1080 Ti still rendering after 4 days (good)

[AndreeOnline]

Four days ago I went away on a family trip. To have something interesting to come back to, I started a simple clay render animation from my apartment that I had modelled in C4D and then exported to Maxwell Render 4. I set it up so that it wouldn't finish before I came back.

The CUDA render support in Maxwell 4 isn't complete and on Mac it's 'experimental'. I've had hits and misses.

The good thing here is that I started the render before I left and it's still going strong! I had half expected a crash for one reason or another. The computer is extremely quiet. I'm using a MSI Gaming 1080 Ti will two pretty large fans on ball bearings.

iStat Menus has the GPU working properly.

Anyway... I just wanted to share this as I feel it's a good sign of stability. On the game side I'm running F1 2016 and that can be a bit flaky with the occasional hang. Hopefully that improves with one or two patches.

 

[Flint Ironstag]

Four days ago I went away on a family trip. To have something interesting to come back to, I started a simple clay render animation from my apartment that I had models in C4D and then exported to Maxwell Render 4. I set it up so that it wouldn't finish before I came back.

The CUDA render support in Maxwell 4 isn't complete and on Mac it's 'experimental'. I've had hits and misses.

The good thing here is that I started the render before I left and it's still going strong! I had half expected a crash for one reason or another. The computer is extremely quiet. I'm using a MSI Gaming 1080 Ti will two pretty large fans on ball bearings.

View attachment 697880

iStat Menus has the GPU working properly.

Anyway... I just wanted to share this as I feel it's a good sign of stability. On the game side I'm running F1 2016 and that can be a bit flaky with the occasional hang. Hopefully that improves with one or two patches.

Nice - are you on cMP or nMP?

 

[AidenShaw]

Nice - are you on cMP or nMP?

Since it's rather hard to connect a 1080Ti to a trash can, and he didn't say eGPU, pretty safe to assume a cMP.

It would be nice if he'd have mentioned the config. I went back through a lot of AndréeOnlines posts and it was hard to figure out exactly what he is running today.

 

[flowrider]

Nice - are you on cMP or nMP?

This should answer your question:

https://forums.macrumors.com/threads/pixlas-4-1-mac-pro-mod.1859652/

Lou

 

[owbp]

The answer is in the title, if it is STILL rendering it has to be cMP.
nMP would crash at least dozen times by now.

 

[AndreeOnline]

Hehe... Yes, still the same 4.1-->5.1 cMP with internal 1080 Ti.

 

[PowerMike G5]

This is really good to hear regarding stability.

Any chance you can provide some benchmarks on the card when you have time (Geekbench, Unigine Heaven/Valley, Luxmark, etc)?

I'm entertaining to the idea of getting this card, but want to compare it to my current Titan X.

 

[AndreeOnline]

Any chance you can provide some benchmarks on the card when you have time (Geekbench, Unigine Heaven/Valley, Luxmark, etc)?

That was the first thing I posted when I installed the card. There are other threads for that, but I'll give you a few screen grabs:



GeekBench 4.1 OpenCL doesn't run to completion. This is Metal and CUDA.





Make sure to compare apples to apples: people post scores in various resolutions for Heaven.

 

[PowerMike G5]

Thank you for these. Very helpful!

 

[AidenShaw]

Thank you for these. Very helpful!

Although this is talking about a very different problem space (deep learning), you might find some interesting points in http://timdettmers.com/2017/04/09/which-gpu-for-deep-learning/ .

In particular one comment he makes is that it often doesn't make sense to upgrade from a top Gen-1 card to a top Gen-current card. Here's one rough chart:

Of course, every situation is different - and some people may be happy with spending extra for a modest improvement (e.g. if you have to meet tight deadlines, double the prices for a 15% improvement might be perfect).

With Volta on the horizon, rushing into Pascal might not be wise.

NB: You might have seen a post that I've ordered twenty 1080Ti cards to upgrade Titan X (Maxwell) cards. There are a couple of extenuating circumstances here.

• Maxwell has a max of 1 TiB of memory accessibility, so we have some systems that we've pulled DIMMs from to bring them down to 992 GiB of RAM. Once the 1080Ti cards go in, those systems will be bumped to 2 TiB of RAM.
• The displaced Titan X (Maxwell) cards will move into systems with weaker cards, and the displaced weaker cards will move into systems without GPUs. After the shuffle is done, I'll have one GTX 970 left over.

 

[PowerMike G5]

Thanks Aiden. Yes, this is partly why I'm on the fence. No doubt the 1080 Ti is faster, but whether cost to do so is worth it is still under analysis for me.

I'm mostly interested in GPU acceleration for various effects, color grading, etc of DCI 4K footage I work with consistently. I've found the Titan X to provide a good experience with Adobe CC.

