16 replies to this topic

[pyrasanth]

Just over a month ago my main processing computer died with a hardware failure. This was somewhat inconvenient when my typical image integration is about 3000 plus full frame subs so needs some pretty beefy hardware. I had a replacement board shipped from China but that failed as well after 3 days- so that generated another delay of a week to get a replacement. A lot of stuff that comes from China seems to fail early especially the refurbished board that I originally purchased. I went for new for the replacement which is now up & running.

 

I am relieved to say that having rebuilt the failed computer and added an additional 256GB of RAM I'm backup & running & ready to crunch those big data sets-

 

A screen grab of all the cores & memory is below. I'm planning to increase the RAM to 1TB and change the CPU's to 64/128 thread in the next few months- so should make a PI processing monster. The system now is pretty quick as it stands but PI loves cores & memory so I'm going to give it what it desires. The CPU's currently are fairy modest at present 2x EPYC 7601 32/64 core/threads supported with 512GB of RAM- 4TB spanned NVME hardware controller and 24TB of WD mechanical hard drives but connected to 138TB Synology DS918 using NFS Shares. PI has its own SSD swap drive and the Linux setup is backed up using time shift to its own SSD as well. The O/S flavour is Linux mint 22.x

 

I wish you all clear skies.

 

Edited by pyrasanth, 29 April 2025 - 03:01 PM.

[amikep1]

WOW, can you be more specific on your hardware?
Motherboard
Processor

Cost??

 

Thanks

[pyrasanth]

WOW, can you be more specific on your hardware?
Motherboard
Processor

Cost??

 

Thanks

This is the motherboard https://www.ebay.co....tm/175902906748 and the processors x 2 https://www.ebay.co.uk/itm/126425138754

 

I then added 16 memory modules DDR4 32 GB

 

Cooling uses the Noctua TR3 x 2 https://www.ebay.co....tm/196058591636

 

You then need to look at a beefy PSU I used this https://www.corsair....y-cp-9020261-uk

 

There of course are all the additional drives and a good case which will easily add another £1200.00

 

I don't think you will get much change out of £4000.00 for the basic build once you've added a good graphics card which helps with the PI processing especially if you want to use CUDA cores to speed up the Xterminator modules.

 

When I upgrade the CPU's shortly they will cost around £500.00 each.

 

So- not a cheap bit of kit. If you want to go really modern you could build an AMD threadripper but that also can quickly get into serious money.

 

I will post a photograph tomorrow so you can have a look at the complete built server.

Edited by pyrasanth, 29 April 2025 - 04:39 PM.

[Astro-Goat]

WOW, can you be more specific on your hardware?
Motherboard
Processor

Cost??

 

Thanks

Bro, I know right?!!! LOL!!

 

That's like me saying "I have a computer that can run the latest Doom game at sustained 120fps, with DLSS set on, and textures set to Ultra".

 

 

 

People "ummmm, okkkaaaay. So what's your hardware?"
me "dunno, i bought the most expensive PC on Amazon and hoped for the best"

[Astro-Goat]

Just over a month ago my main processing computer died with a hardware failure. This was somewhat inconvenient when my typical image integration is about 3000 plus full frame subs so needs some pretty beefy hardware. I had a replacement board shipped from China but that failed as well after 3 days- so that generated another delay of a week to get a replacement. A lot of stuff that comes from China seems to fail early especially the refurbished board that I originally purchased. I went for new for the replacement which is now up & running.

 

I am relieved to say that having rebuilt the failed computer and added an additional 256GB of RAM I'm backup & running & ready to crunch those big data sets-

 

A screen grab of all the cores & memory is below. I'm planning to increase the RAM to 1TB and change the CPU's to 64/128 thread in the next few months- so should make a PI processing monster. The system now is pretty quick as it stands but PI loves cores & memory so I'm going to give it what it desires. The CPU's currently are fairy modest at present 2x EPYC 7601 32/64 core/threads supported with 512GB of RAM- 4TB spanned NVME hardware controller and 24TB of WD mechanical hard drives but connected to 138TB Synology DS918 using NFS Shares. PI has its own SSD swap drive and the Linux setup is backed up using time shift to its own SSD as well. The O/S flavour is Linux mint 22.x

 

I wish you all clear skies.

 

LInux backup & running.jpg

Bro what is your computer rig?

I use Pixinsight on a 2015 Intel i5, 128gb system ram, 800w power supply, a micro gaming motherboard Asus, and an Nvidia 2080.

 

What do you need all that horse power for other then flexing?

[Astro-Goat]

This is the motherboard https://www.ebay.co....tm/175902906748 and the processors x 2 https://www.ebay.co.uk/itm/126425138754

 

I then added 16 memory modules DDR4 32 GB

 

Cooling uses the Noctua TR3 x 2 https://www.ebay.co....tm/196058591636

 

You then need to look at a beefy PSU I used this https://www.corsair....y-cp-9020261-uk

 

There of course are all the additional drives and a good case which will easily add another £1200.00

 

I don't think you will get much change out of £4000.00 for the basic build once you've added a good graphics card which helps with the PI processing especially if you want to use CUDA cores to speed up the Xterminator modules.

