2nd GPU performance improvements

Get answers to your questions about color grading, editing and finishing with DaVinci Resolve.
  • Author
  • Message
Offline
User avatar

MikeMeagher

  • Posts: 339
  • Joined: Thu Oct 03, 2013 8:21 pm
  • Location: Bellingham, WA. USA

Re: 2nd GPU performance improvements

PostMon Aug 28, 2017 9:15 pm

Thank you Peter for the great clarification.

I have one more question for clarification: When working in the Delivery page, and when a timeline job is created for rendering, when that job is rendered is the CPU or GPU card(s) or both tasked to perform the decoding and rendering of the sourcemedia into the final rendered file? (assuming and not assuming that I have created Optimized media)
I am curious to know if the GPU card(s) contribute to this rendering task or if it is solely a CPU/system Ram/storage system process or some of both?
Mike Meagher Filmmaker
Win10 64 pro I7 6core ASUS X99 64 GB RAM
M.2 system drive.
2x SSDs in Raid0 cache.
3x 7200 HHDs in raid0 as video source storage
Two EVGA1080TI cards 422.19 Studio
Decklink MiniM 15.1 4K HDMI to BENQ WS271
Resolve 16.1 studio
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 13875
  • Joined: Wed Aug 22, 2012 7:08 am

Re: 2nd GPU performance improvements

PostTue Aug 29, 2017 3:03 am

Im not sure whats not clear in my message. CPUs do not process the images in Resolve. That's a GPU process. The CPUs will do the encoding into the codec you select. It doesn't matter which page you are on, the GPU processes the images.
DaVinci Resolve Product Manager
Offline
User avatar

MikeMeagher

  • Posts: 339
  • Joined: Thu Oct 03, 2013 8:21 pm
  • Location: Bellingham, WA. USA

Re: 2nd GPU performance improvements

PostTue Aug 29, 2017 6:14 am

Thank you.
Mike Meagher Filmmaker
Win10 64 pro I7 6core ASUS X99 64 GB RAM
M.2 system drive.
2x SSDs in Raid0 cache.
3x 7200 HHDs in raid0 as video source storage
Two EVGA1080TI cards 422.19 Studio
Decklink MiniM 15.1 4K HDMI to BENQ WS271
Resolve 16.1 studio
Offline

Justin Jackson

  • Posts: 670
  • Joined: Thu Apr 28, 2016 3:50 am

Re: 2nd GPU performance improvements

PostTue Nov 07, 2017 6:27 am

old thread.. i know.. but with regards to two identical cards, sli, etc.. i have the desire to get two high end gpus for sli gaming purposes, but also use it for resolve and fusion work. can i set my cards up in sli for gaming, but still get resolve to use the two cards.. or will the sli bridge thingy make it so resolve can not use both cards.. or even one card?

with two high powered gpus, does the studio version use both cards for image processing if you have a 3rd slower card for gui... such that it would parallelize the use of the two higher end cards.. or will only one get used in that configuration.
Custom DIY AMD1950x 16-core/32-thread, liquid cooled, 64GB 3600Mhz RAM, 950Pro-512GB NVMe os/apps, 2x500GB 850 Evo RAID 0 SATA3, Zotac 1070 8GB video, USB 3.1Gen2 RAID0 2x4TB, 2x2TB Crucial MX500 SSD SATA3.
Offline
User avatar

MikeMeagher

  • Posts: 339
  • Joined: Thu Oct 03, 2013 8:21 pm
  • Location: Bellingham, WA. USA

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 2:28 am

On my Win 10 pc i use the nvidia settings control app to disable SLI..i do that for when running resolve. Then I enable sli when i want to do some gaming with my boys..they always win.
Mike Meagher Filmmaker
Win10 64 pro I7 6core ASUS X99 64 GB RAM
M.2 system drive.
2x SSDs in Raid0 cache.
3x 7200 HHDs in raid0 as video source storage
Two EVGA1080TI cards 422.19 Studio
Decklink MiniM 15.1 4K HDMI to BENQ WS271
Resolve 16.1 studio
Offline
User avatar

AndreeMarkefors

  • Posts: 281
  • Joined: Fri Sep 30, 2016 11:41 am
  • Location: Zürich, Switzerland

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 11:14 am

Peter Chamberlain wrote:There is some miss information in this thread, so again for clarity…


Really great post Peter. It's hard to imagine that more information could fit into fewer words. Now, if people would just read it until they understood it...

