Jump to content

SkySurveyBanner.jpg.21855908fce40597655603b6c9af720d.jpg

PC build for processing


Recommended Posts

  • Replies 45
  • Created
  • Last Reply

IMHO getting or building a PC is not what I would do. It ties you down to a fixed point in the house. I use a "high end" laptop. A Dell Precision M4700 i7 Work Station, with an upgraded premier graphics screen set to Adobe RGB. This eats data and I used one in my working life for complex machine designs using SolidWorks 3D CAD design software. I was lucky to purchase mine on fleabay for £499. Ok I've added a SSD drive and replaced the DVD drive with a 4TB hard drive for image storage, but the standard specification is really good.

 

Steve

Link to comment
Share on other sites

11 hours ago, Alien 13 said:

Its hard deciding on PC specs but for compatibility on all software I would go Intel/Nvidia every time. might not always be the fastest on a Friday afternoon in the third week of the month but always works.

Alan

My last Nvidia/Intel desktop rig was a P3 550, never had an issue with using non intel stuff. Laptop is an I7, I cant see any issue with either Intel or AMD, updates and patches are just as frequent, drivers are as good (or bad) on both. Meltdown appears to have hit Intel more than Spectre hit AMD.

Link to comment
Share on other sites

1 hour ago, The Admiral said:

I've just run a test of DSS on the elderly i7 computer I described above, and DSS took about 17 minutes to process 50 darks, 50 flats and 100 lights, each a ~46MB dng file. DSS was v4.1.0, 64 bit.

Ian

Funnily enough, DSS doesn't take much longer for me. My last project, I used it to stack 60 x 600s lights, 40 darks, 150 flats, over 100 bias, 2x drizzle and it was done in around half an hour. Pixinsight though seems to be a law unto itself!

Link to comment
Share on other sites

33 minutes ago, sloz1664 said:

IMHO getting or building a PC is not what I would do. It ties you down to a fixed point in the house. I use a "high end" laptop. A Dell Precision M4700 i7 Work Station, with an upgraded premier graphics screen set to Adobe RGB. This eats data and I used one in my working life for complex machine designs using SolidWorks 3D CAD design software. I was lucky to purchase mine on fleabay for £499. Ok I've added a SSD drive and replaced the DVD drive with a 4TB hard drive for image storage, but the standard specification is really good.

 

Steve

Not too bothered about being in a fixed point. Its no different to how I work now. Laptop on the edge of the sofa in the living room when the kids allow it lol

Link to comment
Share on other sites

13 minutes ago, david_taurus83 said:

Pixinsight though seems to be a law unto itself!

I wonder if it's worth starting a new thread and asking specifically what computer specs PI users have, and whether they are satisfied with performance. I don't think this current thread will ellicit that sort of information.

Ian

Link to comment
Share on other sites

Lots of good points already, but here are few thoughts.

My 'under-the-desk' tower is now on its 5th upgrade.  The case itself must be twenty tears old, but allows things like front USB ports, Card readers and the like in the drive bays.  The current bprocessor  is an i7-4790 running at 3.6 Ghz.  As pointed out this allows multi threading and gives 8 cores which means DSS finishes stacking 30 lights, 30 bias, flats &c in under 10 minutes.  16Gb ram and a 64 bit operating system (Win7 home premium) all help.  It finishes a Star Tools job in 15 minutes or so.  The operating system lives on a small SSD, the software on a 1 Tb HDD and the data on a 4 Tb HDD.  This was more by accident than design and is almost certainly not the most failure proof arrangement.

Getting a good power supply that matches the motherboard is useful and make sure that you can get the cable you need to connect whatever graphics card to the display device you end up using (not all  HDMI /DVI sockets match easily, I found).

Link to comment
Share on other sites

2 minutes ago, almcl said:

The operating system lives on a small SSD, the software on a 1 Tb HDD and the data on a 4 Tb HDD.

I have a similar arrangement. OS on one drive, program software on 2nd drive and data/media on 3rd & 4th drives. The thinking being that more data can be read simoultaneously, whether or not it makes a difference in reality, I'm not sure?

Link to comment
Share on other sites

4 hours ago, david_taurus83 said:

Pixinsight though seems to be a law unto itself!

