Resolve 17 issue on Lenovo P620 and Nvidia Quatro P620

I am running into an issue on Resolve 17. On new machines I have an old project that I use as a rendering baseline to see how the machine will perform compared to other machines. The video project is several 1080p/60fps videos with some static jpg/png shows and some text overlays. There is also a music track in addition to the main audio track.

Very simple process-
1. Install Resolve
2. Copy project files off a USB drive
3. Import project file
4. Go to "Deliver" tab, create a simple MP4 H.264 1080p 59.94 fps output, medium quality
5. Render and record the total time

I tried this last week on a new Lenovo P620 (AMD Threadripper 16 cores, 64GB DDR4, multiple SSDs, Nvidia Quadro P620 GPU). When the render process begins it quickly fails. The CPU and GPU never spool up above 10%.

This is the free version of Resolve 17.2 build 11. I started on 17.1.1 but updated to 17.2 to see if it solved the issue.

Machine details:
Lenovo Thinkstation P620 Gen 1 (model released in 2020, purchased last month)
Windows 10 Version 20H2 build 19042.985 (and fully patched, according to Windows Update)

I updated the Graphics driver to 27.21.14.6259 but that didn't change anything. Resolve support logs from before and after this driver update are attached.

On this machine I only have the 1 project defined. Sometimes when I open the project I see this error-

GPU Memory Full.jpg (30.4 KiB) Viewed 2022 times

When I attempt to render the project in MP4, 1920x1080, 59.94 fps, medium quality, the job starts and fails after 30-60 seconds with this error-

Render job failed.jpg (34.64 KiB) Viewed 2022 times

This job is a project I put together last year and have used it on several machines as a baseline to gauge render times, including Lenovo machines with this same OS and GPU. This is the first machine to give me any issues rendering this project. I do plan to eventually change the GPU to an RTX4000 but I wanted to see how the machine would perform with the Quadro P620 that came with it.

In looking at the logs it might be a memory allocation error on the GPU?

Any ideas on where to go next?

2 gigs of VRAM is barely enough for what you're doing right now. I would recommend a GPU with more memory.

I agree that this GPU is not the optimal one for this scenario but I have rendered this project on this same GPU model in several machines before, and on dozens of machines with only the on-die GPU (i.e. Intel UHD graphics, Iris Xe graphics, AMD Radeon Graphics, etc). There is something going on here with the GPU and the logs points to memory handling-

[0x00002e1c] | GPU.SingleBoardMgr | INFO | 2021-05-12 16:45:57,827 | Flushing GPU memory...
[0x00002e1c] | UI.GLTexPool | INFO | 2021-05-12 16:45:57,842 | Released 0 MiB in 0 unused textures.
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,879 | cudaMalloc failed: Requested size 7.9 MiB
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,879 | DeviceResourceTypeCUDA.cpp:97: CUDA error cudaErrorMemoryAllocation (2)
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,879 | Cuda Memory Status : free 4.1 MiB total 2048.0 MiB
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,884 | cudaMalloc failed: Requested size 7.9 MiB
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,884 | DeviceResourceTypeCUDA.cpp:97: CUDA error cudaErrorMemoryAllocation (2)
[0x00002e1c] | DVIP | ERROR | 2021-05-12 16:45:57,884 | Cuda Memory Status : free 4.1 MiB total 2048.0 MiB
[0x00002e1c] | GPU.SingleBoardMgr | ERROR | 2021-05-12 16:45:57,889 | DVIP exception caught:
DVIP Exception: Position DeviceResourceTypeCUDA.cpp:97
- API: CUDA
- API Error Code: cudaErrorMemoryAllocation (2)
- Call stack:
00007FF68B8B168D
00007FF68B9A9F71
00007FF68B9A8916
00007FF68B9A9096
00007FF68B9A96F3
00007FF68B9AB2E4
00007FF68DE96B00
00007FF9D6C61080
00007FF9D6C626A5
00007FFA0A551406
00007FF68B9AC87A
00007FF68B9635E1
00007FF68B96216B
00007FF68B8D2304
00007FF68D6146C8
00007FF68D6140F8
00007FF68D5E396A
00007FF68D6AA2D8
00007FF68D5FB121
00007FF68D5D0B38
00007FF68D367506
00007FF68D364A56
00007FF68D365F58
00007FF68D361A69
00007FFA07C91BB2
00007FFA0A3C7034

[Dwaine Maggart]

No. Your Quadro P620 GPU only has 2GB of VRAM. That's not enough to do much.

You can see in the log segment you posted that it only has 4.6 MB of VRAM available, and something is asking for 7.9 MB.

If you want to post a project export .drp, we can likely point out the item(s) causing the issue.

DRP is too big to attach (2MB file limit?).
Here is a link to it on my OneDrive- https://1drv.ms/u/s!AoXbpELydADNtsJ2YR_ ... Q?e=nJzlZZ

I have run this same render job on 6 different systems with this same graphics card but only 2 of those had Resolve 17.x, the rest were 16.x.

Like I said, I plan to replace the GPU with an RTX 4000 in the very near future but I wanted to get a baseline with the Quadro P620 first.

Just for grins I uninstalled 17.2 and installed 16.2.8.

Video rendered successfully 29:51, one of the fastest times I have seen on this video. I then installed 17.2 (update in place) but the problem persists.

[Dwaine Maggart]

Comparing your project between free Resolve 16.2.8 and 17.2, and adding a similar sized JPEG to the end of the timeline, yes, 17.2 is using more VRAM than 16.2.8. 16.2.8 is using 1.8GB to play the clip. 17.2 is using 2.8GB to play the clip. (I'm using a GTX 1050Ti, with 4GB of VRAM, and monitoring VRAM usage with the Windows Task Manager Performance tab).

Based on what I see, you should be getting GPU errors just playing the timeline, let alone rendering, when using 17.2.

So yes, 17.2 is using more VRAM. My assumption would be there is a good reason for that, but I don't write the code, and I can't say why that VRAM use difference exists.