Best processor for pixinsight?

M

Madmaz

Member
Am sure this has been raised before, but wondering what is the most cost-effective investment for processing performance for PI?
Processors like AMD Ryzen 9 with 16 cores/32 threads or is there cheaper options which may have less processors etc but work better with PI
Current PC is a i7 with 4 cores/8 threads but faster processing would be a major bonus to reduce overall data processing activities
Cheers
Keith
 
Keith.

Right now it is not possible to give a definite "This is the best value hardware platform for PI" given the current world-wide shortage of processors, motherboards and memory that has driven up component prices, more than treble in some cases, and also resulted in very long backorder times.

As a PI user you will know that there is a built in platform benchmark tester (look under Script > Benchmarks > PixInsight Benchmark) that produces an overall performance score for your particular system and allows you to submit those scores to the on-line benchmark library, the higher the score the faster any particular system. If you visit the PI Benchmark web page, click on the highest scores in the benchmark list and then click on any listed serial number against that score you can read details of the basic configuration of that particular system, CPU model and clock speed, Memory size, Swap disks, Operating System etc plus any added user notes.

You can find the Benchmark scores here:


By comparing your own PI benchmark test score against others in the list you will be able to determine what level of hardware upgrade you would need to double, treble, quadruple, or more, your current system performance.

In general, the higher the number of (real) CPU cores, maximised RAM capacity and multiple hard drives that allow for separate swap partitions on individual drives (using the fastest drive technology) that you have, the better the performance will be.
Graphics cards that support GPU co-processing may become more important in the future but as yet few PI processes can utilise that feature.
Note also that PI is developed on a Linux system and it has been stated previously that a Linux platform delivers the best performance.

As for cost effectiveness only you can be the judge of that given the state of the semiconductor industry at this time.

William.
 
The AMD Ryzen 9 5950X processor is the most cost-effective option currently available for a high-end desktop running PixInsight. We have just completed construction of a new workstation and here is an initial benchmark:

PixInsight — PixInsight Benchmark

pixinsight.com pixinsight.com

This is on Kubuntu Linux 20.04.3 LTS with a default hardware configuration and just 8 swap threads on the system /tmp directory. There are other samples with even better performance on overclocked machines and using more sophisticated swap disk configurations. The performance is spectacular for a 16-core CPU at $750.
 
Juan Conejero said:
The AMD Ryzen 9 5950X processor is the most cost-effective option currently available for a high-end desktop running PixInsight. We have just completed construction of a new workstation and here is an initial benchmark:

PixInsight — PixInsight Benchmark

pixinsight.com pixinsight.com

This is on Kubuntu Linux 20.04.3 LTS with a default hardware configuration and just 8 swap threads on the system /tmp directory. There are other samples with even better performance on overclocked machines and using more sophisticated swap disk configurations. The performance is spectacular for a 16-core CPU at $750.
Click to expand...
Juan,

Just one minor item of note out of curiosity. Are you sure that you’re system isn’t overclocked? That Asus motherboard, and most enthusiasts motherboards, tend to have aggressive, not actually within default AMD spec settings to give them an edge in reviews. As a fellow Asus/5950x owner, your score looks like what a 5950x should be getting with PBO “on” or one of the Asus performance settings (which enable PBO without telling you) and good cooling, not a truly stock 5950x that conforms to AMDs base specs. Technically, anything set higher than the default spec is generally considered overclocked; this includes tools AMD ships such as PBO or motherboard manufacturers playing with settings.

This isn’t an issue, but I wanted to point it out as other users might wonder why their 5950x runs several thousand points lower on a B550 or less aggressive motherboard. Years ago when Intel starting encouraging playing with PL2 settings Anandtech did a write up. They updated it when AMD users on motherboards actually following AMDs spec were reporting lowers scores. You can find that here: https://www.anandtech.com/show/1487...gning-perception-with-amds-frequency-metrics-

Check your settings and check your cooling. This is all intended as educational. Thanks for making great software!
 
Good advice on the AMD Ryzen 9 5950X processor. I choose this and 128GB RAM to build a new workstation for PixInsight last weekend. This wasn’t difficult to put together after watching a couple of YouTube videos and booted up without any problems. The benchmark test results look good.

