandwich... and eat it) but it worked just fine. I take care of my stuff and it takes care of me.

That makes sense to me. We just don't have the capital to be buying this and that every year or other year, try this or that - love to, but right now and for the forseeable, not going to happen. That said, this old 8Gb quad core from 2009 has done well and still delivers, but I am hitting the buffers too many times now - big HD comps, HD edits, big data sets in LW, wanting to do stuff that this machine simply can't handle anyway ... that and the fact that our other machine has gone pear-shaped with what looks like a failing memory manager on the mobo, and luckily we're busy enough to just about have the loot, it's time to step up a notch or ten. Hence the thinking on the Titan - read plenty about it, and other cards, certainly pricey, not the fastest, but it's solid, well thought of, and the 6Gb DDR5 is very appealing. I reckon that could be another four or five years of usage ... famous last words :)

Julian.

Carlo_Jongen
03-12-2014, 05:53 AM
I'm sharing some of my questions to OctaneRender and their answers, I think it fits perfect in this thread:


Question:

The Nvidia Geforce GTX Titan and GTX 780 are rated 3.5 for there compute Capability. And the Geforce GTX 690 is rated at 3.0 for its compute Capability.
But the GTX 690 has more Cuda cores then the Titan and the 780.

- To get the maximum speed out of Octane Render should we look at the higher number next to the compute Capability?
- Or should we look at the amount of Cuda Cores?
- Is the GTX 690 faster then the GTX 780 and Titan because of its amount of Cuda cores?
- Is it possible that a scene can not be rendered with Octane because of a limitation in the Ram on the video card?

Answer:
We usually look at the amount of cuda cores but there are many factors affecting render speed. The compute capability refers to the different features/specifications for the gpu architecture, the compute capability version also gives you an idea of the kinds of calculations the gpu is capable of and how many cores are allocated for such calculations. Of course, if more cuda cores are available in a gpu, then more cuda cores can be allocated for each calculation taking place (http://en.wikipedia.org/wiki/CUDA#Limitations).

Although the 690 has more cuda cores than the Titan, it is not necessarily faster than the Titan when it comes to rendering with Octane. The overall performance of the Titan is also better compared to the 690.

The video ram is very important. Scene resources must fit into the vram of the gpu and the resources must be available at all times during the rendering, otherwise the render will fail.

Question:

What exactly get's pumped into the video ram? Is it the total of the textures, objects and scene file that goes to the vram?
For example, when I render a scene in Lightwave ScreamerNet ( so with CPU ). My RAM fill's up to 12G to 16G. Does this mean I can't render these scenes with Octane? Or is the RAM on the motherboard filled with other tasks then the vram?

Answer:

I'm no expert on gpu architecture but here are just some facts about them. The vram stores all the data that the GPU currently needs so that it can access that information locally, making it faster than if it were to find data from system ram or hard disk. The GPU also has its own scheduler, letting it know which data is currently needed and when to swap data when its not currently needed. This includes geometry data, textures, frame buffers etc. Also, the higher the resolution the GPU needs to render, the more data it stores to be able to continue its calculations.

3GB vram is usually fine for most things, but big textures need much more vram. Typically, a 1000x1000 32bit image can take 4Mb ram, while a 5000x5000 32bit image can take 100mb vram. If you have more complex scenes with large very detailed textures plus a large hdri, the Titans would be a good option.
Here is an example of a scene: http://render.otoy.com/forum/viewtopic.php?f=6&t=37244, the user mentioned that it used up 5.3 gb of Vram and 25 of ram.

Finally, Yes you said it right. The system ram does store more kinds of data in it encompassing the entire pc. So not just for CPU-based rendering but for all other applications running on the machine, not to mention the OS.


I hope this information is useful to somebody. Please feel free to add more information or experience to this thread. Because sometimes the specs of these cards become to technical, one might loose track in finding a good videocard and this might help.

Cheers,
Carlo


Freelance Lightwave Animation at www.c3d.be

JBT27
03-12-2014, 06:07 AM
That is useful - thanks for posting. Especially as I just ordered a Titan :)

Julian.

Carlo_Jongen
03-12-2014, 06:15 AM
Hi Julian,

Maybe you can share your LW & AE experience once you have the Titan up and running?

Greets,
Carlo

JBT27
03-12-2014, 06:25 AM
Hi Julian,

Maybe you can share your LW & AE experience once you have the Titan up and running?

Greets,
Carlo

Yes, be happy to. If the hardware gurus can guide me as to the best way to define it - I am more likely to be subjective, throwing a ton of data at it and reporting what happens, how fast (or not), and so on, but yes - as soon as I have it up and running and get the feel of it, I'll post some experiences of it.

Julian.