R14 vs 12.5.5 significant performance degradation

I've taken an old project, just to test the famous "10x performance improvement". Nothing fancy – 10-12 nodes (secondary keyers mostly), without noise reduction or OFX. The sources are from Alexa, anamorphic 2944x2160 ProRes 4444 XQ, timeline is 1080p, delivery in source resolution DPX sequence.
In 12.5.5 with this project I have a real-time playback in Color page and rendering speed in Delivery is about 10-11 fps. In R14 I couldn't get more that 6 fps, both in playback and delivery. That's totally disappointing

My hardware: customized MacPro 5,1 (12 cores 2x3.06 Xeon 5675), 56Gb RAM, GeForce GTX 980Ti with 6Gb. Latest NVidia drivers (378.05.05.25f01) and CUDA (8.0.90) versions.
On the "old" Resolve 12.5.5 I have real-time performance on 95% of my projects. So what's happening? Where's this 4x performance degradation came from?

Hello Andrew

It's R14 Studio that has the performance improvement. Concerning general performance, if you are running anti-virus software, be sure to add the Resolve application folder and folders used for input/output to the AV app "white list" - the AV app will slow everything down really badly as it tries to perform a real-time check on any of those.

Hello Andrew

It's R14 Studio that has the performance improvement. Concerning general performance, if you are running anti-virus software, be sure to add the Resolve application folder and folders used for input/output to the AV app "white list" - the AV app will slow everything down really badly as it tries to perform a real-time check on any of those.

Forgot to mention, I'm using Studio version. On Mac, macOS 10.12.6. So there's no anti-virus or other "enhancement" software present

I'm seeing the same issues, especially with rendering, I've gone back to 12.5 for now

Maybe it's an issue with OFX plugins? I normally don't use OFX plugins and I have a big performance improvement in R14. I don't have a big graphics card. It sometimes crashes using the new stabilizer, but it's probably because my graphics card doesn't have enough RAM. Overall it works much better for me.

[Rohit Gupta]

I've taken an old project, just to test the famous "10x performance improvement". Nothing fancy – 10-12 nodes (secondary keyers mostly), without noise reduction or OFX. The sources are from Alexa, anamorphic 2944x2160 ProRes 4444 XQ, timeline is 1080p, delivery in source resolution DPX sequence.
In 12.5.5 with this project I have a real-time playback in Color page and rendering speed in Delivery is about 10-11 fps. In R14 I couldn't get more that 6 fps, both in playback and delivery. That's totally disappointing

My hardware: customized MacPro 5,1 (12 cores 2x3.06 Xeon 5675), 56Gb RAM, GeForce GTX 980Ti with 6Gb. Latest NVidia drivers (378.05.05.25f01) and CUDA (8.0.90) versions.
On the "old" Resolve 12.5.5 I have real-time performance on 95% of my projects. So what's happening? Where's this 4x performance degradation came from?

In Preferences, switch GPU type to Manual and CUDA. Resolve 14 defaults to OpenCL on single GPU NVIDIA systems.

Switching to CUDA and MANUAL gives a HUGE performance boost...

Whoa! Just did my first render test. I'm on 2-3 year old SuperMicro dual Xeon system with dual TitanX Black GPU's.

v12 would render 2K Prores from Alexa/Amira with <8 nodes (dallies) to 1080 DNxHD QT at around 100fps. Setting the system to CUDA and manual it is rendering out at around 240fps!

In Preferences, switch GPU type to Manual and CUDA. Resolve 14 defaults to OpenCL on single GPU NVIDIA systems.

Interesting. Why's that if the recommendation here is to manually switch to CUDA?

In Preferences, switch GPU type to Manual and CUDA. Resolve 14 defaults to OpenCL on single GPU NVIDIA systems.

Thanks, that's helping, but not much. I'm still not getting a real-time playback. Funny though, if I play the whole timeline, it slowly (much slower than R12) gets to 25 fps and the the second time in the loop it plays ok. But on the single shot I can't get above 23 fps. And the playback speed is increasing very slow comparing to R12.

