Jump to content

Narrowband

How many processing software packages make use of a GPU?


Recommended Posts

I use PS and using the onboard Intel graphics or the Geoforce GTX doesn't appear to make a lot of difference to astro'  image processing, maybe faster for video editing I guess.

The i7 processor might be helping.

Dave

  • Thanks 1
Link to comment
Share on other sites

Not sure about those software packages mentioned but Affinity Photo have incorporated GPU processing for Mac only at the moment that gives a X10 speed increase with some actions..

With the rise in larger format CMOS in use and the big file sizes I cant imagine that the use of GPU processing will be overlooked for long, for now SSD, lots of fast RAM and a decent processor are your friend but worth having a dedicated GPU anyway for other stuff like VR etc.

Alan

  • Thanks 1
Link to comment
Share on other sites

Just been looking on the net, here is a quote from Ivo on the StarTools forum:
 

Quote

As of version 1.7 supports GPU acceleration in all modules, with many parts of the signal path rewritten for GPU execution (reduced branching, fewer memory transfers/access).

As ST is my current software of choice, looks like I should include one in the spec.

  • Like 1
Link to comment
Share on other sites

The StarNet process in PI (for star removal or creation of star masks) can configured to use nVidia GPUs, reportedly making a significant improvement is speed - see https://pixinsight.com/forum/index.php?threads/how-to-speed-up-starnet-as-pi-module-on-nvidia-gpus-with-cuda-on-windows.14882/

Edited by fireballxl5
  • Like 2
Link to comment
Share on other sites

Astropixelprocessor also supports GPU's now too. If you only have a board with integrated graphics (which most probably do) then it might not make a big difference. But any sort of half decent discrete graphics card should make a noticeable difference. I have a Gtx 780ti in my home PC and even though it's now old tech by today's standards, it still rips through Starnet, App and PS. 

  • Thanks 1
Link to comment
Share on other sites

It's worth noting that CUDA is an NVidia proprietary/only technology, so anything that specifically requires CUDA will not run on AMD cards.

Writing and particularly optimizing for GPUs has been an incredibly interesting experience. Much of what you know about optimisation for general purpose computing does not apply, while new considerations come to the fore.

Some things just don't work that well on GPUs (e.g. anything that relies heavily on logic or branching). E.g. a simple general purpose median filter shows disappointing performance (some special cases not withstanding), whereas complex noise evolution estimation throughout a processing chain flies!

I was particularly blown away with how incredibly fast deconvolution becomes when using the GPU; convolution and regularisation thereof is where GPUs undeniably shine. My jaw dropped when I saw previews update in real-time on a 2080 Super Mobile!

I don't think APP uses the GPU for offloading arithmetic yet by the way. Full GPU acceleration for AP is all still rather new. GPU proliferation (whether discrete or integrated in the CPU) has just about become mainstream and mature. Hence giving it another look for StarTools. Exciting times ahead! 👍

  • Like 7
Link to comment
Share on other sites

On 20/09/2020 at 05:22, jager945 said:

It's worth noting that CUDA is an NVidia proprietary/only technology, so anything that specifically requires CUDA will not run on AMD cards.

Writing and particularly optimizing for GPUs has been an incredibly interesting experience. Much of what you know about optimisation for general purpose computing does not apply, while new considerations come to the fore.

Some things just don't work that well on GPUs (e.g. anything that relies heavily on logic or branching). E.g. a simple general purpose median filter shows disappointing performance (some special cases not withstanding), whereas complex noise evolution estimation throughout a processing chain flies!

I was particularly blown away with how incredibly fast deconvolution becomes when using the GPU; convolution and regularisation thereof is where GPUs undeniably shine. My jaw dropped when I saw previews update in real-time on a 2080 Super Mobile!

I don't think APP uses the GPU for offloading arithmetic yet by the way. Full GPU acceleration for AP is all still rather new. GPU proliferation (whether discrete or integrated in the CPU) has just about become mainstream and mature. Hence giving it another look for StarTools. Exciting times ahead! 👍

You're right Ivo, APP hasn't yet added GPU support. I could have sworn i read that it had, i guess i got it mixed up with something else. 

Thanks for putting the record straight! 👍

Link to comment
Share on other sites

41 minutes ago, jager945 said:

For anyone interested, this is currently a very cheap solution (~40 GBP) to get your hands on some pretty decent compute performance, if you are currently making do with an iGPU or older (or budget) GPU.

Excellent cheap solution Ivo I’m assuming you could add a couple of these together on a system to create a cheap powerful Astro processing dedicated machine space allowing.

Link to comment
Share on other sites

45 minutes ago, bottletopburly said:

Excellent cheap solution Ivo I’m assuming you could add a couple of these together on a system to create a cheap powerful Astro processing dedicated machine space allowing.

Actually, most applications do not make use of multiple cards in your system (ST doesn't). That's because farming out different tasks to multiple cards is a headache and can cause significant overhead in itself (all cards need to receive their own copy of the "problem" to work on from the CPU). It may be worth investigating for some specific problems/algorithms, but generally, things don't scale that well across different cards.

Edited by jager945
  • Like 1
Link to comment
Share on other sites

  • 4 weeks later...

Interesting, this which I've just seen.