 

4 hours ago, The Admiral said:

I wonder if it's worth starting a new thread and asking specifically what computer specs PI users have

The place to look for information specifically regarding PixInsight performance on different platforms is the PixInsight benchmark list.

http://pixinsight.com/benchmark/

Have you ever run PixInsight's built-in benchmark test on your current laptop? It can help you optimise the setup of the swapfiles etc to leverage a little more performance from your current system (swap files should be on the SSD). The bigger the number generated by the benchmark test the faster it processes images. In the benchmark results page linked above click on any of the results and you can read the specification of the system(s) that generated it.

I have a copy of PixInsight on an old back-up Windows laptop with an i3 processor and a SSD but is woefully slow and I never use it for image processing. My main system for image processing is an iMac Pro with an eight core 3.2Ghz Zeon processor, 32GB 2666Mhz DDR4 error correcting memory and a 1TB SSD, this system is lighting fast but of course is way outside the price point you are looking to spend. It was an interesting comparison though with my previous iMac, an i7 5K Retina with 32Gb memory and 500GB SSD, I would say the Zeon processor with its eight real cores is three to four times faster than the Hyper-threading i7 for processing in PixInsight.

Having spent a previous lifetime working with UNIX and SUN Solaris OS's I felt more comfortable with Mac OS and a business meeting with Microsoft execs back in 2005 where they announced they were not expecting future Windows OS to be used in system control and were not prioritising that facet of the Windows OS I decided not to invest too much time with Windows and am quite happy with Mac OS and Linux. I notice quite a few of the highest performance benchmark reports are from Linux systems.

P.I. is a processor intensive application and uses all available cores to the maximum so virtual cores AKA Hyper-threading does not help a great deal in speeding up P.I's performance, real cores would be more of an advantage in this respect. Hyper-threading is more of an advantage where multiple applications are being run concurrently but where the majority of the applications are underutilising an individual core and can share it with an application that is using it's core at 100%.

The P.I. support forum has been abuzz in recent years with talk of developing the core code to utilise the under used graphics card GPU to help with the heavy mathematical processes but the reality is this is a huge undertaking and no-one is talking a timescale. Until a standard way of accessing the GPU processor is agreed industry wide I doubt there will be much progress other than the few demo systems currently doing the rounds that were written around specific graphics cards, bespoke tailored OS's and customised motherboards. As of today, P.I. does not need a high end graphic card since it does not use or need shading engines or multiple pipelines, P.I. is basically a 2D graphics application and money spent on a high end graphics card would be wasted unless you are into gaming or other 3D graphics intensive applications.

For P.I. the processor, memory, disk and motherboard are the prime drivers for system performance.

Thats my five bobs worth.....

Link to comment
Share on other sites

1 hour ago, The Admiral said:

I couldn't get Oddsocks' link to work

It’s working ok for me Ian but it is slow to load, the page has become quite large as users add their own benchmark scores, the page may appear to hang during loading, if you just wait for around ten seconds or so it will complete, I guess it depends too what your broadband speed is and whether your A.V. program is pre-scanning the web page prior to displaying it etc.

The benchmark test to appraise your own platform is built into P.I. You will find it under scripts. You need to run the benchmark test at least three times before you can upload your own report and add it to the database, this is to average out variations in the test due to the O.S. running background processes such as checking for updates, disk indexing, virus scanning etc etc.

 

Link to comment
Share on other sites

Thanks for the clarification Oddsocks, though for info I don't use PI, just an interested bystander! Did I spot correctly from the discussions on the PI forum that Linux swaps to RAM, and Win and OS to SSD, which explains the speed difference?

Ian

Link to comment
Share on other sites

16 minutes ago, The Admiral said:

Did I spot correctly from the discussions on the PI forum that Linux swaps to RAM, and Win and OS to SSD, which explains the speed difference?

I believe you can configure virtual ‘RAM’ swap disks for P.I. in all three O.S’s if you want but this would only be practical if you had masses of spare RAM available and the number/size of images you were processing would comfortably fit in the RAM disk(s), certainly, if you had say 32Gb of RAM on a system it would be no problem to create a couple of 8gb virtual disks in RAM and tell P.I. to use them for swap files.