So, in general, I have lower performance in R14 than in R12 on the same project / same hardware. And my setup, while not brand new, is powerful enough for R12. So I guess I'd have to stick with R12 for now

So much for 10x improvement marketing BS

You may have some other bottlenecks in your systems causing you not being able to benefit from the performance increase lots of other are definately experiencing.

You can not immediately claim BS as otherwise nobody would experiece this. For me the difference was between a good fast car and a ferrari.
But to stick with the Ferrari analogy, would you immediately blame the Ferrari dealer if it does not drive the advertised 380 km/hr when you drive on a forest sandy road ?

So depending on your hardware and setup , you mileage may vary, from very little to a lot of improvement.
That would be the only thing BM could have probable said as an extra, but DR has always had a hunger for more beefy hardware, so nothing new.

p.s. Not always obvious but make sure you also have "performance mode " selected in the menu.

You may have some other bottlenecks in your systems causing you not being able to benefit from the performance increase lots of other are definately experiencing.

So those [alleged] bottlenecks affecting only R14???
I have decent real-time performance in Resolve 12.5.5. Performance of the Resolve 14 on the same system with the same project is lower.

Your Ferrari metaphor is incorrect. My new, 10x powerful (sic) Ferrari is going slower on the same road I'm riding my old one, with the same gas and load.

I don't care about improvement. I never believed in any "10x faster" promises. It's deterioration that bothers me.

I said "may have".
It is always difficult to pinpoint performance issues. Have had lots of cases where ate the exact same system setup the same software behaves different. Only after lengthy troubleshooting i found that some minute difference made the big difference.
Here the problem is more complex as we try to compare 2 versions of the same software that work internaly different as a complete new graphics engine and audio engine.
So what works fine on a certain system with one engine may work completely different on the next.

Recent examples of these sort of things are when you compare performance of a graphics card like the new 1080ti's. The card can increase you performance by a large factor, but if you put the same card in a system with pci bus contention , slower cpu and memory, then that same fast card can not perform to its expected limits.

Thats is more what i was pointing to.

Lets hope you find what is holding resolve back on your system as i do feel looking at your specs you should also feel a significant increase.

[Peter Chamberlain]

Andrew, id like to get to the bottom of this issue.

Can you tell me what playback speed you have on one clip, one say 10 seconds in length on loop, with color correction bypassed.

Then add your nodes to see which parameter is causing the slowdown.

If you don't get real time with color correction bypassed we can look elsewhere.

Andrew, id like to get to the bottom of this issue.

Can you tell me what playback speed you have on one clip, one say 10 seconds in length on loop, with color correction bypassed.

Then add your nodes to see which parameter is causing the slowdown.

If you don't get real time with color correction bypassed we can look elsewhere.

Without corrections I'm getting 25fps almost instantly.
Enabling each node is causing small slowdown in fps acceleration. The order of nodes doesn't matter. With 7 nodes I still have 25fps, stabilizing at ~2.5s mark. After 8 nodes the playback speed starting to behave funky. It gets up to 25fps and began fluctuating between 24.5-26 (!?) fps. The fps light is glowing red, and playback is a bit jerky. At 11th node the playback speed falls to 24.5fps and enabling all 13 nodes gives me 23-24 fps in the loop.
The correction is pretty simple, primaries, LogC LUT, static windows and qualifiers. No noise reduction or OFX. The source footage is ProRes 4444 XQ, 2944x2160 anamorphic, 25fps. Timeline is 1920x1080 25fps. Color science is YRGB.

Thank you for helping. Ask me if you need more information. I'm eager to switch to the new version

Andrew,

Just to tripple-check as you never replied to that. Do you have "performance mode" active in the menu ?
As its pretty important.

Andrew,

Just to tripple-check as you never replied to that. Do you have "performance mode" active in the menu ?
As its pretty important.

Sorry, I must've missed the question.
Yes, performance mode is active. When it's turned off playback rate is reduced by ~2fps.

Ah ok at least that is out of the way.

And if you do a standard candle test on 12 and on 14 do you see the same thing happening.
This should be faster to significantly faster between 12 and 14. At least we have a base to work with.
It is largely IO independant so should show if there are issue that in general slow down or speed up between the 2 releases.
The more baseline standard stuff we have the better so we can work upward from that untill we find the culprit.
If this candle test does not show improvement then there is definately something off on your system we need to find.