But I do have a question: is there a way to at least roughly predict how much performance a certain amount of RAM will buy in a certain timeline resolution? With frame sizes and amount of RAM as known quantities, can I predict how many nodes I can use before "out of memory"? Or, if the amount of serial nodes might be "unlimited", will parallell structures demand more?

Are there certain effects that you know need a disproportionate about of RAM?

I have my own theories, but prefer actual information, with so many variables in play. Do you have time to go a little bit deeper? I'm sure many would appreciate it, and we could spread the word.

Thanks
Current camera: Canon R5C
Mac Pro 2019: 12c | Dual W6800X Duo | DeckLink Mini Monitor 4K + LG 55C8
MacBook Pro M3 Max 16" 16/40, 64GB
Offline
User avatar

Jean Claude

  • Posts: 2973
  • Joined: Sun Jun 28, 2015 4:41 pm
  • Location: France

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 11:18 am

AndreeMarkefors wrote:
Peter Chamberlain wrote:There is some miss information in this thread, so again for clarity…


Really great post Peter. It's hard to imagine that more information could fit into fewer words. Now, if people would just read it until they understood it...

But I do have a question: is there a way to at least roughly predict how much performance a certain amount of RAM will buy in a certain timeline resolution? With frame sizes and amount of RAM as known quantities, can I predict how many nodes I can use before "out of memory"? Or, if the amount of serial nodes might be "unlimited", will parallell structures demand more?

Are there certain effects that you know need a disproportionate about of RAM?

I have my own theories, but prefer actual information, with so many variables in play. Do you have time to go a little bit deeper? I'm sure many would appreciate it, and we could spread the word.

Thanks


Hi,
For me, the temporal effects are the most demanding in terms of resources. No?
"Saying it is good, but doing it is better! "
Win10-1809 | Resolve Studio V16.1 | Fusion Studio V16.1 | Decklink 4K Extreme 6G | RTX 2080Ti 431.86 NSD driver! |
Offline
User avatar

Ricky Scharneck

  • Posts: 21
  • Joined: Mon Jul 03, 2017 5:54 pm

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 1:29 pm

I mainly work in the Color tab and Resolve bogs down with many mattes and adjustment (EXR is the worst in Resolve) Tif is way better it seems, but even so, hitting that magical number of adjustment nodes (and not ridiculously high mind you) makes Resolve work hard WITHOUT actually working... not much CPU activity, Memory activity, Data activity, nor much GPU activity on any of my TitanX cards. Sometimes performance can be frustrating, So some performance issues have nothing to do with hardware or setup and you probably won't gain much out of another card depending on what you do, or if Resolve goes on a holiday and you sit, waiting for it to catch up.. sorry for rant...
Offline

Willian Aleman

  • Posts: 356
  • Joined: Thu Aug 23, 2012 10:08 pm
  • Location: NYC, USA

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 5:51 pm

Ricky Scharneck wrote:So some performance issues have nothing to do with hardware or setup and you probably won't gain much out of another card depending on what you do, or if Resolve goes on a holiday and you sit, waiting for it to catch up.. sorry for rant...


