Improving Resolve Performance on Windows

I have recently put together a new computer. This works a lot better than my previous one, but I notice that Resolve is not using both GPUs.

Any idea why, or how to fix it?

Rendering Performance:

Resolve-RenderingPerformance.jpg (65.96 KiB) Viewed 12955 times

Playback Performance:

Resolve-PlaybackPerformance.jpg (63.71 KiB) Viewed 12955 times

Hi.

I expect it is because you use the FREE version of resolve?
To use more than one GPU you must use the paid STUDIO version of Resolve.

Regards Carsten.

The "Link 0" after the GPU number suggests you are using SLI which DaVinci Resolve does not support. Disable SLI and only connect monitor cables to one of the GPU:s.

I expect it is because you use the FREE version of resolve?
To use more than one GPU you must use the paid STUDIO version of Resolve.

Thanks for the reply Carsten. I'm using the Studio version. (see my signature).

+1
As Roger said: no SLI and connect screens only to GPU Slot # 1 (as PCI-E x16 depand motherboard)

The "Link 0" after the GPU number suggests you are using SLI which DaVinci Resolve does not support. Disable SLI and only connect monitor cables to one of the GPU:s.

Thanks Roger. It was indeed. I hadn't even thought to check SLI, assuming I would need to explicitly enable it.

However, although both GPUs are now being used I don't see any improvement in render time. It looks like the each GPU is running at a low rate. Nothing else seems to me maxed out either, so I don't know where the bottleneck is.

Resolve-RenderingPerformance2.jpg (67.79 KiB) Viewed 12777 times

As Roger said: no SLI and connect screens only to GPU Slot # 1 (as PCI-E x16 depand motherboard)

Thanks for the reply Jean. I already have both monitors connected to the first card.

As Roger said: no SLI and connect screens only to GPU Slot # 1 (as PCI-E x16 depand motherboard)

Thanks for the reply Jean. I already have both monitors connected to the first card.

Hi,

Do not use SLI is very important.

Thanks for the reply Carsten. I'm using the Studio version. (see my signature).

Hi.

I am sorry I didn't see your profile. I think I was to busy scrolling down in you screen dumps.
But I read somewhere in this forum that Noise Reduction is one of the functions that really benefit
from dual GPU's. Can it be an idea to try it?

Regards Carsten.

Do not use SLI is very important.

Yes, see my latest performance graph above.
I have disabled SLI and now both GPUs are being used, but at a much lower capacity than the single GPU was being used before, so the net result is no performance increase.

Since nothing in those graphs is maxed out I don't know where the bottleneck would be.

But I read somewhere in this forum that Noise Reduction is one of the functions that really benefit
from dual GPU's. Can it be an idea to try it?

The render I'm testing this with does have video NR applied, plus a little other grading and some audio processing, mainly audio NR using using iZotope RX 6 Voice De-Noise

Be warned that rendering audio with RX6 fx applied may lead to sync loss and other anomalies.

Have you tried disabling one of the 1070s altogether? You might want to, and see what happens to rendering performance.

Since nothing in those graphs is maxed out I don't know where the bottleneck would be.

Hi.

If you want to search for the bottleneck will I suggest you to make a RAM Disc and use it for testing?
Then will you be able to see if your disc's is causing the bottleneck.

Or if you tell us more about your system, can some member of this forum may be tell you where to look.
What motherboard is you using?
What two slot are you using for the GPU's
How many and what type of your hard drive and SSD?
etc.

Regards Carsten.

Be warned that rendering audio with RX6 fx applied may lead to sync loss and other anomalies.

I've just done about 7 hours of finished content using RX6 and the audio/video sync is perfect, so I guess it doesn't always have problems.

Have you tried disabling one of the 1070s altogether? You might want to, and see what happens to rendering performance.

I just did. The performance seems to be very close to the same. I'm seeing an average of 16 frames per second processed, which was what it has been doing on all these tests.

