Helicon Soft Forum

Just built a new system with windows 10 Pro,an Intel i9-12900, EVERYTHING on SSD, most on NVMe, 64GB of 3600 DDR4, AMD 5700 XT. This is a MASSIVE upgrade from the 7-year-old I7-5820 I WAS using.

Puget Systems Photoshop benchmark is massively higher.
Lightroom tests like building 100 1:1 previews is over twice as fast.
And so on.

In Helicon Focus, I ran the benchmark and got:
Performance 2.48 seconds (846.43 MPixel/s)
Transfer speed: 14712.34 MiB/s
Stack: 100 in-memory images, method B.

In the results comparison, it's now 786 as opposed to somewhere around 367 previously.

I've been consistently testing with a 14-shot, D850 focus stack, and timed in Method B from the press of Render button to ready for Saving. Times with the 5820 were consistently 35 seconds +/- 1 second across a half dozen tests.

With the NEW system, times were consistently 65 seconds, again +/- about a second, so it was taking approximately twice as long.

Performance settings are Memory cache Limit: 10%, Parallel image loading: Auto, Use OpenCL hardware acceleration: AMD 5700XT.

I tried turning off OpenCL hardware acceleration - time increased to about 69 seconds - essentially no difference.
Tried running with OpenCL hardware acceleration on and FailSafe mode - 66 seconds - essentially no difference.
I changed the Parallel image loading from Auto to Enable - which made ABSOLUTELY NO difference on the old system.

Processing time dropped from 65 seconds to less than 12. Most times were around 11.5 seconds.

I'm not sure WHY this odd change would make such a huge difference, or if it will prove consistent over time. BUT, if you're getting sort-of adequate performance try changing the Parallel image loading value from Auto to Enable. Can't hurt, only takes a second, MIGHT help...

Changing rendering method and re-rendering was fairly fast previously, but with these settings, changing to Method C and rerunning provides times just over 3 seconds.

I'll grab some other stacks between 10 and 20 or so images and see what it does, but SO FAR, this minor change seems to help.

Are there any other preferences I should look at? Or systems setting you think may further improve performance? And will Helicon be doing any testing to further optimize the software to make the best use of the 12th Gen CPUs - PARTICULARLY when it gets on Windows 11, which I'll be doing in the next week or 2?

Your benchmark results are as expected, even slightly better. So the question is why the actual performance while rendering wasn't as expected.

The most important question: were you using the same version of Helicon Focus on both computers for benchmarking? Were you using the latest version 8.0.1 on your new i9-12900 PC? If it wasn't version 8, that would explain the results. "Auto" for parallel image loading used to mean "Disabled", but now it means "Enabled". So switching from Auto to Enabled should do nothing in Focus 8, but it provides a performance boost in Focus 7. However, if the source images and / or the cache folder is on an HDD rather than SSD, activating this feature may result in performance drop instead, or in no significant change. This could be the reason your old PC did not benefit from activating this feature.

Also, memory cache limit of 10% = 6.4 GB on your computer (10% of 64 GB RAM), it could be a bit low. I recommend bumping it to at least 20% unless you really need that RAM for other tasks.

Thanks Catherine! For some reason I'm not getting emails when I get a reply - PROBABLY user error and I forgot to check the box!

NO, I was using 7.5.5 on both computers because I DIDN'T want to change to V8. So, that would explain why when I changed it to enabled I got such a HUGE improvement.

I also didn't change the memory because I was running a benchmark, but I'll change to to 20 or 25% or try some tests to see where the diminishing returns kicks in.

I'm going to put V8 on today and try it again. Hopefully, it'll remain as fast or faster than 7.5.5.

I also did all this with the system completely standard, so other than XMP1 for the memory, nothing has been tuned yet - that's next...

Just to clarify, neither the memory cache limit nor the parallel image loading option affects benchmark results. They only apply to actually using the software.

Thanks Catherine.

I put 8.0.5 Beta on last night.

Worked fine. I'm using a hand stopwatch, so my ability to time accurately sub-second is non-existent. BUT, as near as I could tell, 8 is as fast as 7.5.5 - in both cases I was less than 12 seconds for a 14-shot D850 stack.

I DID a bunch of renders, restarting from Lightroom each time.

The AMD 5700 XT is running the newest drivers - the 12/3 set.

The highest GPU Utilization I saw during the render was 25% - doesn’t mean that’s actually the highest, just that’s the number it displayed.

Dedicated GPU memory jumped to about 5.7-6.5GB during processing.

GPU Memory jumped to 6.0-6.1GB during processing.

System memory usage jumped about 9.0-9.2GB during processing.

Shared GPU memory didn't go up much - maybe .1 from .3-.4 GB.

Nothing interesting or exciting as near as I can tell, and it seems happy with the threshold at 20%. The new system is about 3.5X faster than the old one with the same GPU.

It will be interesting to see if Windows 11 really IS any faster once I go to that... I have to get past all my whining and complaining about how ugly and unproductive the UI is (I have it on the laptop), but eventually I'll bite the bullet and upgrade.