GPU comparison for gigapixel stitching

[Conrad Allan]

Hi Joost,

With your depth of experience, do you have any thoughts on which of these two GPU specs would be more efficient in stitching 50,000px and larger HDRI's?

Nvidia RTX 3060 12GB
AMD RX 6800 16GB

As I understand it, we want lots of VRAM so the stitch can be entirely handled in the GPU. I'm not sure how much VRAM ptgui might want with a 50,000px HDRI though?

The Nvidia card has the benefit of CUDA cores, does that have any effect on stitch speed? If 12GB of VRAM is plenty for the stitch, then the Nvidia is the better card because of the nvenc encoder and the cuda cores.

Thanks for the advice,

[PTGui Support]

Hi Conrad,

To be honest I don't think there will be much difference. Both are
plenty fast and the other parts of the system will become the bottleneck.

If you use optimum seam finding: this runs on the CPU and will become
the bottleneck. The old multiband blender can run entirely on the GPU
but the new Zero Overlap blender will (partially) run on the CPU. As
PTGui will need to move data in and out of the GPU anyway, the
difference between 12 and 16 G VRAM will be hardly noticeable.

I've seen occasional problems with nvidia GPU drivers if you run PTGui
and other applications (like Lightroom) at the same time. Especially
when the Game Ready driver is used, but also with their Studio driver.
That could be a reason to go for an AMD card. If PTGui will run alone
this is not an issue.

You do want to have lots of (regular) RAM. Especially if you use
bracketed source images or tone mapping, 128GB of RAM would speed up
stitching a lot.

CUDA cores is just nvidias naming for the number of processors on the
GPU. PTGui uses OpenCL, not CUDA, but both use the same processors on
the card. NVENC is not used by PTGui.

Kind regards,

Joost Nieuwenhuijse
www.ptgui.com

> ​Matte Paint Pty LTD accepts no liability for the content of this email,
> or for the consequences of any actions taken on the basis of the
> information provided, unless that information is subsequently confirmed
> in writing.
> Any views or opinions presented in this email are solely those of the
> author and do not necessarily represent those of the company.
>
> --
> You received this message because you are subscribed to the Google
> Groups "PTGui Support" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to ptgui+un...@googlegroups.com
> <mailto:ptgui+un...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/ptgui/8a09b87c-9ee4-436c-afb1-237658783916n%40googlegroups.com
> <https://groups.google.com/d/msgid/ptgui/8a09b87c-9ee4-436c-afb1-237658783916n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Conrad Allan]

Thanks again Joost!

It is as I expected. Though it's good to know the optimum seam finding runs on the CPU. 

We currently have an RX570 (4gb RAM) and is really the only portion of the pipeline which I believe could speed things up. 

Our CPU is a 5950x (12 core), HDD's are 2x PCIE 4.0 nvme drives and the server has 128GB of ram.

Thanks again,