Op 13-Jan-24 om 11:11 schreef David W. Jones:
On 1/12/24 22:34, Maarten Verberne wrote:
I don't have resources for it yet, either. But if my Dell laptop gives
up the ghost (the Thunderbolt/USB-C port died last year), the
replacement dollars go into the desktop. That currently has a
motherboard running an Intel Pentium 4, so a motherboard replacement/new
memory/new cooling system is inevitable.
in that case even a few gen old i3 will speed up things considerably due
to avx :)
Not in my experience. Stitching starts, 16 threads fire up, and checking
in htop shows none of them waiting for others.
a yes, that part is not single threaded, what i meant was that it starts
with nona>enblend and then back to next image nona>enblend.
if i start one process you'll see so many peaks per minute on the GPU,
where each peak is 2 images that are processed with nona.
if i start 3 (close after each other) you'll see a multitude of peaks,
close to 3x as much per minute.
but after a while they will 'latch up' for quite some time, so the peaks
on the gpu get wider and there are only so many peaks left as with one
script.
every now and then one of the quicker cores matches one of the slower
cores in finishing an image save earlyer and the whole sequence of loose
peaks starts over until they come together again.
I understand that enblend isn't always compiled with multiprocessor
support, maybe that's the difference?
i'm using the precompiled version of hugin in windows, and that seems to
be compiled properly for multriproc, it only lacks support for enblend
gpu out of the box. but i didn't find any improvements speed wise by
using enblend -gpu when i tried that.
Yeah, might be a bit much. But might be more cost effective than
alternatives?
at this moment, that is definitely true for me :)
Yes, Nvidia isn't on my list. If I was using Adobe under Windows on this
machine, I suppose the resident RTX would get used.
I understand AMD's GPUs are more power efficient than the RTX3000 series
GPUs.
yes they are, i used a hd6850 for a while that would be twice the speed
of the uhd630, 6x rtx3060...and that's a 10 year old card :)
however, i think the arc might be the real killer.
Hmm, I'd think that since you're doing this through scripts, you'd have
control over where final images are written. I've never done that, but
might be worth asking about.
i think it is too little gain to pursue that for now.
I survive on a mere 2TB M2 drive, but don't do as much heavy lifting as
you.
for last year images i have 12TB in store, sorting and then stitching
triples that for the duration of the project.
Hmm, I've had Hugin (particularly enblend) consume more than the 64GB
RAM in my laptop when stitching. Probably depends on the sizes of the
source images and the final image. Perhaps the image format, too?
absolutely.
I don't think GPU matters at all, as you pointed out about Intel onboard
GPUs outrunning the fancy GPUs. If the GPU supports OpenGL (without
throwing you out of house and home with its electric bill!), then any
basic GPU is good. :)
It matters in the speed nona works, and that's still significant if you
do a lot of stitching.
i might be able to catch some screen prints if you like.
Ah, as Lukas Wirz wrote, it appears openCL...got those mixed up, but the
point stands. a gpu that does openCL well is what you want, that ain't
nvidea, this is intel and amd.
About speedy cards, is it the FP64 performance that makes specific cards
shine more?
--
A list of frequently asked questions is available at:
http://wiki.panotools.org/Hugin_FAQ
---
You received this message because you are subscribed to the Google Groups "hugin and other free panoramic software" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to hugin-ptx+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/hugin-ptx/253bf783-a2f5-4925-8b2c-e03e50a77205%40gmail.com.