[Peter Chamberlain]

Andrew,

Just to tripple-check as you never replied to that. Do you have "performance mode" active in the menu ?
As its pretty important.

Sorry, I must've missed the question.
Yes, performance mode is active. When it's turned off playback rate is reduced by ~2fps.

Can u post or PM a link to the project please. No media or stills needed.

Ah ok at least that is out of the way.

And if you do a standard candle test on 12 and on 14 do you see the same thing happening.
This should be faster to significantly faster between 12 and 14. At least we have a base to work with.
It is largely IO independant so should show if there are issue that in general slow down or speed up between the 2 releases.
The more baseline standard stuff we have the better so we can work upward from that untill we find the culprit.
If this candle test does not show improvement then there is definately something off on your system we need to find.

I'm really confused
The standard candle test showing a significant improvement. Basically one step higher, i.e. in R14 I'm getting the same performance on 30 Blur and 6 TNR as on 18 Blur and 4 TNR respectively in R12. However the playback rate build-up in R14 is about x2 slower (on the same grades).

But on any real life project (I've tried 3 different projects) I see the same degradation as in my original test project
The most confusing thing is that on some shots I'm having a faster than realtime playback rate (fluctuating between 25.5-26fps), which is jerky, i.e. unusable.

Couldn't resist sorry

But at least we are getting somewhere. So Resolve "is able" to get the improved performance , but there are some factors in your real live stuff that drops it back for some yet unknown reason.
Lets see what BM can come up with.

In the mean time if i where you i would do single node /node by node comparison troubleshooting using different things. I would gable to have a look at resizing. So having Higher format source footage then timeline and see the effect of these things. Check the resizing preferences between the 2 releases and projects. (i believe there where some modifications in quality settings that can effect performance.
Anyway, sounds like a long troubleshoot road ahead.

Couldn't resist sorry

Check the resizing preferences between the 2 releases and projects. (i believe there where some modifications in quality settings that can effect performance.
Anyway, sounds like a long troubleshoot road ahead.

Yeah, that could be an issue. In R12 I had all settings set at "Optimize for playback". But this option is gone in R14. The fact that you can't have both versions installed simultaneously doesn't helping either, doing the reinstall every time is making this process even more fun
I have some small project upcoming, so I'll try to grade it in R14 from scratch, to get the general feeling and whatnot. Maybe there's some issues with importing old projects or something. And maybe BM will find that something and we'll have it fixed in the next update

Thank you, guys, for your help, I'm feeling much better now
Sorry for some harsh words, but I was really frustrated with the first impression.

No worries on the harsh words man. Completely understandable. Have been there.. Thats why we are here, just to try and help each other , as what you face others could also face. Does not always work but it the intention that counts.

Keep us posted if you find out anything. Pitty you where (aparantly) not yet using SQL database as otherwise its on mac with SQL pretty simple to switch back and forward between releases. Each with their own db and the install in its own directory. Zero problems. Reinstalls sounds like as much fun as a painfull tooth.

Keep us posted if you find out anything. Pitty you where (aparantly) not yet using SQL database as otherwise its on mac with SQL pretty simple to switch back and forward between releases. Each with their own db and the install in its own directory. Zero problems. Reinstalls sounds like as much fun as a painfull tooth.

I was using SQL db in the past, up until R9 I think. I can't remember the reason why I've switched to disk based, but I'm having now a separate db for R14. So the database is not a problem. I'm not quite understand, how do you keep two versions separate from each other. Yes, you could install one, move it to the different folder and then install another. But what about preferences folder (the one in /Library/Application Support/)? They share it? Or you just have two copies and switch them?

I dont touch anything on preferences. The system/user pref structure is overhauled and completely different between the 2 releases and have yet to see a single conflict. Project prefs are saved in the DB with the project.
I can start 12 then 14 then 12, not a single issue and only thing i have is as said a separate db (in the same db server) and a separate install directory (just renamed after each install).

edit : just a disclaimer as surely this is not standard supported or recommended by BM. Just giving you my experiences. Have been using it during the entire beta as such with sometimes 6-10 different versions next to each other to facilitate testing my own developments on each different release.
And always had full backups of everyting, the golden rule when testing.