Totally agree with your statement here. Although DR14 as an NLE is progressing in fast pace and Blackmagic design effort in putting all in one is changeable, and at the same time admirable, I haven't seen the 10 times faster performance Davinci Resolve 14 is advertises regardless the combined configuration when it comes to editing a long form feature film. Without doubts, the future is flattering.
Willian Aleman
New York City
USA

Resolve Studio
iMac (Retina 5K, 27-inch, 2020)
MacOS: 10.15.7 (19H2)
processor: 3.8 GHz 8-Core Intel Core i7
Memory: 40 GB 2667 MHz DDR4
Graphic: AMD Radeon Pro 5700 XT 16 GB
Offline
User avatar

Uli Plank

  • Posts: 21292
  • Joined: Fri Feb 08, 2013 2:48 am
  • Location: Germany and Indonesia

Re: 2nd GPU performance improvements

PostSat Nov 11, 2017 10:55 pm

Did you really believe it? That's typical marketing wording: "up to 10 times faster".

Some functions are probably that much faster, but it doesn't mean that you can run DR 14 on a machine 10 times weaker than needed before.

Just like Diesel engines are not cleaner than what you burn in them…
No, an iGPU is not enough, and you can't use HEVC 10 bit 4:2:2 in the free version.

Studio 18.6.5, MacOS 13.6.5
MacBook M1 Pro, 16 GPU cores, 32 GB RAM and iPhone 15 Pro
Speed Editor, UltraStudio Monitor 3G, iMac 2017, eGPU
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 13875
  • Joined: Wed Aug 22, 2012 7:08 am

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 1:34 am

We have measured a number of processes and transport actions that are 10 times faster in v14 but as expected not every process or every configuration or every resolution will see 10 fold gain, which is why we say “Up to”

The single GPU config has many optimization’s and overall has the best price/performance until you can’t get real-time. We still can measure performance gains with multiple GPUs up to 8 for many processes. For some items each additional GPU adds a linear gain with some diminishing returns past six GPUs on a price/performance ratio.

If you can’t see a benefit of four over two GPUs or even over a single it’s likely you’re just not using enough of the demanding processes or something else is limiting your throughput, usually CPU or disk I/O but can also be as simple as motherboard slot use. Thus the configuration guide is a worthy read if you are building a system for a specific performance.
DaVinci Resolve Product Manager
Offline
User avatar

Uli Plank

  • Posts: 21292
  • Joined: Fri Feb 08, 2013 2:48 am
  • Location: Germany and Indonesia

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 7:21 am

I can only second this, Resolve needs a very balanced system if you want to process demanding formats.

Since you are mentioning delivery in particular, disk I/O is an important factor. Never write back to the same system your footage is coming from. And with high data rate formats a small SSD RAID is not a bad idea.
No, an iGPU is not enough, and you can't use HEVC 10 bit 4:2:2 in the free version.

Studio 18.6.5, MacOS 13.6.5
MacBook M1 Pro, 16 GPU cores, 32 GB RAM and iPhone 15 Pro
Speed Editor, UltraStudio Monitor 3G, iMac 2017, eGPU
Offline
User avatar

Jean Claude

  • Posts: 2973
  • Joined: Sun Jun 28, 2015 4:41 pm
  • Location: France

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 9:46 am

If you want to do GPU load tests, create an HDD in RAM: it will avoid the I / O HDD bottlenecks.
"Saying it is good, but doing it is better! "
Win10-1809 | Resolve Studio V16.1 | Fusion Studio V16.1 | Decklink 4K Extreme 6G | RTX 2080Ti 431.86 NSD driver! |
Offline
User avatar

Ricky Scharneck

  • Posts: 21
  • Joined: Mon Jul 03, 2017 5:54 pm

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 11:48 am

Peter Chamberlain wrote:If you can’t see a benefit of four over two GPUs or even over a single it’s likely you’re just not using enough of the demanding processes or something else is limiting your throughput, usually CPU or disk I/O but can also be as simple as motherboard slot use. Thus the configuration guide is a worthy read if you are building a system for a specific performance.