 

When I upgrade the CPU's shortly they will cost around £500.00 each.

 

So- not a cheap bit of kit. If you want to go really modern you could build an AMD threadripper but that also can quickly get into serious money.

 

I will post a photograph tomorrow so you can have a look at the complete built server.

Bro i just looked at these specs. Out of curiosity, what do you need a 1200W. power supply for?  Are you running like 2 video cards in a SLI configuration (i dunno if they even do SLI).
Would LOVE to see the full build whenever you post a picture.

[pyrasanth]

Bro what is your computer rig?

I use Pixinsight on a 2015 Intel i5, 128gb system ram, 800w power supply, a micro gaming motherboard Asus, and an Nvidia 2080.

 

What do you need all that horse power for other then flexing?

This is not a desktop computer- it is a server. When the CPU's are upgraded in a couple of months those CPU's will pull 600 watts alone before any other component. I process an immense number of subs- my integrations reach TB's of data with typically 5.000 subs going into WBPP- try & crunch that data on anything other than a very powerful HEDT or server- I'm flexing nothing- that is needed to process the amount of data that I handle. The PC you use, as nice as it is, would not handle the amount of data in a time that you'd happy with if 5000 full frame subs were needed to be processed. My office PC which sometimes picks up a few post integration tasks is a Ryzen 5950, 128GB RAM and a GTX3080

 

Bro i just looked at these specs. Out of curiosity, what do you need a 1200W. power supply for?  Are you running like 2 video cards in a SLI configuration (i dunno if they even do SLI).
Would LOVE to see the full build whenever you post a picture.

No I'm not running an SLI gaming rig- this is a high end server and when you power the CPU's I'm proposing to upgrade to they will take 600 watts alone before I've powered any other components- I like to run my PSU's at no more than 85% utilisation so 1500 watts is about right for the config I'm using.

Edited by pyrasanth, 29 April 2025 - 06:49 PM.

[csrlice12]

Yea, but can it run Johnny Castaway?

[pyrasanth]

Yea, but can it run Johnny Castaway?

This is a sensible post- I guess I could quickly build my number of posts to the same as you by posting irrelevant garbage.

[bluesteel]

How much RAM do you eat up when processing a large dataset?

[pyrasanth]

How much RAM do you eat up when processing a large dataset?

This is a good question. When running Linux Mint WBPP will crash when running large data sets with an out of memory error. I don't think it works as well as it could with regard to memory management. The process through registration/normalisation works fine. I finish WBPP at those stages then integrate manually where I can set memory management from auto to custom controlled thus I can chunk the integration of the data. If I set the cache setting to 8192/16384 an integration of 1400 OSC subs (APSC size) will run to about 300GB of RAM being used. This is why I plan to take the RAM to 1TB which means using those settings manually set I should be able to integrate around 5000 subs without changing those settings.

[ayadai]

3000 plus full frame subs

Just curious: are you imaging in the UK? If so, where do you find enough clear nights to amass that volume of data?

[pyrasanth]

Just curious: are you imaging in the UK? If so, where do you find enough clear nights to amass that volume of data?

Yes- always in the UK imaging. I never complete my integrations in short time frames- each target can take me months or years to get the data sets I need. I've started to use dual telescopes to leverage the clear nights I have.

 

LRGB data on my telescopes never need longer than 60 second subs under my light sky conditions. I have many collections of data for a target- I just append more subs to those data sets thus growing the target's data over time.

 

Rome was not built in a day- never more true in this hobby.

Edited by pyrasanth, 03 May 2025 - 10:44 AM.

[ayadai]

Yes- always in the UK imaging. I never complete my integrations in short time frames- each target can take me months or years to get the data sets I need. I've started to use dual telescopes to leverage the clear nights I have.

I'd imagine you don't make changes to the imaging train often, if at all. How frequently do you find you need to shoot calibration frames?
 

[pyrasanth]

I'd imagine you don't make changes to the imaging train often, if at all. How frequently do you find you need to shoot calibration frames?
 

I renew darks/bias every six-nine months. Flats with the Altair 115 are taken after every session as a precaution, they might not be used, as automation with the DeepskyDad automated flat panel makes them very easy to take. The RASA flats only when the camera is removed- I might do a very long OSC session that may last months then swap to a mono which would need a change of flats. 

[joshman]

Out of interest, how long does it take WBPP to crunch through 5,000 subs?

 

Is that 5k subs from the 6200 or the 2600 camera?

 

The most I've ever done is 3,700 subs from my 2600MM camera, and that took ~8 hrs at drizzle 2x with my much more modest (and affordable?) 7950X and 64GB RAM.

[pyrasanth]

Out of interest, how long does it take WBPP to crunch through 5,000 subs?

 

Is that 5k subs from the 6200 or the 2600 camera?

 

The most I've ever done is 3,700 subs from my 2600MM camera, and that took ~8 hrs at drizzle 2x with my much more modest (and affordable?) 7950X and 64GB RAM.

The speed of the drizzle depends on the size of the sub and of course whether you use the full frame. I seem to remember that the last OSC integration from the ASI2600 being drizzled to x2 from a region of interest which was about 70% of the full frame completed 1400 subs in under an hour- so the drizzling is quite fast.