Resolve-RenderingPerformance3.jpg (46.78 KiB) Viewed 12628 times

If you want to search for the bottleneck will I suggest you to make a RAM Disc and use it for testing?
Then will you be able to see if your disc's is causing the bottleneck.

I'll have a go at that later. I'm not sure yet how to create RAM disks on Win10. Haven't done that for many years.

Although I don't really expect the disk will be the bottleneck. My video files aren't that big; 150Mb/s UHD MP4s. I can see the decoding being more of an issue, but neither the CPU or the GPUs are working hard at all.

What's puzzling me the most is how it looks like the same amount of GPU processing is used regardless of if Resolve is using one of them or both. The job is just shared and each works half as hard.

Or if you tell us more about your system, can some member of this forum may be tell you where to look.
What motherboard is you using?
What two slot are you using for the GPU's
How many and what type of your hard drive and SSD?
etc.

Motherboard: ASUS STRIX Z270F
https://www.asus.com/au/Motherboards/RO ... 0F-GAMING/
CPU: i7-7700K @ 4.2GHz
Memory: 32GB 2400
OS SSD: 256GB M.2 NVMe
Second SSD: 256GB M.2 NVMe
Hard drives: 3TB 7200RPM x 3
Source Media drive for project: 2TB 7200RPM

What are your source files precisely?
Do you have any other 3rd part effects?

What if you disable NR and RX6?

I've just done about 7 hours of finished content using RX6 and the audio/video sync is perfect, so I guess it doesn't always have problems.

There's no latency communication between these plugins and Resolve, but if real time playback from the timeline is good, the rendered file may be acceptable. Just note that any observed anomalies during playback will get baked into the render. You may survive one or two rx6 filters, depending on which ones and how much processing they do.

BTW, 15 comes with a good noise reduction filter, and no latency issues encountered, so far.

Three things for multi GPU on Windows.

1) Studio Version - Check
2) Disable SLI in Nvidia drivers - Check

Did you set this check box?

3) Preferences>System Config>Use display GPU for compute

If that doesn't help, it looks like a hardware config issue causing a bandwidth bottleneck somewhere. Make sure you have components in the optimum slots, and you aren't choking the PCIe bus somehow with the wrong NVME/GPU/SATA arrangement or a poor memory settings slowing you down.

it looks like a hardware config issue causing a bandwidth bottleneck somewhere. Make sure you have components in the optimum slots, and you aren't choking the PCIe bus somehow with the wrong NVME/GPU/SATA arrangement

Hi.

I agree. The reason I suggested to make a RamDisc is that a test will show if the hardware bottleneck is in the connection to the GPU's or if it is in the connections to the Harddrives. If you test with a RamDisc and still have a slow system will I expect the problem to be with the connection to the GPU's. If you suddenly get a faster system will I expect your problem to be related to the Disc System. Both the GPU's and the Disc's are connected by PCIe lanes.

Lets look at it. In your link did I found the hardware manuel:

http://dlcdnet.asus.com/pub/ASUS/mb/LGA ... V2_WEB.pdf

Here I see that your motherboard have 7 slot:

2 x PCIe X16 (X16, X8/X8)
1 x PCIe X16 (max X4 mode)
4 x PCIe X1 Slot

If you only have one Gpu must it go in slot no 2.
If you have two GPU's must they go in slot no 2 and no 5.

If you only use one GPU will it use 16x PCIe lanes. If you use two GPU's will each use 8x PCIe lanes.

It is not the most optimal way to do it. But I think that is OK. But a test with a RamDisc will show.
What do other people reading this thread think?


To look into your disc system will I need some more information.

Please post the exact type numbers of your two NVMe SSD's and
tell us what 4 SATA ports your harddrives are connected to?

And I wonder, if you use RAID?


Regards Carsten.

Having closer look at usage of PCIe lanes sounds the like a good point to me.

If source is 150mbit h264 then ramdisk test is not needed. It's very unlikely to have anything to do with disks reading speed.