When I specc'd my current (Old) tower I had two cards, as I was running DaVinci Resolve, which needed (Possibly still does, haven't checked) two cards, one for the display (I used a Quadro 600) and one for the heavy video stuff, (A GeForce 570, I think).

Wondering of Star Tools (Or any other package) has caught up with Blackmagic yet.

Link to comment
Share on other sites

On 23/09/2020 at 04:57, jager945 said:

For anyone interested, this is currently a very cheap solution (~40 GBP) to get your hands on some pretty decent compute performance, if you are currently making do with an iGPU or older (or budget) GPU.

That looks quite interesting as a way to try out some stuff I've been playing with.  At the moment I can't find a UK source at a delivered price that is close to £40 though :(

James

Link to comment
Share on other sites

As an add-on to the original question:

  • APP has promised GPU support to speed up processing. Currently, it only uses the GPU for (re)drawing the image. There is no pointer as to when real GPU support will be implemented.
  • PI has promised GPU support (since 2015). Currently it has none, but I found a reference from Juan (the developer) to "hopefully" before the end of this year.
  • StarTools has GPU support (1.7 alpha)
  • PS/Affinity Photo all have very limited GPU support as far as I know

CPU performance still is very significant if you want speedy processing across the board.

I have written elsewhere in this forum what GPU acceleration in StarTools does on my PC.  Still, when building or upgrading a PC, I would suggest to balance the expenditure:

  • A modern CPU with a lot of cores (AMD 3700x comes to mind, or 3600 if you want to keep it a bit cheaper) as APP and PI do make proper use of multiple cores
  • A decent GPU but not the latest and the greatest as they are a) very expensive and b) hard to get
  • 32 GB of RAM as a sweet spot
  • SSD for OS and SSD for image and swap space

 

Edited by Annehouw
  • Like 3
Link to comment
Share on other sites

On 21/10/2020 at 23:26, JamesF said:

That looks quite interesting as a way to try out some stuff I've been playing with.  At the moment I can't find a UK source at a delivered price that is close to £40 though :(

James

Honestly, something like a used RX 570 also offers very good value for money (though since it's an AMD offering, it will have no CUDA cores). A bit more expensive (as it's a general purpose GPU), but quite a bit more powerful for compute purposes. There are mining versions of these around as well.

Link to comment
Share on other sites

I haven't come across anything that requires more than a few seconds for most heavy things - multiple deconvolution examples, commercial photoshop actions, longest wait is pixinsight match and integration process which can takes minutes for a  couple hundred images (full spec 2018  macbook pro i9  with 32gb  ram and gpu). Using AI super resolution can take several seconds with a 300mb image

 

I have a  video /image/deep learning machine (my ex crypto rig) with 32  core threadripper and  2 AMD Vega (each  25Tflop in single precision mode) all water cooled (plus 2 spare gpu), 96GB RAM with a 1K high spec monitor  but I rarely use it since I find the macbook more convenient to use

I would get a decent cpu and plenty of RAM personally plus fast ssd card (as big as possible)

Edited by billhinge
changed to TFlop
Link to comment
Share on other sites

On 24/10/2020 at 14:41, billhinge said:

I haven't come across anything that requires more than a few seconds for most heavy things - multiple deconvolution examples, commercial photoshop actions, longest wait is pixinsight match and integration process which can takes minutes for a  couple hundred images (full spec 2018  macbook pro i9  with 32gb  ram and gpu). Using AI super resolution can take several seconds with a 300mb image

 

I have a  video /image/deep learning machine (my ex crypto rig) with 32  core threadripper and  2 AMD Vega (each  25Tflop in single precision mode) all water cooled (plus 2 spare gpu), 96GB RAM with a 1K high spec monitor  but I rarely use it since I find the macbook more convenient to use

I would get a decent cpu and plenty of RAM personally plus fast ssd card (as big as possible)

Holy smokes! That must be one beast of a machine.

A stacking run of 400 subs in APP can easily take 4 hours on my machine (i5-2500k / 32gb ram), 16MB subs. 

Link to comment
Share on other sites

I  built it to do Ethereum crypto mining 3 years ago when  ether was cheap < $30.  Fully loaded with all 4 Vega's it runs at over 100TFlops, what I made from crypto more than covered its cost. If you recall that was people were buying top end gpu's as though they were going out of fashion.  It does use a lot of electric though, over 1KW/h full tilt  running 24/7! I just re-purposed it later with the addition of a BenQ SW271 27 Inch 4K Photography Monitor

Its all running on  an open air crypto mining rig frame so no thought of aesthetics 🙂

Edited by billhinge
Link to comment
Share on other sites

  • 2 weeks later...

"Actually, I've been working on this for some time, but preferred to keep it silent until now. I already started exploring the possibility of adding optional GPU support to several critical parts of our PCL development framework a few months ago. As a starting point, a version of PCL (and hence of PixInsight) with OpenCL kernels implementing some important bottlenecks (such as inner convolution loops, statistics and linear algebra routines, etc) should be ready sometime this year.

As I already have said many times, however, the amount of pending work and ongoing projects that we have ahead is huge, and right now we have more important things to do than adding GPU support. So this is not at the top of the stack, but is undoubtedly important and I am taking it into account"

 

This is what Juan said back in April

  • Like 1
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • 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.