Seeking Pixinsight swap/system performance Masters

[photonhunter1]

I'm seeking help from the PI performance masters out there. I installed AMD Radeon Ramdisk and have run multiple scenarios. Using the PI Benchmark script, system performance with and without Ramdisk is crazy good - a major increase in performance:

Total Performance without (5787) and with (12304)
CPU Performance without (10576) and with (11261)
Swap Performance without (2019) and with (20017)

However, when processing 50 darks with Image Integration, the Process Console shows run times of 1:30 without Ramdisk (Memory in Use was 5GB per Task Manager) and 1:23 with Ramdisk (Memory in Use was 16.8GB per Task Manager) which I would expect since I've allocated 12GB to Ramdisk. Why would I only see a process time improvement of 7 seconds?? Is Image Integration not the best measure of Ramdisk performance?

Per the PI Benchmark script the best all around configuration seems to be 12GB Ramdisk and 6 swap directories pointed to my Ramdisk (drive P:/) with an additional swap directory pointed to C:/...temp (which resides on my 256GB M.2 SSD drive). Below is a screenshot of PI configuration.

The balance of my desktop configuration is ASRock AB350 Pro4 mother board. AMD Ryzen 1700x CPU 3400 mhz. 2-16GB sticks of Ram, overclocked to 3266 mhz, running on win10 pro.

Do I need to configure my Ramdisk or PI Directories and Network differently? Any help would be appreciated! Thanks.



Mike

[pfile]

i'm not even sure the PI swap system is involved in ImageIntegration; how did you run this test? you'd probably have to point PI at the disk for swap, reboot, test, and then reboot again, point PI at the ramdisk and then test again.

the speedup you saw might have just been due to windows caching the input files in memory; the memory in use by the ramdisk might just be statically allocated and the fact that it shows up in task manager might not reflect whether or not it was actually accessed...

rob

[photonhunter1]

Rob, my testing (using the PI Batch Test script) actually included running Image Integration with and without Ram disk with a reboot in-between. I have PI pointed to 6 instances of the Ramdisk swap directory in the global Networks and Directory setting (image in the original post), along with one instance of a swap directory from my 256gb SSD M.2 drive.

To your point, perhaps Image Integration isn't the best program to run/test. I'll continue to test as I progress on my processing of my latest image.

Mike

[pfile]

it would be good if juan commented on this - i see a lot of people on CN tuning their system for the benchmark. it's not clear at all to me that swap performance is ever the long pole in PixInsight performance - i think at most any process that runs on an image reads the swap once (if there is a mask) and writes the swap once (to update the swapfile with the results of the current process.). i might be off by one here, there might be another read. but my point is that if your process is CPU bound (and most are - even the disk intensive ones juan rejiggered a couple of years ago to make them as cpu-bound as possible) then even if the swap read/write time were 0 it might only increase your overall performance by a couple of percent.

rob