If source is 150mbit h264 then ramdisk test is not needed. It's very unlikely to have anything to do with disks reading speed.

Hi.

No I don't believe that either. What I am looking for is if the PC is actually shareing Disc Controller PORTs or may be PCIe lanes.

The manual says:

M2_2 is closest to CPU.
M2_1 is clostet to the bottom.

Under Storage I read:

M2_1 share SATA 1 port
M2_2 share SATA_56 ....

So what I ask for is exact type numbers of his two NVMe SSD's and
please tell us what 4 SATA ports your harddrives are connected to?

Regards Carsten.

Please post the exact type numbers of your two NVMe SSD's and
tell us what 4 SATA ports your harddrives are connected to?

The OS drive is a Samsung MZVPV256
The other is a Samsung 960EVO 250GB

And I wonder, if you use RAID?

No.

If source is 150mbit h264 then ramdisk test is not needed. It's very unlikely to have anything to do with disks reading speed.

That was my reasoning.

please tell us what 4 SATA ports your harddrives are connected to?

I don't have time to pull the computer apart at the moment. It in a difficult to get at position.

I just set up a basic test.
1 Clip: H.264 70Mb/s UHD 25FPS with stereo audio. 1 minute long. Rendered to DNxHR HQX 10bit UHD.
Both source and destination are the EVO SSD.

No processing:
- Playback speed: 25FPS, Render time: 1:12

Video NR and a bit of colour grading:
- Playback speed: 10FPS, Render time: 2:58

Hi.

Thanks for the type numbers.

All my own 3 SSD are SATA only. But as I understand the M2 slot can you use either SATA or faster PCIe x4 connection.

I don't know if you set it up in the BIOS?

I don't expect the slower SATA to be a problem itself. But I don't know what will happen if SATA is activated and at the same time it is sharing the SATA Controller PORT with a SATA harddrive. So please tell us which 4 SATA ports your harddrives is connected to.

If you don't have the possibilities to make a RamDisc test, can you may be try to use the OS SSD for a test. Just to change the drive to make sure it isn't it or the way it is connected.


I also like to give you all a link to a customer review on Amazon.com where SATA and PCIe x4 version got mixed.

'Ordered 2 and received 2 different - one with Slower SATA B+M Interface ???'
By GG-Prime Member

https://www.amazon.com/Samsung-960-EVO- ... filter-bar

Please see his picture for differences.

Regards Carsten.

If source is 150mbit h264 then ramdisk test is not needed. It's very unlikely to have anything to do with disks reading speed.

That was my reasoning.

please tell us what 4 SATA ports your harddrives are connected to?

I don't have time to pull the computer apart at the moment. It in a difficult to get at position.

I just set up a basic test.
1 Clip: H.264 70Mb/s UHD 25FPS with stereo audio. 1 minute long. Rendered to DNxHR HQX 10bit UHD.
Both source and destination are the EVO SSD.

No processing:
- Playback speed: 25FPS, Render time: 1:12

Video NR and a bit of colour grading:
- Playback speed: 10FPS, Render time: 2:58

Those numbers are quite poor.
You can't even in real time render to DNxHR. I can do this on MacBook Pro with much weaker spec.
What is your source- is it 8bit 4:2:0 h264 (or eg. GH5 4:2:2 h264)?
Your problems have to be GPU related- drivers etc (maybe your system runs on bad PCI-E resource share, eg. GPUs are running on x4)
Have you done on your SSD speed test to make sure it's working fine?
In the same time 50% CPU usage on multithreaded system is actually not bad.

i7-7700K Processor specs

Cores: 4

PCI Express Configurations: Up to 1x16, 2x8, 1x8+2x4

Max of PCI Express Lanes 16

1. Gtx 1070 PCI Express Lanes 16
Your Intensity Pro 4K need 4 Lanes !

I got 2. 16x graphic cards and a Intensity Pro 4K card !
ASUS X99-Deluxe II motherboard - Cpu i7 5960x, 8 cores, 16 Threads, 40 PCI Express Lanes !
PSU: EVGA SuperNOVA 1300 G2, 1300W

Hi.