I am afraid that isn't quite as cut and dry as that at times, I have a 8 core i7 running @ 4Ghz, DDR4 Dominator Pro at 2800Mhz, SSD OS drive 500read 450write (not super but enough), scratch/project M.2 at 3000read 2000write and multiple Titan X's using X8 pci. Yet when i use monitoring software, Resolve doesn't utilize anything enough to find a bottleneck anywhere, not in CPU, data rate, GPU or RAM. Nuke, Premiere, Aftereffects, Modo, etc... all work fine, yet Resolve starts to take ages to update or not at all when it hits that stage. I cannot speak too much about the other tabs but certainly color tab needs to be pushed harder by your QA with multiple mattes and EXR performance fixed :) I am not criticizing as Resolve is awesome and loving it thoroughly. I know you have heard it before, I will probably continually bring it up every update release as a reminder :) sorry for being a pain...
Offline

Dermot Shane

  • Posts: 2720
  • Joined: Tue Nov 11, 2014 6:48 pm
  • Location: Vancouver, Canada

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 6:52 pm

i'm not sure it's the software alone, my use case is; working in a EXR centered project today, DCi4k scope, many layers of embedded mattes, no issues really, no hangs , no waits, just pretty much a normal day, no diffrent from RED or ArriRAW in terms of artist responce

z820 / dual 2680v2s / dual 1080Ti's / 128gig / SAS array's - so more CPU / fast storage, and alot less GPU, but on x16 slots, all built to follow BMD's guidelines, all un-needed w10 services/garbage turned off, we followed Nucoda's guidelines for that
Offline

PeterMoretti

  • Posts: 928
  • Joined: Sat Aug 03, 2013 12:12 am

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 9:29 pm

Ricky Scharneck wrote:... and multiple Titan X's using X8 pci...

I'm sure you know more about this than I do, but don't you want to be using 16X pcie slots?
Resolve 14.3 Studio. GTX 970 with GeForce 390.77 driver. Desktop Video 10.9.10. Intensity Shuttle USB 3.0. Windows 10 Pro.
Offline

Dermot Shane

  • Posts: 2720
  • Joined: Tue Nov 11, 2014 6:48 pm
  • Location: Vancouver, Canada

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 10:12 pm

he has a i7, so not enough 16x lanes for dual, much less mutiple GPU's, may be a reason for less than stellar performance?
Offline
User avatar

waltervolpatto

  • Posts: 10502
  • Joined: Thu Feb 07, 2013 5:07 pm
  • Location: 1146 North Las Palmas Ave. Hollywood, California 90038 USA

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 10:57 pm

Dermot Shane wrote:i'm not sure it's the software alone, my use case is; working in a EXR centered project today, DCi4k scope, many layers of embedded mattes, no issues really, no hangs , no waits, just pretty much a normal day, no diffrent from RED or ArriRAW in terms of artist responce

z820 / dual 2680v2s / dual 1080Ti's / 128gig / SAS array's - so more CPU / fast storage, and alot less GPU, but on x16 slots, all built to follow BMD's guidelines, all un-needed w10 services/garbage turned off, we followed Nucoda's guidelines for that


Dermot, do you have a link for the W10 garbage clean?
W10-19043.1645- Supermicro MB C9X299-PGF - RAM 128GB CPU i9-10980XE 16c 4.3GHz (Oc) Water cooled
Decklink Studio 4K (12.3)
Resolve 18.5.1 / fusion studio 18
GPU 3090ti drivers 512.59 studio
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 13875
  • Joined: Wed Aug 22, 2012 7:08 am

Re: 2nd GPU performance improvements

PostSun Nov 12, 2017 11:17 pm

Ricky Scharneck wrote:
Peter Chamberlain wrote:If you can’t see a benefit of four over two GPUs or even over a single it’s likely you’re just not using enough of the demanding processes or something else is limiting your throughput, usually CPU or disk I/O but can also be as simple as motherboard slot use. Thus the configuration guide is a worthy read if you are building a system for a specific performance.


