CPU upgrade and software performance

[mikeyj1]

I am thinking of upgrading my home computer (i5-3570) to an AMD Ryzen 5 or 3600 or 5600, and I wonder what sort of performance improvement I will get on software like APP / DSS and Photoshop 2019.  Although the processor clock speeds are similar, there are more 'cores and threads', and I will also be upgrading the memory from DDR3 to DDR4.

Thoughts and advice welcome..

Mike

[Budgie1]

If this is a desktop then one thing I can highly recommend is swapping out the old HDD and replacing it with a SSD. Boot up times will be drastically reduced, when I did my desktop it want from about 4-5 minutes boot up to less than a minute from pressing the power button. Software start up seems a lot faster as well.

I've always had AMD processors and they've never given any issues. I'm currently running an AMD FX 6300 6 core with 32Gb RAM, which is getting a bit long in the tooth now, I was looking at the Ryzen 7 as an upgrade, although it would need a new motherboard as well.

Sorry I can't give you any more of the performance you're likely to get.

[mikeyj1]

yes it's a desktop, and I already did the swap to  SSD (in RAID striped config), which is a good upgrade i agree.  

[licho52]

Here are some benchmarks that might give you some idea:

Astronomy software performance tests with different CPUs - astrojolo

[Shimrod]

17 hours ago, mikeyj1 said:

I am thinking of upgrading my home computer (i5-3570) to an AMD Ryzen 5 or 3600 or 5600, and I wonder what sort of performance improvement I will get on software like APP / DSS and Photoshop 2019.  Although the processor clock speeds are similar, there are more 'cores and threads', and I will also be upgrading the memory from DDR3 to DDR4.

Thoughts and advice welcome..

Mike

The performance gains you get from increased cores depends on how the software is written - some software cannot exploit all the cores available - this is especially true if using older versions of the software. You might also want to consider whether the software can exploit GPU processing which can dramatically increase performance if you have a relatively new (DX12 support) mid-range card. Photoshop and darktable are two examples of software that will exploit graphics cards. I have a Ryzen 9 3900x and DSS will use all the cores/threads available, much quicker than my previous i7 4 core/8 thread processor.

If upgrading, you might also want to consider a motherboard with PCI 4.0 support. Getting one of these m.2 drives  will give you around 10x the performance of a standard SATA interface.

[mikeyj1]

3 hours ago, Shimrod said:

you might also want to consider a motherboard with PCI 4.0 support. Getting one of these m.2 drives  will give you around 10x the performance of a standard SATA interface.

 

10 hours ago, licho52 said:

Here are some benchmarks that might give you some idea:

Astronomy software performance tests with different CPUs - astrojolo

Thanks, I like that website article, and I also subsequently found another on photoshop multi-core usage I thought I would share .  This comment from the article is interesting:

"Performance in Photoshop does improve directly with increases in processor clock speed, the higher, the better.  It is, therefore, better to utilise a processor with fewer cores but a higher clock speed if budget is limited.  If you use the PC for a variety of tasks including video editing and image processing, you may still benefit from higher core counts from the other applications".

 

I also found an article on the APP forum here which says it will use more threads and cores to improve stacking and processing performance..

I like the m2 drive idea, maybe I would use the SATA drives as a 'scratch disk' for PS and Premiere.

So my 3rd Gen i5-3570k has 4 cores and 4 threads, at 3.4Ghz.   It sounds like I would need to go for increased GHz and a 'couple of extra cores/threads' for the best price performance ratio.. which makes me think the better choice would be the 10th Gen core i5-10600k as it has higher clock speed (4.1Ghz, higher than the Ryzens) and 6/12 cores/threads.

does that sound logical?

[stash_old]

The oldest quote in IT "you are only as fast as your slowest component" is still true. E.G Not all SDD's are the same speed wise , Using 2 SSD's with data and programs spilt across the 2 SSD's(1 all data etc) , Getting your Virus checker to ignore data files in a folder and of course motherboard design.

Is the final saving in time worth it against the cost - only you can say as its down to the individual and how deep your pockets are - like Astro in general 🙂

I would use an abacus if I could have clear skies most nights LOL!

