Question Low to Moderate Low AM4 Observatory Build (PC)

[MalVeauX]

Hey all,

My laptop gave up the ghost after I dripped fluids into its keyboard and things started to not work anymore, USB port here, keys there, etc. So time to replace it. This time with a desktop, so I can service it in the future instead of total overhaul like this when it happens again.

This PC lives in my observatory. It's duties are to run Windows (7) and it will power software that operates my observatory mount, planetarium software, video capture software and data capture software. None of it is particularly heavy stuff. The video capture is the heaviest stuff, and it will max out what's possible on USB3 bandwidth with my CMOS USB3 cameras. The rest is easy to run on virtually any hardware. I need fast USB bus though. It will be writing its RAW data feed to a 1TB SSD (Samsung). I was going to build around the latest AM4 3000 series APU's, but the Athlon went up in price a lot and/or is backordered. Either way, a newer motherboard is needed and RAM serves as the video buffer, so I'm happy to put 16Gb of DDR4 into it. That's just to help provide context. I already have the other things that will be needed (case, PSU, etc). So the goal is maybe an APU, 4c/8t class CPU with 16GB RAM for the above purposes.

I'm not sure what will work, as I'm familiar with the latest AMD stuff, but not the yesteryear stuff.

My goal is around $300 USD to rebuild the main parts of this (motherboard, CPU/APU, RAM).

Here's what I'm currently looking at, but I'm open to suggestions and pointers as long as we can stay in a similar price range.

AMD Ryzen 5 3400G (4C/8T APU) - $140
MSI ProSeries B450M PRO-M2 Max (B450MPM2MAX) - $75
TeamGroup T-Force Vulcan Z DDR4 16Gb (8x2) 3200Mhz CL16 Memory - $58

That's $273 currently. Will use stock cooler.

Any ideas to have any significant improvements for similar cost? Especially considering the above motherboard & memory? Up to the $300 USD mark roughly?

Thanks!

Very best,

 

[sdifox]

It was a laptop, which started to malfunction (keys stopped work, a usb port stopped working, I was dripping sweat into the keyboard, it was in there for a few years but it got problematic after I got it wet a few times). So I replaced it with some parts I put together (spare stuff laying around) and made a workable desktop and put it out there instead. The performance of the desktop however is lacking compared to the old laptop even. So I'm exploring just doing a rebuild using existing parts for the desktop (like PSU, case, SSD, etc) and just getting new MB, RAM and CPU.

Very best,

Ah so it was a Frankenstein to start with. It doesn't surprise me that it has bandwidth issue then if you added usb 3.0 card to a laptop mb.

 

[MalVeauX]

Ah so it was a Frankenstein to start with. It doesn't surprise me that it has bandwidth issue then if you added usb 3.0 card to a laptop mb.

It's a USB3 card added to a desktop motherboard that was only USB2 (via PCIe). The laptop is completely out of this at this point. I only harvested the SSD. So it's an older AM3/AM2+ desktop motherboard (model was posted above) and a PCIe USB3 card added. I don't need USB3 for the functions of the system at all, other than the speed of the CMOS cameras throughput when I'm imaging.

But yea I didn't expect to just fix the issue by adding the USB3 card to the old desktop. I figured it would be better than USB2 but not quite fully USB3 just because its an old system and wasn't designed around the bus having that throughput perhaps. I'm not sure. But since I had the parts, I put it together and it works. But it's just not fast. So, back to the drawing board which means spending money. So if I'm going to spend, might as well make something as good as I can for the budget.

Very best,

 

[sdifox]

It's a USB3 card added to a desktop motherboard that was only USB2 (via PCIe). The laptop is completely out of this at this point. I only harvested the SSD. So it's an older AM3/AM2+ desktop motherboard (model was posted above) and a PCIe USB3 card added. I don't need USB3 for the functions of the system at all, other than the speed of the CMOS cameras throughput when I'm imaging.

But yea I didn't expect to just fix the issue by adding the USB3 card to the old desktop. I figured it would be better than USB2 but not quite fully USB3 just because its an old system and wasn't designed around the bus having that throughput perhaps. I'm not sure. But since I had the parts, I put it together and it works. But it's just not fast. So, back to the drawing board which means spending money. So if I'm going to spend, might as well make something as good as I can for the budget.

Very best,

Ah, got it. Ok then try updating firmware to card and add the driver I mentioned. Or better yet install win 10. Maybe you don't need a new machine at all xd.

 

[MalVeauX]

Ah, got it. Ok then try updating firmware to card and add the driver I mentioned. Or better yet install win 10. Maybe you don't need a new machine at all xd.

Thanks, will check it out!

Very best,

 

[MalVeauX]

Update,

Decided to go fairly low on the CPU to test things; there's literally zero testing out there for my application, so I'm interested to at least test things out and report. If the platform doesn't perform, I'll try a higher end CPU to see if that handles it and report there too. I'm thinking though that in all reality the CPU isn't the biggest part of this system and it's more an architecture, instruction set, memory and bus speed for throughput for RAW data transfer and less so much the actual CPU's horsepower and core/thread count. But, we will see.