I am afraid that isn't quite as cut and dry as that at times, I have a 8 core i7 running @ 4Ghz, DDR4 Dominator Pro at 2800Mhz, SSD OS drive 500read 450write (not super but enough), scratch/project M.2 at 3000read 2000write and multiple Titan X's using X8 pci. Yet when i use monitoring software, Resolve doesn't utilize anything enough to find a bottleneck anywhere, not in CPU, data rate, GPU or RAM. Nuke, Premiere, Aftereffects, Modo, etc... all work fine, yet Resolve starts to take ages to update or not at all when it hits that stage. I cannot speak too much about the other tabs but certainly color tab needs to be pushed harder by your QA with multiple mattes and EXR performance fixed :) I am not criticizing as Resolve is awesome and loving it thoroughly. I know you have heard it before, I will probably continually bring it up every update release as a reminder :) sorry for being a pain...


Ricky, your low CPU core count, PCIe x8 and multiple cards are not optimised well for anything other than basic HD. Which motherboard do you have and what card are in each slot.
DaVinci Resolve Product Manager
Offline
User avatar

Ricky Scharneck

  • Posts: 21
  • Joined: Mon Jul 03, 2017 5:54 pm

Re: 2nd GPU performance improvements

PostMon Nov 13, 2017 8:47 am

Peter Chamberlain wrote:
Ricky Scharneck wrote:
Peter Chamberlain wrote:If you can’t see a benefit of four over two GPUs or even over a single it’s likely you’re just not using enough of the demanding processes or something else is limiting your throughput, usually CPU or disk I/O but can also be as simple as motherboard slot use. Thus the configuration guide is a worthy read if you are building a system for a specific performance.


I am afraid that isn't quite as cut and dry as that at times, I have a 8 core i7 running @ 4Ghz, DDR4 Dominator Pro at 2800Mhz, SSD OS drive 500read 450write (not super but enough), scratch/project M.2 at 3000read 2000write and multiple Titan X's using X8 pci. Yet when i use monitoring software, Resolve doesn't utilize anything enough to find a bottleneck anywhere, not in CPU, data rate, GPU or RAM. Nuke, Premiere, Aftereffects, Modo, etc... all work fine, yet Resolve starts to take ages to update or not at all when it hits that stage. I cannot speak too much about the other tabs but certainly color tab needs to be pushed harder by your QA with multiple mattes and EXR performance fixed :) I am not criticizing as Resolve is awesome and loving it thoroughly. I know you have heard it before, I will probably continually bring it up every update release as a reminder :) sorry for being a pain...


Ricky, your low CPU core count, PCIe x8 and multiple cards are not optimised well for anything other than basic HD. Which motherboard do you have and what card are in each slot.


Hi Peter, I don't get that, low core count? I would understand if my load was at 99% for everything but it doesn't usually go over 20% in Resolve and as for the GFX card, they are plenty optimised and works very well for Octane and Vray GPU and any other application that supports CUDA. Besides, the more GPU's you have in a system the less lanes will be available for the cards and that is normal... I only have 40 lanes to work with. Other software have their own other issues but general performance on my system is not one of them. I am currently using an ASUS Deluxe x99 motherboard. The issue is not on my end from testing other apps... Did you even test the file link you asked me to PM? Test first then judge if it is hardware related please.

@shane, please test https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=63030 If you have stellar performance then I can re-look at things somehow, but tell me why EVERYTHING else works 100% nuke, premiere, Aftereffects, my 3d apps.. but Resolve in Color with multiple mattes? 1 app out of everything... doesn't seem right now does it? :)
Offline

PeterMoretti

  • Posts: 928
  • Joined: Sat Aug 03, 2013 12:12 am

Re: 2nd GPU performance improvements

PostMon Nov 13, 2017 12:25 pm

Ricky, there very well may be a deficiency with how Resolve handles EXR files, I honestly don't know but it sounds plausible, esp. given some other things that I've read on the topic. But that all said, reading the specs of your motherboard, it just doesn't sound like your graphics cards are getting x16 bandwidth.