[mikeyj1]

yes sure, fair point... but that's not my question (or a relevant  answer)... 😆

I guess I am hoping someone has done a similar upgrade and can say:  "it cuts the stacking time in half" or "photoshops actions take less than 2 seconds now compared with 1 minute before"

Edited December 20, 2020 by mikeyj1

[Shimrod]

24 minutes ago, mikeyj1 said:

yes sure, fair point... but that's not my question (or a relevant  answer)... 😆

I guess I am hoping someone has done a similar upgrade and can say:  "it cuts the stacking time in half" or "photoshops actions take less than 2 seconds now compared with 1 minute before"

It depends on what software you are running, as as Stash_old says, what other components you have in your system. The latest AMD processors (3 series upwards) have a faster interface to GPUs, so if you are running an application (like Darktable or Photoshop) that exploits GPU processors, then that might be better. Of course, you need a decent graphics card, and that will mean getting something of a decent gaming standard - like an RTX 2060 or better.

More cores will benefit in other applications besides photoshop, so you need to look at the other applications that you run. Adobe themselves don't say there is an upper limit, just that you hit a law of diminishing returns against cost as you increase the cores. Of course, a software update could change all that so given the relatively small cost differential, you may well choose to get 8 cores/16 threads rather than 6. I can't give you benchmarks (I went from 4 cores/8 threads and GTX1060 to 12 cores/24 threads and RTX2060), but I can tell you everything was noticeably quicker, and I expect  this PC to last several years.

For photoshop,  here's a set of benchmarks for 8 core through 32 core systems - there are timings for the various actions carried out and you can see that there is not a huge difference between them.

[licho52]

3 hours ago, mikeyj1 said:

yes sure, fair point... but that's not my question (or a relevant  answer)... 😆

I guess I am hoping someone has done a similar upgrade and can say:  "it cuts the stacking time in half" or "photoshops actions take less than 2 seconds now compared with 1 minute before"

In DSS and PXI I get pretty much full usage on all 12 threads of my 10400.  My guess is that they will take all the threads one throws at them.  My next upgrade will be 8C/16T CPU.  If the weather was better and I was getting more subs everyday I'd consider 12/24.

In your case instead of 10600 I'd go with 10700.  Difference in price is small but 8/16 is worth it.  The software and picture complexity/size only grows with time and I really don't like to wait while processing images.  AMD offers some hefty 12/24 CPUs and I even think it may be worth taking them into consideration for heavy use.

Edited December 20, 2020 by licho52

[malc-c]

Can't speak for the latest gen Ryzen (which on most reviews have blown team blue out of the water), but investing in a decent main board that supports the latest generation of Nvme drives is a key factor.  Don't cheap out on the Nvme drive either - The Samsung Evo Pro, whilst a tad pricey is blisteringly fast.

My system is a 1st Gen Ryzen 5 1500x (4 core 8 threads) built just over three years ago. Based on a Asus 370X mainboard, 16GB Corsair Vengeance DDR4 RAM, and a Samsung 960 Pro Nvme  system drive.  Graphics is an RX550 gaming, but as I don't play games I don't really need the latest Nvida / AMD graphics cards costing a grand.  Storage is a 500GB SSD for programs (the Nvme is set as the system drive) and two mechanical drives totalling 3TB for file storage.  The thing boots in a blink of an eye, can edit and render 4K  HDR video footage without dropped frames, and stacking  40 x 5min RAW subs taken with my D400, plus 40 darks of the same in DSS  takes less time than boiling the kettle  

The thing is though, all the benchmarking in the world is a tad irrelevant as processing images will vary from one user to another.  Benchmarking games is one thing as the game is standardised and thus can be used to compare performance.  Also, whilst it's nice having lots of horsepower, you often will never use it to its full potential as most of the time we tend to use our PCs for browsing and sending e-mail

  

Edited December 20, 2020 by malc-c
spellings

[licho52]

9 hours ago, malc-c said:

 

Benchmarks are very relevant IMO.

Most picture processing (outside of some PS procedures) is multithreaded and thus multicore CPUs are a great upgrade for astrophotography work (big bang for the buck).  Disk drive speed is only relevant up to some point and buying overpriced Samsung NVME is paying twice for miniscule differences.  Also graphics cards are becoming relevant as nVidia CUDA units are being used by StarNet.

