Impact of RAM Capacity on PixInsight Performance: 64GB vs 128GB

Good question to ask to the PI team over there in the PI forum.

64 gigs is a lot of RAM. I don't use PI so I don't know how much of a RAM hog it is but for most applications, 128 would be overkill. I would look more for a balanced system, large number of CPU cores, fast memory buss, fast IO.

I put 128GB in my machine a few months back. I'm not sure I've ever seen even half of it used before.   Even running full frame data sets.    I upgraded from the 32GB I had because pix would always run into memory errors when doing 7 channel full frame.     Adding more RAM isn't going to make things faster in pixinsight.   Speed in WBPP going to improve with core speed and core count….  and some NVME drives for scratch/loading speeds.

You’ll see an exponential increase in PI performance by switching the processing box to Linux. Not even remotely close performance on other platforms.

I have 128GB. When I stack with Pixinsight I see my RAM get to 80-90% quite often. When I process photos I have seen it get to 50%. Often I will have 3 masters open with a 2x Drizzle to start. Then as I process I will clone and try some other things and I might have 10-15 different images open at once. My AIO CPU cooler has a screen that displays system information so I watch RAM often actually. Went to 88% when I stacked my Rosette Nebula. 500 50mb images. When I stacked my horsehead I had 725 photos and it went to 96% or something just below 100%

If you can afford it, get 128gb. So many times I have been close to using it all. A con is I can only run it at 5200mhz (DDR5) due to hardware limitations of quad sticks of ram.

Dark Matters Astrophotography:

Tony Gondola:
You can get the same speed increase just by switching to Siril ;')
Honestly though, what's the exponent?

PI runs very optimized on Linux not to mention the platform has much improved memory management of its own.

Try it. You’ll be very surprised at the performance difference. I even tested it in a VM on a Windows Host machine and the VM easily beat its own host with the exact same data.

Do you have an actual performance data to back this up? It sounds like you do. If the difference is large people can save a lot on buying new hardware.

Dark Matters Astrophotography:
From 2018:

https://www.cloudynights.com/topic/612450-threadripper-1950-linux-or-windows-shootout-need-some-linux-guys/

Those are much smaller subs than in use today and I can tell you that we process significant amounts of IMX461 data from two of our systems (cdk14, dr500) monthly and the actual performance increase is staggering for Linux. I can take a 600 sub integration of extremely large subs (461 subs are 200mb each in 16 bit raw format) and run it on Windows in 9 hours or Linux in 3 hours.

But don’t take my word for it. You can test this with your own data and see for yourself. The difference is not minor at all.

Thank you! How much memory do you have in your linux  setup?

Dark Matters Astrophotography:
From 2018:

https://www.cloudynights.com/topic/612450-threadripper-1950-linux-or-windows-shootout-need-some-linux-guys/

Those are much smaller subs than in use today and I can tell you that we process significant amounts of IMX461 data from two of our systems (cdk14, dr500) monthly and the actual performance increase is staggering for Linux. I can take a 600 sub integration of extremely large subs (461 subs are 200mb each in 16 bit raw format) and run it on Windows in 9 hours or Linux in 3 hours.

But don’t take my word for it. You can test this with your own data and see for yourself. The difference is not minor at all.

That thread is really all over the place with some results showing a huge Linux advantage and others showing the opposite, some people overclocking, etc.  There's no question about the stability of Linux but an exponential speed increase is really not proven. There are just too many variables. Also, benchmarks are nice but what really needs a speed increase in PI is WBPP, the rest is really not an issue.  I guess the only way to really prove the point would be to run PI on a current (2025) machine setup to dual boot win/Linux both staking the same data set. I really don't want to muck around with my main imaging/processing machine myself but maybe someone out there is willing.

I don't question your knowledge in this area but some real world numbers would be nice. phrases like "Absolutely obliterate", "exponentially faster" or significantly faster" just don't do it for me. You work with linux all the time so I would thing it wouldn't be hard to put some real world numbers out there. I'm not trying to bust your xxxx but if a big improvement can be had just by dual booting Linux and slapping PI on it, I think the community would love to know. It would be a shame if someone spent several thousand dollars on a new rig when all they really had to do is install some free software.