Nice finding Ole.


Here is the specifikation for the Z270 chipset:

PCI Express Configurations x1, x2, x4
Max # of PCI Express Lanes 24

https://ark.intel.com/products/98089/Intel-Z270-Chipset

But I have no idea how the Asus motherboards uses the Z270's 24 PCI Express Lanes.

Regards Carsten.

Hej Carsten

Intel Z270 Chipset

Supported Processor PCI Express Port Configurations: 1x16 or 2x8 or 1x8+2x4

7700K only have the 16 from the CPU itself. The chipset controls the extra Z270 = 24total.

[Peter Chamberlain]

Looks like a hardware issue, not Resolve at all.
Please review the config guide on the support web site for recommendations on hardware.

i7-7700K Processor specs

Cores: 4

PCI Express Configurations: Up to 1x16, 2x8, 1x8+2x4

Max of PCI Express Lanes 16

1. Gtx 1070 PCI Express Lanes 16
Your Intensity Pro 4K need 4 Lanes !

That looks like a good explanation. Thank you. Looks like a rebuild is in order. I'm not sure there is a CPU with more lanes that will fit that motherboard, so it looks like I need to change both.

The updated Threadripper is due soon. Can't be beat for cores and lanes at the moment. All your current parts will drop right in, just a motherboard and cpu needed.

The updated Threadripper is due soon. Can't be beat for cores and lanes at the moment. All your current parts will drop right in, just a motherboard and cpu needed.

I'm currently looking at a ASUS Prime X299 deluxe and i9 7820X. That's what BMD recommend in their Resolve config guide.

I expect Threadripper CPUs and motherboards will be too expensive for me for a while.

Hej Carsten

Intel Z270 Chipset

Supported Processor PCI Express Port Configurations: 1x16 or 2x8 or 1x8+2x4

7700K only have the 16 from the CPU itself. The chipset controls the extra Z270 = 24total.

Hi.

No I am sorry, the correct numbers of PCIe lanes coming out from the Z270 is 30 according to this link:

https://www.tomshardware.com/news/intel ... 33305.html

But the 6 is preallocated to USB3 and 8 is preallocated to PCIe SATA. Giving 16 PCIe lanes left.

In your case goes 2x4=8 PCIe lanes to the M.2 PCIe x4 SSD's and 4 PCIe lanes to the 4x slot.

I expect 1 PCIe lane for LAN and 1 PCIe lane for Sound.

So left is 2 PCIe lanes for the 4 pcs 1x slots. That must be enough, specially if you leave 2 empty.

That together with the 16 PCIe lanes coming out from the CPU which is used as x16 for one GPU or 2 times x8 for two GPU Finished this calculation. I can't see any SHARING of PCI'e lanes.


But before going out and buy a new Desktop, we all want new Desktops. Will I suggest two things.

Check the location of your GPU's?
Are they in the right positions, which are:

If you only have one Gpu must it go in slot no 2.
If you have two GPU's must they go in slot no 2 and no 5.


And if you are not sure about the harddrive sharing SATA ports with the M.2 SSD, then decrease your number of harddrives to max 3. If you only use SATA Port 2, 3, and 4 you don't have think about any SATA sharing. The spare Harddrive can you mount in a empty USB3 enclosure.

But I understand. We all like new desktops. I will personally look at the next generation AMD Threadripper expected some time in august. It is recommend in the DaVinci Resolve 15 Configuration Guide on page 18:

http://documents.blackmagicdesign.com/D ... _Guide.pdf

The Intel 10 core Core i9-7900X has 44 PCIe Lanes and the CPU cost $999

The AMD Ryzen Threadripper 1950X 16 Cores / 32 Threads is reduced to $869
And AMD Ryzen Threadripper 1920X 12 Cores / 24 Threads is reduced to $669
All AMD Threadrippers have 60 PCIe lanes.