But you are correct that LINUX uses RAM for page swaps by default where Windows and Mac write page swaps to disk, one of the reasons that LINUX is so popular for servers which are normally just shuffling files from one location to another and not actually carrying out ‘work’ on the files.

Link to comment
Share on other sites

I'm gobsmacked that folks seem to be needing such high spec machines.

I just used DSS to stack 42 lights, 24 darks, 42 flats (I think DSS found a master flat as it too a minute and half to stack the darks and didn't take more than a flash to stack the flats) and a master bias. It took eight minutes. (Edit 12Mb RAWs, just in case you think I'm stacking small images!)

I've got a pretty pedestrian machine, no super duper graphics/games machine:

image.thumb.png.61f8e9779e45565abe2da80f7b75d10f.png

 

Link to comment
Share on other sites

52 minutes ago, Stub Mandrel said:

I'm gobsmacked that folks seem to be needing such high spec machines.

I can only assume that the way PI works makes a high demand on system resources, presumably quite different to the way DSS works. From what I've seen it seems the swap file activity of PI has quite an effect.

Ian

Link to comment
Share on other sites

16 minutes ago, The Admiral said:

I can only assume that the way PI works makes a high demand on system resources, presumably quite different to the way DSS works. From what I've seen it seems the swap file activity of PI has quite an effect.

Ian

I'm confounded how it seems to take so much longer to do the same things. DSS uses all four of my 'virtual' cores and slows down low-priority programs, but normally I just leave it chuntering and do something else and don't notice a performance hit.

I can see how PI's advanced processing may require more resources, but surely applying control frames and stacking is going to demand essentially the same from any program?

Link to comment
Share on other sites

28 minutes ago, Stub Mandrel said:

I'm confounded how it seems to take so much longer to do the same things. DSS uses all four of my 'virtual' cores and slows down low-priority programs, but normally I just leave it chuntering and do something else and don't notice a performance hit.

I can see how PI's advanced processing may require more resources, but surely applying control frames and stacking is going to demand essentially the same from any program?

DSS runs fine on my laptop also. No more than half an hour for a big stack. PI, not so. I restarted my current project again this morning. I am now currently on the Approved based on SNR values part of preprocessing. Next up is Star alignment prior to Image integration. Applying drizzle and local normalisation also....sigh.... To be fair though, I have 6 sets of calibration frames for the 6 nights I captured data. The calibration of darks and flats doesn't take that long. But when I get to these latter stages of preprocessing it starts to drag out. Example, it took almost 20 minutes for the Subframe selector script to measure my images and another 15 to output the approved. Good job it's cloudy!

Link to comment
Share on other sites

23 hours ago, The Admiral said:

I've just run a test of DSS on the elderly i7 computer I described above, and DSS took about 17 minutes to process 50 darks, 50 flats and 100 lights, each a ~46MB dng file. DSS was v4.1.0, 64 bit.

Ian

This gave me an idea Ian, drop the files into an online dropbox and we could have a "process time-off" using the same files ?

Link to comment
Share on other sites

10 hours ago, The Admiral said:

According to the discussion here, I read

"PI has all the files being integrated open at the same time. this means they are mapped into memory… big fits files then mean a lot of memory. "

Is this the same as DSS?

Ian

Doesn't seem to be. This is stacking 90 lights, nothing else. It's loading then registering each light in turn and the total memory usage isn't changing:

image.thumb.png.211567d15102b93d5c32f30cb1896c58.png

 

The interesting thing is that DSS is only using a fraction of the memory used by Firefox! It clearly only loads images when it is using them.

image.png.3cbc3199e0822c9618f1c2a130d88d31.png

 

My conclusion is that lots of memory isn't needed for DSS, but fast disc access will be a boon. Also with CPU usage between 75-100%, it's fastest when not doing other things and a half-decent CPU will help.

Stacking the 90 12MB lights to 16 minutes.

Obviously PI is doing a lot of other things, not just stacking and may use some other stacking algorithms so a direct comparison may not be fair, but it does seem DSS is a more practical choice for stacking if you have a relatively feeble computer!

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue. By using this site, you agree to our Terms of Use.