Images generated by RastaConverter

[GravityWorm]

On 9/7/2023 at 8:30 PM, Atari8man2004 said:

Being a noob I have 175 message in this thread to catchup on here are a few more conversions.  I downloaded GIMP but would require lots of time to learn the program.

 

Everyone was once a noob

[invisible kid]

Re: Gimp. There is also MyPaint, but it might not be any easier to deal with.

[Atari8man2004]

I have been letting Raster picture converter work to completion and it's a timely process days at best.  The pc I'm using has 8 cores and I will use my other Pc that has 12 core later this week.  I'm wonding would a processor that has AVX512 speed up the process ?  I'm starting to get the hang of it tho   and having a blast !

[+MrFish]

5 hours ago, Atari8man2004 said:

...it's a timely process days at best.

 

I processed this image for an estimated continuous 6 to 8 months (24-hours a day). On an old machine, but still.

 

Green Acres

 

[Atari8man2004]

5 hours ago, MrFish said:

 

I processed this image for an estimated continuous 6 to 8 months (24-hour a day). On an old machine, but still.

 

Green Acres

 

WOW you where rewarded with a awesome conversion also you have great patience green Acres is awesome. Raster converter when completed will no longer load the source file so basically it is done.  I have 5 pictures now that have been coming along nicely "fingers crossed"  I have been doing just still pictures I have not done any landscapes.

Edited September 21, 2023 by Atari8man2004

[+MrFish]

4 minutes ago, Atari8man2004 said:

WOW you where rewarded with a awesome conversion also you have great patience green Acres is awesome.

It worked out well. A lot of details resolved by giving it so much time.

 

4 minutes ago, Atari8man2004 said:

Raster converter when completed will no longer load the source file so basically it is done.

Not sure exactly what you mean here.

 

[Atari8man2004]

1 hour ago, MrFish said:

It worked out well. A lot of details resolved by giving it so much time.

 

Not sure exactly what you mean here.

 

I had several conversion where Raster stops the conversion and will not let me continue even if I try to reload the source it just closes the program.

[+MrFish]

11 minutes ago, Atari8man2004 said:

I had several conversion where Raster stops the conversion and will not let me continue even if I try to reload the source it just closes the program.

I've never had that happen before. I did have it, one time, not continue as it was supposed, but it just started over when tried to continued. Other than that, continue always worked for me.

 

You should make sure you're using the latest versions of RastaConverter and RastaConverter GUI.

 

[Atari8man2004]

1 hour ago, MrFish said:

I've never had that happen before. I did have it, one time, not continue as it was supposed, but it just started over when tried to continued. Other than that, continue always worked for me.

 

You should make sure you're using the latest versions of RastaConverter and RastaConverter GUI.

 

I'm experimenting with the options to see what would be the best for a certain picture lots of trial and error what should "number of solutions" be set at ? the default is one.  Anyways like you said getting the results as close to the destanation is the botton line and I fully understand.  

[+MrFish]

1 hour ago, Atari8man2004 said:

what should "number of solutions" be set at ?

For me, it depends on complexity of the image and how long I plan to run it. The more complex and and longer I plan to run it, I set it higher.

 

[GravityWorm]

1 hour ago, MrFish said:

For me, it depends on complexity of the image and how long I plan to run it. The more complex and and longer I plan to run it, I set it higher.

 

I think It depends how powerful PC you have...

[GravityWorm]

Testing if it's possible to upload GIfs here...
GravityWormBS01.gif
 

[GravityWorm]

XEX files:

GravityWormBS_01.xex GravityWormBS_02.xex GravityWormBS_03.xex

[Rybags]

Has anyone created an "RC animator" yet?

16K frames would probably allow a narrow DMA mode with 100, maybe more scanlines with the remainder to do the kernal.

Though from looking at some RC binaries a fair while back, there is a fair bit of wasted space in there with long strings of NOPs that could be reduced by using other instructions for same cycle consumption but less memory.

[drpeter]

12 hours ago, Atari8man2004 said:

what should "number of solutions" be set at

Fewer solutions will lead to an image that resolves and stabilises more quickly, but at a level of accuracy that, on average, is less than one run for much longer with more solutions.

 

For a given run length (say 24 hrs) there is an optimal number of solutions that will, on average, produce the most accurate result (as the algorithm sees it) after that length of time.

 

For short runs, that will be a lower number of solutions.  For long runs, that will be a higher number of solutions.

 

All runs will tend to a point after a period of time where further improvement is very slow and often virtually invisible to the naked eye.

 

Added to that, because of randomness in the algorithm, repeated runs with identical parameters may tend to slightly better or worse results in each case, both visually and as the algorithm sees it.

 

So there's a case for starting several runs (say 4-5) each with a fairly high number of solutions and, once they are beginning to clearly resolve, picking the visually best of them and continuing to run that one for a much longer time until it ceases to visually improve.

 

My strategy based on this and the speed of the PC I'm using (a fairly slow 10yr old 4-core laptop) is to run 4 conversions with between 4000 and 16000 solutions (on a sliding scale depending on the image complexity and how much time I want to invest in it) for long enough to pick a 'winner', then continue that one until it stops significantly improving.

 

Doing this has taught me that you need to run a conversion with 16000 solutions for a VERY long time before it generally overtakes a 4000 solution conversion run for a similar length of time- roughly, doubling the number of solutions means the conversion will take 4 times as long to reach the same level of accuracy, so a 16000 solution conversion will take roughly 16x as long to reach the same point of accuracy as where the 4000 solution conversion has stopped improving.  The difference is that the 16000 solution conversion will continue to slowly improve for a long time beyond that point, although the discernible visual difference may not be all that much.