I picked up:

Athlon 3000G (2c/4t) APU ($50)
MSI B450M PRO-M2 Max motherboard ($85)
TeamGroup T-Force Vulcan Z DDR 16Gb Kit (2x8Gb) 3000Mhz CL16 ($52)

I will test the following CMOS cameras on the system for achieving the documented limit of the full resolution FPS potential via the throughput of the system to the writing medium (Samsung SSD):

ZWO ASI224MC (1.2Mp sensor, 150 FPS max full res) <-- I've never gotten the 150 FPS on past systems
ZWO ASI290MM (2.13Mp sensor, 170 FPS max full res) <-- I've never seen more than 85 FPS on this one on my past systems
ZWO ASI174MM (2.35Mp sensor, 164 FPS max full res) <-- This one has never seen 164 FPS on my past systems, so I'm interested in cranking this up
ZWO ASI183MM (20Mp sensor, 19 FPS max full res) <--- This one is the one to really test, it generates a ton of data fast

If the above CPU proves to be a bottleneck in the system, I'll try the Ryzen 2600 CPU ($140).

Very best,

 

[MalVeauX]

Update,

I have the system up and running and tested the cameras. I had to upgrade to Windows 10 of course, as this hardware doesn't support Windows 7 anymore. Thankfully all the software works the same in Windows 10 without any problems. This again is to look at my old hardware and old laptop moving forward to see what matters most to maintain maximum throughput of our typical planetary imaging CMOS cameras to have the maximum FPS at both the maximum resolution and the minimum resolution to see where the bottlenecks exist and what hardware influences what.

My previous Intel i3 CPU laptop with USB3 and SSD was not achieving maximum FPS on any of the cameras, but was faster than my oldest-older stuff.
My desktop AM3/AM2+ platform with DDR2 memory and a Quadcore CPU (Athlon II x4 600e AM3) was the slowest even with a PCIe USB3 card installs.

I recently picked up some modern hardware to really test what matters. The focus was on the system bus and the architecture between the devices, so the motherboard and memory lanes basically to allow for bandwidth to not be the limit and instead let the hardware (CMOS camera in this case) be the limit. I wanted to see if CPU mattered, so I got the lowest end desktop class APU I could that was modern (AM4, Zen Architecture). This is still not the latest hardware, but it's recent enough to be reasonable. I went with an AMD platform because of cost, while there are also reasonably low cost Intel platforms for this the motherboards were more costly, so I just went with what provided overall value for cost and went with an APU to avoid needing any sort of display output extra in the system (not that it matters, just adds to cost and power consumption).

$185 hardware I picked up, new:

MSI B450M PRO-M2 Max (AM4 motherboard, 450x chipset, USB3.1 Gen II)
TeamGroup T-Force Vulcan Z DDR4 16Gb (8x2) 3000Mhz CL16 RAM
AMD Athlon 3000G (Zen series APU, 2 Cores, 4 Hyperthreads, 35watt, 3.5Ghz)

The hardware already provided is the Samsung EVO 860 1TB SSD that the cameras will write to.

This is a very low power system in general, with humble gear. The goal again was to stress if the CPU mattered for acquisition and throughput of the CMOS cameras. I've used higher end CPUs and it didn't matter much. What does matter is the system bus and memory lanes and then finally the medium that is being written to, so ideally a SSD or NVMe M.2 SSD to avoid any bottlenecking and dropped frames.

Cameras tested (ZWO):

ASI224MC (IMX224)
ASI290MM (IMX290LQR)
ASI174MM (IMX174)
ASI183MM (IMX183)

I collected their documented throughput specs to compare my data to.

Capture Parameters and Method:

I used FireCapture (latest release). I did a fresh install with default values to avoid any custom stuff interfering. I left the preview at default. I enabled High Speed in the camera USB options and set the traffic to 100 on all cameras.

The test is 60 seconds capture time. Real capture, not preview. No gamma or any overlay used.

I captured at the maximum resolution of the sensor and the documented lower resolution of 320x240p on each sensor (as they listed specs down to this resolution).

Captures are 8/10 bit.

I let FireCapture give me the averages and information in log files.

++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Data and Observations:

ASI224MC (IMX224)

IMX224 Sensor Specs from ZWO

1304x976 150 fps
320x240 577.9 fps

Test Platform Throughput

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI224MC
Duration=59.985s
Frames captured=8961
File type=SER
ROI=1304x976
FPS (avg.)=149
Shutter=1.000ms
Gain=527 (87%)

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI224MC
Duration=60.012s
Frames captured=34338
File type=SER
ROI=320x240
FPS (avg.)=572
Shutter=1.000ms
Gain=527 (87%)

Results: Essentially achieving theoretical throughput at maximum resolution and a minute difference at lowest resolution, for a 0.0103% difference.