But some effects like NeatVideo can utltize all CPU cores and all GPU cores and VRAM. And I do use this pretty extensively, so I am always interested in ways too decrease render times.

And the occasional League of Legends game in DCI 4K, which mostly plays at 60fps already with the Titan X, except in heavy fight scenes.

 

[AidenShaw]

Four days ago I went away on a family trip. To have something interesting to come back to, I started a simple clay render animation from my apartment that I had modelled in C4D and then exported to Maxwell Render 4. I set it up so that it wouldn't finish before I came back.

The CUDA render support in Maxwell 4 isn't complete and on Mac it's 'experimental'. I've had hits and misses.

The good thing here is that I started the render before I left and it's still going strong! I had half expected a crash for one reason or another. The computer is extremely quiet. I'm using a MSI Gaming 1080 Ti will two pretty large fans on ball bearings.

View attachment 697880

iStat Menus has the GPU working properly.

Anyway... I just wanted to share this as I feel it's a good sign of stability. On the game side I'm running F1 2016 and that can be a bit flaky with the occasional hang. Hopefully that improves with one or two patches.

Still rendering?

 

[SoyCapitanSoyCapitan]

Still rendering?

He refunded the card

 

[AndreeOnline]

Still rendering?

I'm sure it would have. If it's stable for four days non stop, it's good in my book. I stopped it myself.

The CUDA accelerated Maxwell Render (stand alone) was one of the highlights with the card. That and CUDA speed in DaVinci Resolve.

But the card didn't work in Maxwell's FIRE render inside Cinema 4D, which would mean extremely fast look-dev. But this is more a Maxwell issue and that their GPU support is pretty new. Also, GPU rendering in Maxwell isn't fully supported.

He refunded the card

Gaming isn't my main thing by a long shot, but I did some F1 2016 (new game) driving and the game kept hanging there (also: the RX480 is faster than the 1080 Ti in that game). BruceX was 1 minute and a couple of OpenCL benchmarks didn't work properly. <--- these things combined left a feeling in my gut that the drivers aren't quite there yet. Maybe they never will be. I didn't want to wait and find out, so I returned the card on the last possible day. I just don't like computers that hang or are unstable. That's a no-go for me personally.

I'm going to check out AMD Vega, but I also think the Radeon Pro Duo looks interesting. We'll see what happens now. I'll finalise my solution when DaVinci 14 is out of beta.

I do think it's a strong card and for someone doing DaVinci and Octane render it must be pretty sweet.

 

[AidenShaw]

BruceX was 1 minute and a couple of OpenCL benchmarks didn't work properly. <--- these things combined left a feeling in my gut that the drivers aren't quite there yet. Maybe they never will be.

I suspect that will be the case until (unless?) Apple supports Nvidia cards in the mMP.

Once Apple has some skin in the game of driver support the "perpetual beta" should end rather quickly.

 

[CapnDavey]

4 days that must be some large stuff to take that long hehe

 

[Prince134]

I'm sure it would have. If it's stable for four days non stop, it's good in my book. I stopped it myself.

The CUDA accelerated Maxwell Render (stand alone) was one of the highlights with the card. That and CUDA speed in DaVinci Resolve.

But the card didn't work in Maxwell's FIRE render inside Cinema 4D, which would mean extremely fast look-dev. But this is more a Maxwell issue and that their GPU support is pretty new. Also, GPU rendering in Maxwell isn't fully supported.



Gaming isn't my main thing by a long shot, but I did some F1 2016 (new game) driving and the game kept hanging there (also: the RX480 is faster than the 1080 Ti in that game). BruceX was 1 minute and a couple of OpenCL benchmarks didn't work properly. <--- these things combined left a feeling in my gut that the drivers aren't quite there yet. Maybe they never will be. I didn't want to wait and find out, so I returned the card on the last possible day. I just don't like computers that hang or are unstable. That's a no-go for me personally.

I'm going to check out AMD Vega, but I also think the Radeon Pro Duo looks interesting. We'll see what happens now. I'll finalise my solution when DaVinci 14 is out of beta.

I do think it's a strong card and for someone doing DaVinci and Octane render it must be pretty sweet.

My experience with Titan X pascal, the BruceX is like 30-32 seconds. 1080 Ti should be close. I tested with 2x x5690, 96GB. Did you have single CPU?
[doublepost=1494023993][/doublepost]Yes it was 50+ seconds before FCP 10.3.3. If you update your final cut pro to 10.3.3 you should see it runs BruceX at 30 seconds. The FXP 10.3.3 came out right after Nvidia provides the web driver for Pascal.

 

[irked]

I'm sure it would have. If it's stable for four days non stop, it's good in my book. I stopped it myself.

