Computer requirement for OpenPNP

[Marek T.]

To whom plan to build the machine using Openpnp:

When I have started my play with OpenPNP and retrofitting Philips, I was not thinking to much about the computer requirements. Finally, it's only simple two cameras service and Java, nothing specially complicated... Mistake, later I have spent a lot of time and nerves why nothing works like happy people write here... So decided to share with my experiences, maybe someone next will have a bit easier as I would probably have if find a description like this.

So, my first computer used for OpenPNP was some AMD Opteron-185, dual core 2.6GHz. 4GB of RAM and not bad graphic card. In one word: nothing special but thought that it's more than enough. Connected two cameras (ELP 2Mpx + ELP 1Mpx) to Windows7-64bit and tested it using popular vmcap.exe soft - everything perfect. So run Openpnp and one big sh*t from this point:
- only one camera working under Openpnp (opencvcamera) moreless properly. When connected the second camera the first one die no matter what the resolution or refreshment I set.
- ok, it's WIndows for sure! Installed Linux instead and came back to Windows after 2 days (don't say that Linuks is bad, just don't friendly for me and effects the same like with Windows)
So first important information: it's nothing worth to test that two cameras are working at full res and full fps under Windows or Linux. Drivers used by Java it's completely other story.
- wow, probably I must use separated USB host for the second camera as people advise here. So bought USB@PCI and did it as advised. Uff, both cameras works now, no matter that max some 10fps, no matter that at 1Mpx+VGA no more... But the success is that something begun to work at all! Strange that people writes about some 10-12Mpx cameras but probably they say some bullshits :-).
- so next we try to start mount some pcb. And another surprise: why it all works so slow??? One part is picked, alligned and placed in time more than 10sec. After every single command an OpenPNP is thinking about next operation like about something saving life... what the shi*??? Java for sure ;). I had a machine making 4000cph before and made an "upgrade" to 500 - really not sure if it was a good decission :-). But will have the vision, high precission etc, no matter that slow but what the high quality now :-). Echh...
- what now? Lets try with stronger PC, no other idea has left. So bought AMD-A8 7670K with graphic card in processor, 8GB of RAM and AsRock FM2A88X motherboard, plus fast SSD disk, the system still the same old good Win7. Bottom vision connected to separated host USB3.0 (but camera 2.0) and nothing more to it.
- connected to the machine and can't believe: one part placement takes max 3sec including vision process. Everything works 2-4x faster!
- so what now? Downlooking camera picture at VGA is not specialy impressing. So let's go to 1Mpx what was not possible at old computer. And surprise again - it still works :-). But 5fps is not too much, so let's set 30, it must die! But not, it still works with the quality like dreaming :-)
- next step the bottom vision. On old system 1Mpx was max that wanted to work. But it's probably a bit too low to "cover" the parts from 0402 to some 40x40mm as I wanted. So inceased now from 1 to 2Mpx and it still works, so maybe faster? Refresh 30fps, 100ms for settle time - just nothing may stop this system now :-).

Conclusion? If you are not sure what the computer you need for Openpnp to be sure that it will work properly - an answer you have found above. Maybe it could be a bit cheaper but potential savings are not worth your time I think. Maybe it could work still a bit better than at me if you pay for still stronger machine, hard to say, but I don't think.

Good luck!

[Jason von Nieda]

One thing to note on this: When running jobs you should turn the logging system down to WARN or INFO, not DEBUG or TRACE. DEBUG and TRACE cause the vision system to save debug images and this can cause delays plus eat up hard drive space.

Jason

--
You received this message because you are subscribed to the Google Groups "OpenPnP" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openpnp+u...@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/755248a5-1277-46d1-b4d4-161de6e84a61%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Marek T.]

Valuable information, specially when the slow disk.
However I've had visible delays even i.e. between the ending of the pickup and begining to move the part from feeder to camera position. Hard to say why, but not effect like this at the new computer now. I suspect the delays between java machine and windows less visible or not existing when computer much faster, maybe I'm wrong.

[Jason von Nieda]

It would be worth looking at your logs to identify the delay. The logs are timestamped so it should be very easy to see what is happening. In addition there is very little value in running your cameras at 30 FPS. This just causes extra CPU load. OpenPnP won't do anything with all of those extra frames. I run mine at 5 FPS.