ASI290MM (IMX290LQR)

IMX290LQR Sensor Specs from ZWO

1936x1096 170 fps
320x240 737.5 fps

Test Platform Throughput

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI290MM
Duration=60.000s
Frames captured=10202
File type=SER
ROI=1936x1096
FPS (avg.)=170
Shutter=1.000ms
Gain=473 (78%)

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI290MM
Duration=60.000s
Frames captured=43665
File type=SER
ROI=320x240
FPS (avg.)=727
Shutter=1.000ms
Gain=473 (78%)

Results: Achieved maximum theoretical throughput with the maximum resolution, nearly achieved maximum on the lowest listed resolution (727 fps vs 737.5 fps) for a 0.0144% difference.

ASI274MM (IMX174)

IMX174 Sensor Specs from ZWO

1936x1216 164 fps
320x240 740 fps

Test Platform Throughput

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI174MM
Duration=60.007s
Frames captured=9814
File type=SER
ROI=1936x1216
FPS (avg.)=163
Shutter=1.000ms
Gain=377 (94%)

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI174MM
Duration=60.006s
Frames captured=43749
File type=SER
ROI=320x240
FPS (avg.)=729
Shutter=1.000ms
Gain=377 (94%)

Results: Theoretical throughput achieved at maximum resolution, 0.0061% difference and at the lowest resolution a difference of 0.0151%.

ASI183MM (IMX183)

IMX183 Sensor Specs from ZWO

5496x3672 19 fps
320x240 308.17 fps

Test Platform Throughput

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI183MM(ASI183MM)
Duration=60.006s
Frames captured=1142
File type=SER
ROI=5496x3672
FPS (avg.)=19
Shutter=1.000ms
Gain=398 (88%)

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI183MM(ASI183MM)
Duration=60.013s
Frames captured=18492
File type=SER
ROI=320x240
FPS (avg.)=308
Shutter=1.000ms
Gain=398 (88%)

Results: Theoretical throughput maintained at maximum resolution (100%) and at lowest resolution (308 vs 308.17, a 0.0005% difference likely only exists due to rounding in log file).

Observation conclusion, CPU has no impact on this throughput. The chipset, architecture, memory bandwidth, system bus and the write-to medium matter the most. I noted that maximum resolution throughput was achieved on all of the tests and that the only differences were found, even though they were all around the 0.01% range (completely negligible), it was only found when reducing to 320x240fps and increasing potential FPS in the 500~700 fps range. This may indicate averages because of more data collected over time, or maybe it indicates the small file write speed of the SSD is showing a tiny bit here. Either way, the differences are completely not noticeable and barely measurable, so I am happy with the current performance and don't see a reason to get fastest modern multi-core CPU components at the moment for acquisition purposes like this with the CMOS cameras. Modern architecture, system bus speed and memory speed and SSD write speed I think matter in this environment and purpose most.

++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++
++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Real Time Flat Frame Overlay

I threw in an extra test that would show a difference in throughput as the CPU would be involved (I think?). I often use real time flat frame overlay when doing some solar imaging at full FOV and this requires CPU performance and it slows down throughput because it takes the data stream, applies the overlay and then writes it to the write-to medium, so it adds another pathway and involves the CPU to do some minor work. This did show performance impact. Again, only theory in terms of how the CPU is involved.

I'm not sure how the flat frame overlay application works under the hood, maybe Emil could shine light on that to us if he's around (and thanks for your awesome software!!), so I'm only speculating that it's CPU intensive to do this (it may not be, it may be simply adding micro-seconds of time to each frame to avoid dropping frames, or maybe its dropped frames difference, etc).

I used the ASI290MM camera and a real time flat overlay with the following measurements and observations:

IMX290LQR Sensor Specs from ZWO

1936x1096 170 fps

IMX290LQR Sensor with Flat Overlay in FireCapture

FireCapture v2.6  Settings
------------------------------------
Camera=ZWO ASI290MM
Duration=60.015s
Frames captured=6284
File type=SER
ROI=1936x1096
FPS (avg.)=104
Shutter=1.000ms
Gain=484 (80%)

Results: Originally achieved 170 fps, 100% of theoretical spec, however with the flat overlay and CPU involvement, this dropped to 104 fps of 170 potential fps for a 63.4% throughput performance. This is not good! The CPU made a major impact here, costing many frames over the course of time, losing 66 fps recorded due to flat frame overlay processing in the pathway.

This can be eliminated by not doing the real time overlay and instead doing a separate flat video (this mainly applies to solar and potentially lunar, rarely if ever planetary). This could be potentially alleviated with a faster performing CPU with more cores, but I would need to get one to test this concept. If I do get another CPU I will test it. Until then, I think I will do separate flat frames to avoid losing 66 fps, which matters a lot in solar with limited time to capture before features change.

Very best,