Did you play it back? If i left a spittoon mac pro rendering for four days i'd have to scrub through frame-by-frame and look for glitches.

 

[Gazember]

One can get those BruceX numbers with a dirty old 7950 for $200 (most recent FCPX on Sierra). Not sure if I want to upgrade to a Pascal card anymore. I never play games and FCPX is about the only application I need the Macs for.

 

[Asgorath]

One can get those BruceX numbers with a dirty old 7950 for $200 (most recent FCPX on Sierra). Not sure if I want to upgrade to a Pascal card anymore. I never play games and FCPX is about the only application I need the Macs for.

That's more a function of the fact you're using an ancient Mac Pro, I was getting 14 seconds on a GTX TITAN X (GM200) in my Hackintosh system.

 

[AidenShaw]

That's more a function of the fact you're using an ancient Mac Pro, I was getting 14 seconds on a GTX TITAN X (GM200) in my Hackintosh system.

That's the sad truth.

The cMPs have old processors, old RAM, old PCIe, old SATA, old USB, old 1394 (but few care about 1394 any more).

I got a notice this week that HPE was moving all of their E5-vx6xx v2 servers to legacy support. What Apple is selling as the "New Mac Pro" is on life-support from other tier 1 vendors. Ivy Bridge is legacy at HPE - but Apple still thinks that it's leading edge.

To quote our liar-in-chief: SAD!

 

[Prince134]

That's more a function of the fact you're using an ancient Mac Pro, I was getting 14 seconds on a GTX TITAN X (GM200) in my Hackintosh system.

It's not because of the ancient Mac Pro. I was getting 18-20 seconds on GM200 Titan X as well. But that was with El Capitan.

With Sierra, it became 30 seconds. It's 50% FCP efficiency loss in Sierra, comparing to the El Capitan, as was discussed before.

However before new FCP 10.3.3 update, i.e., in the beginning 3 days (that people still have fop 10.3.2) with Nvidia's web driver for Pascal, the BruceX test needs 54 seconds.

Barefeats test results were before that FCP 10.3.3 update. It's faster now actually.

 

[joebclash]

That's the sad truth.

The cMPs have old processors, old RAM, old PCIe, old SATA, old USB, old 1394 (but few care about 1394 any more).

I got a notice this week that HPE was moving all of their E5-vx6xx v2 servers to legacy support. What Apple is selling as the "New Mac Pro" is on life-support from other tier 1 vendors. Ivy Bridge is legacy at HPE - but Apple still thinks that it's leading edge.

To quote our liar-in-chief: SAD!

Yet they are still selling mac pros that completely outdated for over 3K. Every time I watch that video I want to punch my monitor.

 

[AndreeOnline]

4 days that must be some large stuff to take that long hehe

No problem at all, unfortunately. Depending on scene and resolution, it's easy to set it up so that a frame of production quality video takes an hour to render. 24h then becomes one second of animation. So four days means 4 seconds of 'movie'. I'm not talking about video encoding here, but output from an unbiased renderer.

But like I said, this was by design since I wanted a job that kept on going. I could also have set it up to only use 4 min per frame, which would have resulted in images with a lower sampling level.

Did you play it back? If i left a spittoon mac pro rendering for four days i'd have to scrub through frame-by-frame and look for glitches.

There shouldn't be glitches. What you describe wouldn't work for me. It was the reason I had to swap my R9 280X since it introduced artefacts when rendering out from Resolve (probably some memory corruption). At the end of the day it doesn't matter if it's glitchy software or faulty hardware—you need to be able to trust it. But naturally anyone charging for a product better watch the final product a couple of times to make sure everything is OK. Image sequences are good that way: if you need to make a small correction, just re-export those exact frames.

 

[kennyman]

My experience with Titan X pascal, the BruceX is like 30-32 seconds. 1080 Ti should be close. I tested with 2x x5690, 96GB. Did you have single CPU?
[doublepost=1494023993][/doublepost]Yes it was 50+ seconds before FCP 10.3.3. If you update your final cut pro to 10.3.3 you should see it runs BruceX at 30 seconds. The FXP 10.3.3 came out right after Nvidia provides the web driver for Pascal.

So to conclude - does that mean that Titan X Pascal (2016) and 1080 Ti are now on par on FXP 10.3.3?

I can confirm on Resolve, there is virtually no difference. 1080 Ti works well and performance is similar to a Titan X Pascal (2016). We have 2 cMPs at work, both are equipped with 1 x 1080 Ti FE, no problem with Adobe and Resolve.

I am currently using a 980 Ti, just saw a used EVGA 1080 Ti FE for around $613 on Amazon. If FCP 10.3.3 is optimized for Pascal, then I think I will get a 1080 Ti for Resolve and FCP, I can throw the 980 Ti on eBay to recoup some $$$