[Atari8man2004]

3 hours ago, drpeter said:

Fewer solutions will lead to an image that resolves and stabilises more quickly, but at a level of accuracy that, on average, is less than one run for much longer with more solutions.

 

For a given run length (say 24 hrs) there is an optimal number of solutions that will, on average, produce the most accurate result (as the algorithm sees it) after that length of time.

 

For short runs, that will be a lower number of solutions.  For long runs, that will be a higher number of solutions.

 

All runs will tend to a point after a period of time where further improvement is very slow and often virtually invisible to the naked eye.

 

Added to that, because of randomness in the algorithm, repeated runs with identical parameters may tend to slightly better or worse results in each case, both visually and as the algorithm sees it.

 

So there's a case for starting several runs (say 4-5) each with a fairly high number of solutions and, once they are beginning to clearly resolve, picking the visually best of them and continuing to run that one for a much longer time until it ceases to visually improve.

 

My strategy based on this and the speed of the PC I'm using (a fairly slow 10yr old 4-core laptop) is to run 4 conversions with between 4000 and 16000 solutions (on a sliding scale depending on the image complexity and how much time I want to invest in it) for long enough to pick a 'winner', then continue that one until it stops significantly improving.

 

Doing this has taught me that you need to run a conversion with 16000 solutions for a VERY long time before it generally overtakes a 4000 solution conversion run for a similar length of time- roughly, doubling the number of solutions means the conversion will take 4 times as long to reach the same level of accuracy, so a 16000 solution conversion will take roughly 16x as long to reach the same point of accuracy as where the 4000 solution conversion has stopped improving.  The difference is that the 16000 solution conversion will continue to slowly improve for a long time beyond that point, although the discernible visual difference may not be all that much.

Thank you for passing this information to me.   I now have a better understanding on how to aproch a conversion process Thanks!

Edited September 21, 2023 by Atari8man2004

[Atari8man2004]

12 hours ago, GravityWorm said:

I think It depends how powerful PC you have...

I have many pc the one I'm using is 8 cores 16 threads at 4ghz . I been converting 4 or 5 images at a time and it puts a load on the processor almost like crypto mining at temps of 70c.   I was wondering if a processor with AVX512 would shorten the converion time. 

[pseudografx]

For some reason, RastaConverter runs at above average priority which affects response of some systems rather substantially. When I start a conversion, I always set the RastaConverter thread's priority to lowest via Task Manager. It has no effect on the conversion speed but system's response is not affected any more.

Edited September 21, 2023 by pseudografx

[+MrFish]

3 hours ago, pseudografx said:

For some reason, RastaConverter runs at above average priority which affects response of some systems rather substantially. When I start a conversion, I always set the RastaConverter thread's priority to lowest via Task Manager. It has no effect on the conversion speed but system's response is not affected any more.

I always set RastaConverter to run in less threads, if I know I'll need the bandwidth. So, on a 4 core machine, I just set it to 3 threads, and then I've got plenty of bandwidth for most things I want to do (1 core free).

 

    

 

    

 

[Rybags]

Modern instruction extensions would probably make no difference unless the program is recompiled and can take advantage of them.

With the CPU temps, multiple jobs running total thread count of all jobs to be -1 of physical or logical processors would give good performance but having a round-robin core idle state would knock the temperature down somewhat.

[+Stephen]

No way man - I set that bitch on highest priority, max number of threads, max solutions, and let her rip for as long as it takes!  If my machine is not screaming for mercy, I am not using it properly!

[Atari8man2004]

34 minutes ago, Stephen said:

No way man - I set that bitch on highest priority, max number of threads, max solutions, and let her rip for as long as it takes!  If my machine is not screaming for mercy, I am not using it properly!

lol    Put the pedal to the metal!  

Edited September 22, 2023 by Atari8man2004

[Atari8man2004]

4 hours ago, MrFish said:

I always set RastaConverter to run in less threads, if I know I'll need the bandwidth. So, on a 4 core machine, I just set it to 3 threads, and then I've got plenty of bandwidth for most things I want to do (1 core free).

 

    

 

    

 

I'm converting 4 images you can see in the graph the cores.  When these are done I'm switching to a better system 12 cores 24 threads at 4.8ghz  I'm also going to set Raster at 4000 solutions . I been doing these conversion at (1) solution  no wonder they where done in just 4 days 

[Atari8man2004]

2 hours ago, Rybags said:

Modern instruction extensions would probably make no difference unless the program is recompiled and can take advantage of them.

With the CPU temps, multiple jobs running total thread count of all jobs to be -1 of physical or logical processors would give good performance but having a round-robin core idle state would knock the temperature down somewhat.

It would be interesting if Raster the author would move CPU computation to the GPU in the future.  The video cards we have these days have thousands of core that would shorten conversion time greatly.

Edited September 22, 2023 by Atari8man2004

[Beeblebrox]

2 hours ago, Atari8man2004 said:

It would be interesting if Raster the author would move CPU computation to the GPU in the future.  The video cards we have these days have thousands of core that would shorten conversion time greatly.

This was/is potentially on the cards. Or at least it had been a possibility. Not from the author of RC who has said many times in the past he'll not be revisiting RC. Rather from another famous A8 Coder (not naming names as don't wanna put any pressure in him) who was potentially going to look into creating a GPU based RC engine.