Jason

To view this discussion on the web visit https://groups.google.com/d/msgid/openpnp/b29f74cb-1ac0-4124-ba74-9a2220205838%40googlegroups.com.

[Eagle Media]

The Opteron is a 10 Years old CPU-2 threads, 90nm and so on...
The new CPU is a completely different beast 4 threads, 28nm technolgy and so on...

Basically it's important to take care about the memory bandwith!
Every new system with new CPU and quad channel memory should work very well-also with multiple streams and higher resolution...

[Marek T.]

I was checking logs and it was looking normally there but system worked really very slow. But timestamps are stored with 1sec resolution, so stored timestamps distance 1sec may mean in some situations 0.01sec and 1.99 as well - and in this specific situation next operation was stamped in the next sec. I have not analyzed every operations and their timestamps one by one, just few and there was no clearily visible delays. I think there was a sum of different computer hardware problems, history now already.

I know that no big sense for OpenPNP to run it with 30fps, but it looks better on the screen than at 5 :-) and CPU usage is not high in my case (may check it concrete on Monday). But if it works with this paramaters well, it says about the PC performance, that's all.

It's clear that it's an abyss between these processors. But have not expected so big difference in working with Openpnp when saw how good these cameras worked under Windows (vmcap) on old computer, rather thought it was problem of Openpnp/Opencvdriver/Java but not computer... Now I know how misleading it was.
Dual channel memory there in this new computer to be precise.

[Eagle Media]

Sorry, Intel is using 4 channel memory architecture.

I understand You, I cant look in anything slower than 20 FPS either.

[Lisandro B]

All of this sounds like missconfigurations all over the place. Go slow and use a "divide and conquer" methodology

[Marek T.]

The system on this PC set has been changed few times. Win7-32, Debian64, Win7-64 fresh installations. Forget misconfiguration. More I can believe was old motherboard with dried capacitors maybe, however nothing wrong visible while tests. For me opencvdrivers just need pc power and that's all. Everything works like dreaming after changing to A8-4core.

[Webconnector]

 

Hi all, I need your advice I choose a computer (Components) for OpenPnP machine it will only work for pnp as it will be in the machine itself 

CPU:  Pentium® Processor G4560 14 nm

Cores 2

Threads 4

Frequency 3.50 GHz

Cache 3 MB

.............................................................................

RAM

8 Go (2x 4 Go) DDR4 2400 MHz CL15

..............................................................................

Motherboard

Gigabyte  GA-Z270N-WIFI - Socket LGA 1151

.............................................................................

SSD

Samsung SSD 250 Go 960 EVO M.2 Type 2280

................................................................................

this assembly is enough for comfortable use with vision ?

Thanks

[Webconnector]

and the second question needs an external video card or built-in enough ?

Thankssss

[Marek T.]

I'm not experienced with Intel at all, only AMD in all my machines.

But can say that using A8 built in graphic or additional card gives no difference, not any few % of better/worse CPU usage or performance.

Pay attention on the USB host controllers better. If there is only one in your MB - you will need another one for sure.

[Webconnector]

Thanks for the quick response !!

On the motherboard there are USB 3.0 and 2.0 means that there are 2 host controllers?

and from your experience need more Ghz or multi cores ? openpnp (Java) uses all the cores ?   I'm not sure about my CPU choice 

Thanks

[Marek T.]

No it doesn't. If you have there 3.0 it means you must have 2.0 as well... It's not meaning more than one controller.

I have made a short look on the gigabyte terrible site but can't see it specified too clear for the first look.

Note you may be convinced that now Openpnp doesn't need two controllers since there is new camera driver. For someones it is ideal now, for me it is better than before but still nor working good two cameras on the same host. However, you can always add some USB PCI-e card controller if needed.

I would take some i5 rather. I have A8 4-core/3.3GHz and don't feel it's too much. But as said - Intel is not my story, maybe in his case 2c is good enough.

Java eats how much you feed it ;).

[Webconnector]

Thank you very much for the explanation. I'll try on  G4560 and I'll write about the results I do not want to pay 150 euro instead of 50 :) 

In my opinion, many people have questions about PC for openpnp , probably it would be better if people just wrote what kind of computer they use if this topic contains data on 10-20 computers Such questions would no longer be 

thanks again and sorry for bad english