Regards Carsten.

Can you plz post a screenshot of compute tab of your GPUs in task manager.

When I built my Windows computer with 2-piece Sapphire NITRO R9 390X 8GB GDDR5 graphics cards and an ASUS X99-Deluxe II motherboard - I got a problem when placing the two graphics cards in the right ports - they cover the other ones ports - so you can not mount an Intensity Pro 4K card - look at the picture that I had to place the Intensity Pro 4K card somewhere else and use a Pci extension cable!

When using more than one graphics card - remember that one's PSU should be able to provide enough power!

I have recently put together a new computer. This works a lot better than my previous one, but I notice that Resolve is not using both GPUs.

Any idea why, or how to fix it?

Did you try MSI Afterburner for monitoring? For me the results are very different from Windows's Task Manager

SLI.jpg (542.53 KiB) Viewed 12195 times

The updated Threadripper is due soon. Can't be beat for cores and lanes at the moment. All your current parts will drop right in, just a motherboard and cpu needed.

I'm currently looking at a ASUS Prime X299 deluxe and i9 7820X. That's what BMD recommend in their Resolve config guide.

I expect Threadripper CPUs and motherboards will be too expensive for me for a while.

They are much cheaper than Intel equivalents in spec. If you match price, they are faster and have greater potential. A 7820X is virtually what you have now, you're simply moving sideways.

But I understand. We all like new desktops. I will personally look at the next generation AMD Threadripper expected some time in august. It is recommend in the DaVinci Resolve 15 Configuration Guide on page 18:

http://documents.blackmagicdesign.com/D ... _Guide.pdf

The Intel 10 core Core i9-7900X has 44 PCIe Lanes and the CPU cost $999

The AMD Ryzen Threadripper 1950X 16 Cores / 32 Threads is reduced to $869
And AMD Ryzen Threadripper 1920X 12 Cores / 24 Threads is reduced to $669
All AMD Threadrippers have 60 PCIe lanes.

As Carsten notes, Threadripper is in the config guide. I have a 1950X and Gigabyte X399 Aorus 7 with 2 x 1080TI running full lanes and three NVME with full lanes and a 4K Mini running full lanes, along with a half dozen SATA drives, all full lanes.

I'm currently looking at a ASUS Prime X299 deluxe and i9 7820X. That's what BMD recommend in their Resolve config guide.

I looked more closely at the Threadripper system suggested by BMD and it does seem to be better suited.

As far as I can tell, it will enable running both my GPUs at 16x.

I also prefer the M.2 implementation from a thermal perspective.

As far as I can tell, it will enable running both my GPUs at 16x.

I also prefer the M.2 implementation from a thermal perspective.

Hi.

I understand that the AMD Threadripper is new to you. Here is some short information I hope will help you when start looking into it.

Yes, it will enable running both Your GPUs at 16x and if you prefer M.2 will I recommend you to look at this link for:

'Free NVMe RAID upgrade for AMD X399 chipset'

https://community.amd.com/community/gam ... 99-chipset

If you don't have enough M.2 sockets, have several motherboards manufacturers just released PCIe x16 Riser Cards with up to four extra M.2 sockets.


The first generation Threadrippers was introduced around 1 year ago together with the first generation of X399 motherboards.

At the Computex tradeshow, AMD both talked about and showed 2nd generation 24 core and 48 threads and 32 core and 64 threads Threadripper systems. It is official a Q3 release. But I personally expect it some time in August.

First Gen and 2nd Gen Threadrippers uses the SAME motherboards. But at the same time as the 2nd Gen Threadrippers, we expect what they call new Refresh X399 motherboards.

I will personally look into one of the new Refresh X399 motherboards, as the will be one year newer in design. But others may be looking for, good deals on one of the original motherboards.