Rafael Sampaio:

Dark Matters Astrophotography:

Tony Gondola:

Dark Matters Astrophotography:
From 2018:

https://www.cloudynights.com/topic/612450-threadripper-1950-linux-or-windows-shootout-need-some-linux-guys/

Those are much smaller subs than in use today and I can tell you that we process significant amounts of IMX461 data from two of our systems (cdk14, dr500) monthly and the actual performance increase is staggering for Linux. I can take a 600 sub integration of extremely large subs (461 subs are 200mb each in 16 bit raw format) and run it on Windows in 9 hours or Linux in 3 hours.

But don’t take my word for it. You can test this with your own data and see for yourself. The difference is not minor at all.

That thread is really all over the place with some results showing a huge Linux advantage and others showing the opposite, some people overclocking, etc.  There's no question about the stability of Linux but an exponential speed increase is really not proven. There are just too many variables. Also, benchmarks are nice but what really needs a speed increase in PI is WBPP, the rest is really not an issue.  I guess the only way to really prove the point would be to run PI on a current (2025) machine setup to dual boot win/Linux both staking the same data set. I really don't want to muck around with my main imaging/processing machine myself but maybe someone out there is willing.

In real world usage PI on Linux will absolutely obliterate PI on Windows. If there were an Apple Silicon optimized build of PI it too would obliterate PI on Windows.

Windows does a lot of things well. One thing it has never done well is managing memory. 

We process terabytes of data monthly. I can assure you that Linux being significantly faster is a fact, and one that can easily be tested by anyone. The platform is completely optimized to run on Linux so this should not come as a surprise to anyone.

Thank you! 😊 And what memory amount would you recommend? I will be using a Macbook Pro M4 Max . 64gb is enough, or would you recommend 128gb?

The more the merrier. What you want to minimize is the paging the system is doing to swap. I’ve seen large integrations use 150GB of page files on Windows so increasing the system memory to 256GB in a case like that would be warranted.

i would just get the most you can comfortably afford and watch the system statistics while your integrations are running. It’s really a shame that the current PI client still uses emulation instead of native Apple Silicon support.

I need to save money then to build a PC that is up to the job/task for PI, i am a windows user, i tried Linux few times but i have no idea how to use, so if i will use PI then i have to check out people builds, but i am not rich like most of people here, i am barely using 64GB RAM on to systems but not strong specs, so even if i upgrade to 128GB i am afraid that it won't give me any big improvement if the specs overall isn't to high standard even if i use Linux, so i would like to see people here what they are using as computers so then i can decide if i have something reasonable or must build one if i can afford or just give up astrophotography before it is getting more expensive spending for me.

Rafael Sampaio:

Tareq Abdulla:
I need to save money then to build a PC that is up to the job/task for PI, i am a windows user, i tried Linux few times but i have no idea how to use, so if i will use PI then i have to check out people builds, but i am not rich like most of people here, i am barely using 64GB RAM on to systems but not strong specs, so even if i upgrade to 128GB i am afraid that it won't give me any big improvement if the specs overall isn't to high standard even if i use Linux, so i would like to see people here what they are using as computers so then i can decide if i have something reasonable or must build one if i can afford or just give up astrophotography before it is getting more expensive spending for me.

I don’t think you need to give up astrophotography because of that. You can definitely use an affordable computer. While a high-end one might speed things up, you can still create amazing images without it, no doubt

It is about speed what i am trying to say, i mean i can use PI on my old 7th gen laptop, i did stack some on that laptop, or even my old build that is using 3rd gen Intel, but if i will build one latest gen i can't just afford 16-100 cores CPUs maybe, and going with 128GB RAM isn't cheap either, will need a beefy motherboard too, all that to make things faster, it is no longer about "It is a budget computer that is still working fine", we all buy new stuff and upgrade about setup, even my imaging setup is upgraded so i won't keep using old computers and i won't just do 20-40 frames only like before anymore, even PI itself updated with many scripts and addons that will make computers scream, so i just don't go that far there as you only because my computer can't handle heavy tasks so i don't try, amazing images can be done these days with mobiles too but i am not spending $$$$$$$$$$$$$$$ for just amazing images that are like mobiles images and they aren't winners and top picks here and there, no point for me, but i will watch this topic anyway.