Edited December 21, 2020 by licho52

[malc-c]

2 hours ago, licho52 said:

Benchmarks are very relevant IMO.

Most picture processing (outside of some PS procedures) is multithreaded and thus multicore CPUs are a great upgrade for astrophotography work (big bang for the buck).  Disk drive speed is only relevant up to some point and buying overpriced Samsung NVME is paying twice for miniscule differences.  Also graphics cards are becoming relevant as nVidia CUDA units are being used by StarNet.

The point I was trying to make was that there is no standard by which to benchmark.  An image of the orion nebula may contain more data than say a globular cluster, an image of the same object taken on a dedicated astro camera may have more data or more pixels than a Canon 400d.... The ONLY way to benchmark a system would be to process the same image using the same software and software settings.  I agree that for any application that is written to use multiple cores, a processor with a high core count will give more performance, but equally the various levels of cache on the processor can make a difference - it all depends on how the application is written.  Graphics cards IMO are not that critical, mainly because the final result is displayed after processing, unlike a game where refresh rates and fps are the key factors 

[kirkster501]

I have a reasonably modern, fast machine with i7-8700K (12 cores, hyper-threaded) with 32G of fast memory and Samsung EVO 970 and a NVidia 1060 GC.

Stacking 150 x 6000x4000 resolution images from my QHY268C camera in PixInsight can take 45 minutes plus.

[malc-c]

22 minutes ago, kirkster501 said:

I have a reasonably modern, fast machine with i7-8700K (12 cores, hyper-threaded) with 32G of fast memory and Samsung EVO 970 and a NVidia 1060 GC.

Stacking 150 x 6000x4000 resolution images from my QHY268C camera in PixInsight can take 45 minutes plus.

Just out of interest, if you stacked those same 150 images in deep sky stacker would it take longer or be quicker ?, thus showing that it's how the application is written to take advantage of multiple cores, or hyperthreading

[kirkster501]

28 minutes ago, malc-c said:

Just out of interest, if you stacked those same 150 images in deep sky stacker would it take longer or be quicker ?, thus showing that it's how the application is written to take advantage of multiple cores, or hyperthreading

I am not sure.  I will try it.  PI does not use graphic card acceleration and is run on the CPU.  I know you can get Starnet to use the CUDA on the GC but I am not sure other PI processes can use that.

[licho52]

Bottom line we're in a very CPU intensive area which might start getting also coded for use of graphics card engines.  As long as the apps are multithreaded (AS3, DSS and PI are) then getting as many threads as possible is the best way to go.  If price were no object I'd get an AMD Threadripper with huge core counts.  At the same time I am getting no material over the last month and a half so it would just sit there anyway...

Edited December 21, 2020 by licho52

[Starflyer]

One thing that makes a big difference in PI is to use a ram disk as your scratch / swap space.

I have a Ryzen 7, 32GB of fast RAM and a fast M2 drive. Loading a 16GB ram disk at startup and using that in PI / PS makes it blisteringly fast.

Search through the PI benchmark site for your old Vs proposed system and it'll give you an idea on performance gains you can expect.

[rnobleeddy]

I haven't read all of this but my take is:

- most processing is CPU limited so you can just rely on the CPU benchmarks, so CPU benchmarks absolutely matter

- for most people, an upgrade won't yield massive increases in single core performance (but 20% single core performance will help!) so check if your software is multi-threaded.

- I use DSS and Startools and they're happy to use all the cores, with Startools also running well with a GPU

- Regardless of whether it helps overall processing time, getting the fastest storage you can will make for a happy life! 

[mikeyj1]

That's my conclusion alsornobleeddy, although most benchmarks are for gaming, but some, like userbenchmark.com rate them for the 'workstation' also, which is what we need for the number-crunching.  here's the difference with a Ryzen 5 3600 against my processor.  I will also do my own benchmark -  processing a couple of images in DSS and APP and PS, and then repeat them when I've  done the upgrade to give an indication of the improvement in time.  

Edit: just saw the PI benchmark link from @Starflyer (Thanks, I don't use PI so would never have found this!), and it shows my 133 secs reducing to 38-60 secs on the same comparison Ryzen5 3600 mentioned 

thanks for the inputs all 👍 🙂

Merry Christmas!