The only known difference between the 1st Gen and the coming 2nd Gen Threadrippers, is that the coming 24 core and 32 core will have a TDP of 250 Watt compared to the 180 Watt of the 1st Gen.

So if you buy a 1st Gen motherboard and like to have the possibilities to upgrade to 24 or 32 core CPU's in future, then look into a motherboard with a heavy internal power distribution.


Many current Threadripper users are using Water cooling. But at Computex both the 24 and the 32 cores Threarippers was AIR cooled. I don't mind if you end up with a 1st or 2nd gen motherboard,
but I will recommend you to always buy the new AIR cooler that is disused in this 16 minutes long YouTube::

'AMD's Jim Anderson discusses 32-core Threadripper 2'

Regards Carsten.

I understand that the AMD Threadripper is new to you. Here is some short information I hope will help you when start looking into it.

Thanks Carsten.

I've decided to go for exactly what BMD specify in the config guide for a Threadripper system.

Parts have been ordered.

Thank you all for your advice.

Okay, the parts arrived yesterday and I have put the system together.

New parts:
Gigabyte X399 AORUS Gaming 7 motherboard
AMD 1950X CPU
Corsair 1000W PSU

Parts I already had:
2 x 16GB 2400 RAM in optimal slots (DDR4_2_2A & DDR4_2_2B)
2 x GTX1070 graphics cards in optimal slots (PCIEX16_1 & PCIEX16_2)
BMD Intensity Pro 4k (in PCIEX4)
Samsung EVO 960 250GB M.2 NVMe SSD in M2P_32G (C: Windows 10 Home)
Samsung MZVPV256 M.2 NVMe SSD in M2Q_32G (D: used for source and output for my test project)
3 x 3TB 7200RPM hard drives (not used in my test project)
2TB 7200 RPM hard drive (not used in my test project)

Test project: 1 clip 2160p25 H.264 70Mb/s 8 bit 4:2:0, 1 minute long with stereo audio.

Render output format: DNxHR HQX 10 bit 1920x1080 25p

Test render 1: Temporal and spacial NR plus a minor colour grade.
Render time: 2:21 (about 11 FPS)

Test render 2: No processing
Render time: 0:11 (136 FPS)

I guess this will do me, but honestly, I was expecting a bit better. On the render with a bit of processing nothing is even close to maxed out. Curiously, the simple render was very fast and did make good use of the hardware.

Maybe I still don't have something configured optimally.

It's clearly related to GPU processing power. It's GPU which is holding you in case of NR is used.
Try different GPI driver, single GPU as test (in Resolve preferences).

I'm not sure what you should expect from GTX 1070 when it comes to NR, but I guess 2 of them should do at leats realtime for 25p.

It's clearly related to GPU processing power. It's GPU which is holding you in case of NR is used.
Try different GPI driver, single GPU as test (in Resolve preferences).

I'm not sure what you should expect from GTX 1070 when it comes to NR, but I guess 2 of them should do at leats realtime for 25p.

I agree.

Even with my 6850k and 1070, the gpu is the limiting factor when it comes to NR.

It's clearly related to GPU processing power. It's GPU which is holding you in case of NR is used.

Neither of the GPUs are being pushed hard. One averages 55% and the other 13%. And yet, with no processing both GPUs are much better utilised.


Since this should be a fairly common hardware configuration, considering it's one of only 2 Window configs specified by BMD, I'm hoping someone with similar hardware can comment if the performance of my machine is in the ballpark.

Try different GPI driver,

What's a GPI driver? Or is that a typo? I'm using the most recent nVidia driver. Are you suggesting I try some older ones?

single GPU as test (in Resolve preferences).

Do you mean unchecking "Use Display GPU For Compute"? I tried that. GPU0s average usage increased to 60% and the render took 3:08 [8 FPS]

I'm not sure what you should expect from GTX 1070 when it comes to NR, but I guess 2 of them should do at leats realtime for 25p.

Indeed.