The below is about your x99 Deluxe from the Asus website:

*1: The PCIe x16_5 shares bandwidth with M.2 x4. Triple PCIe 3.0/2.0 configuration is default set at x8/x8/x8. Adjust PCIEX16_5 Slot Bandwidth in BIOS.

*2: The PCIex4_1, USB3_E12 and SATAEXPRESS_E1 connectors share the same bandwidth. The SATAEXPRESS_E1 will be disabled when there is a device installed on PCIEX4_1 slot. Set this option to X2 Mode or X4 Mode when the installed PCIe device is higher than X4 interface.

Now I don't have that motherboard, so I can't speak to exactly how it works. But I do know that Resolve relies heavily on the GPU. Good luck with this, and I hope you are getting closer to an answer.
Resolve 14.3 Studio. GTX 970 with GeForce 390.77 driver. Desktop Video 10.9.10. Intensity Shuttle USB 3.0. Windows 10 Pro.
Offline
User avatar

Ricky Scharneck

  • Posts: 21
  • Joined: Mon Jul 03, 2017 5:54 pm

Re: 2nd GPU performance improvements

PostMon Nov 13, 2017 1:40 pm

Thanks Peter, yeah unfortunately i cannot get the configuration to 16/16/16 as that is above the 40 lanes the CPU/motherboard can handle. I don't really think it is due to the PCi-e bandwidth speed mind you. I do appreciate you looking into it a little for me. I hope I can get a resolution on EXR too (hardware or software related either way), thanks. If you have a chance, could you have a look at my file too? It would help to know which builds work and which don't. viewtopic.php?f=21&t=63030
Offline

PeterMoretti

  • Posts: 928
  • Joined: Sat Aug 03, 2013 12:12 am

Re: 2nd GPU performance improvements

PostMon Nov 13, 2017 3:01 pm

I will check it out when I get back to LA. I'm in NYC right now through Thanksgiving and think my i3 laptop would bomb with the project.

BTW, can you say exactly what's in which slot, and what are all of the drives on the computer and how each is connected?
Resolve 14.3 Studio. GTX 970 with GeForce 390.77 driver. Desktop Video 10.9.10. Intensity Shuttle USB 3.0. Windows 10 Pro.
Offline
User avatar

Ricky Scharneck

  • Posts: 21
  • Joined: Mon Jul 03, 2017 5:54 pm

Re: 2nd GPU performance improvements

PostMon Nov 13, 2017 8:27 pm

PeterMoretti wrote:I will check it out when I get back to LA. I'm in NYC right now through Thanksgiving and think my i3 laptop would bomb with the project.

BTW, can you say exactly what's in which slot, and what are all of the drives on the computer and how each is connected?


Sure no problem. Nothing spectacular, but gets my work done.

i7 5960X at 4Ghz OC - watercooled
Asus Deluxe x99 with latest bios
Titan X's in PCI 1, 3 and 5 (all 3 set to PCIe x8 and compatible configuration with latest bios)
Samsung 850 Pro in SATA3 port 2 - OS and Apps.
M.2 Samsung 960 Pro in a slot near RAM (at PCIe x4) - Scratch Disk and current Project files.
2x 3Tb Samsung Barracuda SATA3 (Raid 1 - I think there were only 2 ports supporting Raid) - Backup Storage Only. Then from there Uploaded automatically to Cloud Storage.
32Gb DDR4 Corsair Dominator Platinum 2800Mhz - Black Slots, Grey Empty
1700w PSU

A number of USB Peripheral Drives and Audio interfaces, only things connected were the keyboard and mouse.
Previous

Return to DaVinci Resolve

Who is online

Users browsing this forum: dirk-pel, flow85, gebarnett, Google [Bot], greenfun2, Jim Simon, KudzuRaccoon, Lefthandluke, mywald and 186 guests