PixInsight 1.8.0 RC3 'Ripley' Released

[derrickf]

Thank you for your report. We have detected a performance regression in version 1.8.0 RC3, which affects irregularly to some tools. I have already identified and fixed this problem today, and right now I'm preparing a new 1.8.0 RC4, which hopefully will be the last one before the final 1.8.0 release.

Thank you - I'll watch out for RC4 and feedback my observations when I have tried it.

One possibility is that the performance loss be hardware-dependent, being worse on AMD processors. Not too surprising because the problem is related to atomic operations (reference/dereference operations in multithreaded code).

It will be a great pity if this performance issue persists on AMD processors; with the level of performance I experienced with RC3, PI is pretty much unusable.
In version 1.7 the results I was achieving with the "standard" benchmark referenced in another thread were comparable to results reported for machines running Intel I7 processors so I am somewhat surprised that  there is such a huge difference in 1.8 RC3

It would be interesting to know if any other forum members have tried running 1.8 on machines with AMD processors and what their experience of performance has been.

Thanks
Derrick

[Juan Conejero]

Hi Derrick,

I have just released two updates with new versions of the MaskGeneration and MultiscaleProcessing modules. Please let me know if they run faster on your AMD machines. Thank you!

[derrickf]

Hi Derrick,

I have just released two updates with new versions of the MaskGeneration and MultiscaleProcessing modules. Please let me know if they run faster on your AMD machines. Thank you!

Hi Juan,
I reloaded RC3 and applied the updates on the 6 core machine; I'll do some more exhaustive testing tomorrow but here's my initial impression

• Star Mask is now probably a little faster (maybe 10% or so) than in 1.7

• Realtime preview of noise reduction with ATWT is still slow when run on a preview(subframe of the image) However running the ATWT noise reduction on the whole image is pretty fast (similar to 1.7) - in fact it takes the same time as the sub-framed real time preview irrespective of the preview (subframe) dimensions so maybe something has gone awry here

• Realtime preview on a subframe of noise reduction using MMT is fast (about 3.5X faster than ATWT on the same subframe). MMT noise reduction of the whole image is at least as fast as 1.7

• I tried Deconvolution and a few other processes and, subjectively, they appear to run a little faster (approximately 10%) than in 1.7

So it appears that the fixes you made available have worked the required magic  8)
If I find anything useful tomorrow in my further use of this version I will let you know. If there is anything specific you would like me to try just let me know.

Thanks & Regards
Derrick

[kwiechen]

Thank you all for your nice words!

Right now I'm writing several documents describing the new features, including some key development topics (the new threading model, the new real-time generation model, etc...). I'll post them on the forum as soon as I have something finished.

I ask for a bit of patience while I manage to write something coherent

maybe, I have missed something?

Kai