Even with my 6850k and 1070, the gpu is the limiting factor when it comes to NR.

Are you seeing significantly under 100% GPU usage during a render?

Even with my 6850k and 1070, the gpu is the limiting factor when it comes to NR.

Are you seeing significantly under 100% GPU usage during a render?

Nope, it's into the low 90's! The thing is the core load isn't the only thing on your gpu that can bottleneck you.

Windows task manager really isn't good enough to help you find bottlenecks, you need to use something more advanced, like hwinfo.


On my machine I run rainmeter hwinfo, and a rainmeter plugin for hwinfo so that I can monitor all the important metrics. Rain meter isn't needed, its just a nice tool that lets you make desktop widgets.



here is a test I just ran:

10bit 4:2:2 150Mbit UHD h.264 gh5 footage rendered to DNxHR HQX 10 bit UHD.


These are the settings. i used.

settings.jpg (87.28 KiB) Viewed 11756 times

While the render was running fps bounced between 10.5 to 11, and the gpu core never got above 92%.

However take a look at some of the more important metrics of my overclocked gtx 1070. I took this about 45 seconds into the render.

display.jpg (63.67 KiB) Viewed 11756 times

Specifically look at the voltage and TDP. Voltage bounced between 1.081 & 1.094. 1.094 is the max voltage you can dump through the gpu without flashing it with a custom bios. as you can see TDP was bouncing of the ceiling as well so it was sucking down a lot of amps. In short the GPU is red lined.

What this is, is a memory access bottleneck, in other words the VRAM isn't fast enough. NR is memory access intensive, because you are trying to read from it and write to it very fast. If the memory was faster the core would most likely hit 100%.


take a look at the 10 series gpu memory speeds and bandwidth.

Code: Select all

model      speed(Gbps)   bandwidth(GB/sec)
1050               7.0               112.0
1050 ti            7.0               112.0
1060               8.0               192.0
1070               8.0               256.0
1070 ti            8.0               256.0
1080              10.0               320.0
1080 ti           11.0               484.0
Titan Xp          11.4               547.7


The other thing I think you will find, is the NR is not something that can be parallelized (split up) across multiple GPUs, hence the reason 1 of yours is at 99% and the other at 0%. This along with a few other reasons is why BM recommends a single higher end gpu compared to two lower ones.

The other thing I think you will find, is the NR is not something that can be parallelized (split up) across multiple GPUs, hence the reason 1 of yours is at 99% and the other at 0%.

During the render with NR, the maximum I ever saw the main GPU get was 62%. The second GPU never got above about 13%.

This along with a few other reasons is why BM recommends a single higher end gpu compared to two lower ones.

A support person I talked to from BMD's Australian agent didn't say that at all. He said Resolve loves multiple GPUs and lots of graphics memory. "Adding more GPUs makes an exponential difference." That's the reason I went for two 1070s instead of one 1080.

Thanks for your detailed reply. I downloaded HWiNFO. Attached is what it showed during a render
I notice that GPU Computing Usage is at or close to 100% on both GPUs, so I guess that's all I can expect.

I'm currently testing the eGPU with an iMac (non Pro) and I get nearly half the rendering times for NR with both GPUs. But maybe it's a Mac thing.

Hi.

I remember reading the Conclusion in this link:

https://www.pugetsystems.com/labs/artic ... n-Xp-1060/

'That said, with temporal noise reduction the performance gain was very respectable - we saw an average 25-35% performance gain with two GPUs and a further ~10% with three GPUs.'

So I am a little surprised.


And I see that they later published a newer article looking at the the performance of 1-4 GTX 1080 Ti and 1-4 Titan V GPUs:

https://www.pugetsystems.com/labs/artic ... n-SP-1121/

Under Conclusion:
3 More GPUs is NOT always faster:
2nd paragraph:
'Adding TNR was really where we started to see the benefit of multiple GPUs.'

Can somebody please explain the difference in the test setups and why we see the different results?

Regards Carsten.