PixInsight Benchmark test

The processing speed improvement compared to my Mac Pro which is a few years old was exponential. Starnet took 75 seconds on a ASI6200 full frame size file, as opposed to over 20 minutes before.
 
NeilCorke said:
Good advice on the AMD Ryzen 9 5950X processor. I choose this and 128GB RAM to build a new workstation for PixInsight last weekend. This wasn’t difficult to put together after watching a couple of YouTube videos and booted up without any problems. The benchmark test results look good.

PixInsight Benchmark test

The processing speed improvement compared to my Mac Pro which is a few years old was exponential. Starnet took 75 seconds on a ASI6200 full frame size file, as opposed to over 20 minutes before.
Click to expand...
Hey Neil, fantastic results you got there. My 12900k is a bit slower but I am amazed by your swap speed. May I ask what Drive you are using? I guess NVMe 4th Gen? What specific one? My Samsung Evo 970 Plus takes about 10 seconds to swap files. Seems a bit slow with the 1600MiB/s although the SSD can do 3500.

My results:


CPU Identification

CPU vendor ............. GenuineIntel

CPU model .............. 12th Gen Intel(R) Core(TM) i9-12900K



System Information

Platform ............... Windows

Operating system ....... Microsoft Windows 10 Pro

Core version ........... PixInsight Core 1.8.8-12 (x64)

Logical processors ..... 24

Total memory size ...... 31.747 GiB



Execution Times

Total time ............. 00:23.93

CPU time ............... 00:13.77

Swap time .............. 00:10.13

Swap transfer rate ..... 1635.953 MiB/s



Performance Indices

Total performance ...... 19656

CPU performance ........ 27495

Swap performance ....... 9061
 
Coldt_Astro said:
Hey Neil, fantastic results you got there. My 12900k is a bit slower but I am amazed by your swap speed. May I ask what Drive you are using? I guess NVMe 4th Gen? What specific one? My Samsung Evo 970 Plus takes about 10 seconds to swap files. Seems a bit slow with the 1600MiB/s although the SSD can do 3500.
Click to expand...

Coldt_Astro,
Like Neil, I built a custom box primarily for processing images with PI - which then makes it ideal for just about everything else: PS or whatever.
I also have 128Gb memory, the fastest that was available - and with timing tuned lower as well.
I've found two ways to get the swap speed over 8Gb/sec in Windows or over 11Gb/sec in Linux:
- Ramdrive. I use SoftPerfectRAM disc software (in Windows, or tmpfs in Linux) to create ram 4x ram drives (L:, M:, N:, O:) that I use like RAID in the PI preferences (L,M,N,O,L,M,N,O,L,M,N,O,L,M,N,O). With 128Gb, I have plenty of memory to spare to creat 4x 8Gb RamDrives.

PixInsight — PixInsight Benchmark

pixinsight.com pixinsight.com

PixInsight — PixInsight Benchmark

pixinsight.com pixinsight.com
- NVMe v.4 drives. I have 3x 980Pro drives in my system that I tried as temp files and was pleasantly suprised to get nearly the same xfer rate out of them as I did memory. Again, I put them in RAID type config (C,D,E,C,D,E,C,D,E,C,D,E,C,D,E) to get nearly 8Gb/sec

PixInsight — PixInsight Benchmark

pixinsight.com pixinsight.com
 
I used a Samsung 980 Pro 2 TB M.2-2280 NVME drive in the Linux machine. I didn't do anything else, like the steps suggested by BHammy above, but will look at these now!
 
I achieved less than 2.5s now. If anyone wants to replicate it, I wrote it down in the comment section of the report: https://pixinsight.com/benchmark/benchmark-report.php?sn=8H2Y2DDD4MDO3SQ9NYHSPQHSM3MR4ZZ9
Anything more than 4GB Ram Disks gave me worse results. (I have a fixed budget). Do not make a Ram Disk with more than 4gb. Instead, do multiple 4GB ones. Same for Locations on 980 Pro. More than 8 locations gave me worse results.
Your results may vary.
Have a pleasant day
 
I have done some fairly extensive testing regarding # Swap Files following an upgrade from a Ryzen 3700X to 3900X and was quite surprised at the results.