Mike

Edited December 22, 2020 by mikeyj1
updated info

[prusling]

13 hours ago, rnobleeddy said:

I haven't read all of this but my take is:

- most processing is CPU limited so you can just rely on the CPU benchmarks, so CPU benchmarks absolutely matter

- for most people, an upgrade won't yield massive increases in single core performance (but 20% single core performance will help!) so check if your software is multi-threaded.

- I use DSS and Startools and they're happy to use all the cores, with Startools also running well with a GPU

- Regardless of whether it helps overall processing time, getting the fastest storage you can will make for a happy life! 

+1 for StarTools (latest version 7) with a GPU - I have a reasonably-priced Ryzen 4600 equipped laptop (12 cores) with a GeForce GTX 1650 (14 compute units) and it absolutely flies!

[Shimrod]

On 21/12/2020 at 21:04, Starflyer said:

One thing that makes a big difference in PI is to use a ram disk as your scratch / swap space.

I have a Ryzen 7, 32GB of fast RAM and a fast M2 drive. Loading a 16GB ram disk at startup and using that in PI / PS makes it blisteringly fast.

Search through the PI benchmark site for your old Vs proposed system and it'll give you an idea on performance gains you can expect.

My observation from those benchmarks is that disk transfer speeds play a much bigger part in the performance of that application than overall cores. Looking at the benchmarks of the 3900x processor they range from 16 to 86 seconds, with the Transfer rate (assuming that is the disk performance) being the key factor.

[mikeyj1]

Hello All, 

So I thought i would give you an update now that i have finished the upgrade of the PC.  If you just want the bottom line without reading any further,  I reduced the processing time by 33% on DSS and over 44% on APP (without overclocking the new processor)!  

I opted for a Ryzen 5 3600 and a Gigabyte Motherboard with PCI4.0 support.  I also installed an M.2 NAND drive for the OS, and 32 GB of DDR4 3200 RAM.  This processor is somewhat less than I had planned, as the money saving was considerable over the Intel i5-10700k originally discussed,. However the Mobo will take the latest Gen Ryzen 9, so i have plenty of upgrade options in the future.

The benchmark i constructed was based on Deep Sky Stacker 4.2.5 (64bit) and Astro Pixel Processor 1.082. 

My data was 62 images of NGC3000 a master dark and a master bias, (No flats or original dark files as i couldnt find the originals!).  Data sets were copied to separate directories for both the original and upgraded system, on a separate internal 3TB SATA drive.  PC was restarted between each run to help clear cache.  2 runs of each setup.  These were pretty light benchmarks, but probably typical for general imaging.  It seems the more intensive the processing the more time will be saved...see below!

DSS:  Default settings, 100% image stack, star selection limited to ~60 on both runs. Run until full draw of final picture.  62 CANON CR2 files (Lights).  No calibration frames (The master dark and Bias were 1 pixel different, and DSS wouldnt accept them). 

APP: Default settings and 100% image stack using full cores and threads and 16GB RAM (more on that later!) Run until full draw of final picture.  Same light files+master dark and Bias to process.

For Ref: Original system:  i5-3570k overclocked to 4.1GHz with 16GB RAM and Raid-0 SSD array.  Windows 10 64bit Pro current build version 19042

Times in minutes and seconds to process:

i5 system DSS:  5'43" 

Ryzen 3600 system DSS:  3'50"  (-33% approx)

i5 system APP:  24' 35"

Ryzen 3600 system APP:  14'34" (I realised i had left the deafult 2GB RAM setting unchanged, so re-ran the test as follows):

Ryzen 3600 system APP:  13'56" (images and working drive changed to M.2 Super fast drive, 16GB RAM Setting)

Ryzen 3600 system APP:  13'35" (images and working drive changed to SSD Raid drive, Progs and OS on M.2 drive.  I suspect this gives the best result as ther is less 'bottlenecking') (-44.7% appox)

These are signifiant improvements, and I'm really pleased with the upgrade.  In addition, PC start up time is around 7 secs to password on normal boot option, and games and everything else run maxed out without issue.  I have only tried Photoshop once, so can't comment on this yet.  I also managed to sell my old Mobo and processor/RAM for a good price, so really happy

thanks for all your comments and  suggestions!

regards

Mike

Edited January 6, 2021 by mikeyj1
typos