I am using a Samsung 980 1TB M.2 NVMe (PCIe 4.0 X570)

For my system the optimum number of Swap files is 40.

1661014001508.png
 
I'm in the process of buying an off the shelf desktop PC to upgrade my PI processing capability. I read a post over a year ago that a GPU is not really relevant for PI processing but it may be important in the future. Before I buy the desktop, I thought I'd bounce out a post to see if there is any further update on the utilization of graphics cards/GPU for PI processing. I'm planning on putting my money into a fast CPU, SSD and RAM and if I don't have to worry about spending money on a high end GPU, I'll stick with one that's cheaper. Any thoughts would be much appreciated. Paul
 
Takman120 said:
I'm in the process of buying an off the shelf desktop PC to upgrade my PI processing capability. I read a post over a year ago that a GPU is not really relevant for PI processing but it may be important in the future. Before I buy the desktop, I thought I'd bounce out a post to see if there is any further update on the utilization of graphics cards/GPU for PI processing. I'm planning on putting my money into a fast CPU, SSD and RAM and if I don't have to worry about spending money on a high end GPU, I'll stick with one that's cheaper. Any thoughts would be much appreciated. Paul
Click to expand...

the star removal tool starnet and russ croman's Xterminator tools can use the GPU, but you have to do a bunch of work to get CUDA installed on your machine and then get CUDA-enabled tensorflow libraries. most mainstream NVidia GPUs will work for this purpose; you don't need to get a video card with a billion gigs of VRAM.

outside of that, PI itself does not yet utilize your GPU for compute.
 
pfile said:
the star removal tool starnet and russ croman's Xterminator tools can use the GPU, but you have to do a bunch of work to get CUDA installed on your machine and then get CUDA-enabled tensorflow libraries. most mainstream NVidia GPUs will work for this purpose; you don't need to get a video card with a billion gigs of VRAM.

outside of that, PI itself does not yet utilize your GPU for compute.
Click to expand...
Perfect! Thanks a lot for your fast reply, pfile. I can get around a computer all right but I'm no expert so CUDA is something I'm not familiar with although I've seen it referenced in a few forums and that it is a pain to install. Are you saying that I would need to install CUDA in order for realize the benefits of a GPU? I'm looking at a MSI or ASUS desktop with 128GB RAM, i9-13900 CPU (24 core/32 thread), Nvidia GeForce RTX 4070Ti (12GB) and 2-4 TB SSD. Do you think that is overkill for the GPU? Paul
P.S. - I run all of Russ Croman's tools.
 
Takman120 said:
Perfect! Thanks a lot for your fast reply, pfile. I can get around a computer all right but I'm no expert so CUDA is something I'm not familiar with although I've seen it referenced in a few forums and that it is a pain to install. Are you saying that I would need to install CUDA in order for realize the benefits of a GPU? I'm looking at a MSI or ASUS desktop with 128GB RAM, i9-13900 CPU (24 core/32 thread), Nvidia GeForce RTX 4070Ti (12GB) and 2-4 TB SSD. Do you think that is overkill for the GPU? Paul
P.S. - I run all of Russ Croman's tools.
Click to expand...

yes, if you want the Xterminator tools to use the GPU you need to install CUDA and then the CUDA-enabled tensorflow. as you probably know the Xterminator tools are pretty slow when running on the CPU. it will get marginally faster with your new CPU.

i don't know a lot about nvidia's product line, but 12GB seems pretty normal these days for GPU memory.

rob
 
Yea, Croman's tools are excellent but they sure do run slow on my current laptop. Good to know about the GPU memory since I've never paid attention to it before. Course, until I started processing with PI, I never had to worry too much about my laptop's processing capabilities for normal work. Taking 53 hours to process a 50 hour integration of 120MB subs was a pain so I decided to sell some gear and upgrade. Thanks for your quick responses, Rob.

Paul
 
You must log in or register to reply here.
Back
Top