Best PC Hardware (Q4 2016) for a new build

I ran machinekit on Beaglebone black for a short time on my milling machine. That worked admirably well, that small card could drive the motors faster than they could handle. One-card, really cheap system, kind of perfect for control. But there were of course limitations. Limited amount of memory so after x 100.000 lines of codes it froze. And the graphics was really, really slow. Loading a large file was a pain. It could do the control thing, but it wasn't workable. So I bought a new small pc and a mesa card. What a difference. Not from control point of view because the 7i76e worked of course, but from a usage perspective. Really big files load in a few seconds, snappy graphics. I use two monitors one pointing to the milling machine and the other one pointing to the lathe. I am more than happy. Once i tried it I knew it isn't just the control thing, it comes down to user expectations as well.

Am I right in that there is no lower limit on the pc configuration when using a mesa card? Among currently marketed systems I mean. How to select a good pc wil be a personal system optimization exercise. Driver availability, size of the system, chassi and power supply or naked, fanless, cpu, memory, network ports, pci?, wifi, hdmi are examples of choices to be made. A powerful system may be overkill and just cost extra in many situations. The most low-end configuration might not meet the impatient person's expectations when say loading big files. And some must have a really low-power system without fan. What is best for one is not best for everyone. If there is a lowest configuration it would be good to know that, and let the user's expectations determine how far above that to go.

This was the post, I followed when selecting hardware.
forum.linuxcnc.org/38-general-linuxcnc-q...7-any-good-usff-pc-s

I really wanted to find one of the dual NIC Zotecs PCW used but they are not available in my country.

The USFF is more work as to get the best out of the newer hardware you will probably need a later version of LINUX.
To use the Mesa ethernet cards, you need to patch the kernel to get pre-emptive support. PCW has posted a script that works well on the forum.

Then if not using Debian, you need to compile LCNC from the source. Once again, when you finally put it all together, it is easy.

Mint 17.3 is by far and away better to use and has better hardware support in my experience than Debian Wheezy (which was a fail) or Jessie.

And my journey getting it going in the last few weeks is here.

forum.linuxcnc.org/9-installing-linuxcnc...ng-uspace-for-master

Thanks @rodw & @PCW, I missed the message limit and delays on the ClearPath page (and of course their sales rep specifically suggested that the direct control would work perfectly for a CNC machine ). The Step and Direct was our back up plan, but without the direct control the ClearPath might not have an advantage over something like the Hudson series servos using a proper servo controller. My understanding is that we can plug our absolute linear encoders directly into most servo controllers for improved feedback. What is still appealing about the ClearPath and Step & Direct is being able to drop in a stepper if we want to control costs.

This leads me back to our design goals - I'm really looking to build a milling testbed, not just a working machine. I appreciate that LCNC is very lightweight, but I want to both make sure that hardware can be scaled and maintained. I might want to make a few dozen or more of these machines, and want to be able to keep parts available for maybe 4 years. Stability is important for us, so is the ability to run things on the side without too much worry.

I'm going to RTFM this weekend and get a better understanding of how LCNC works under the hood. I can do an install on one of the mini ITX boxes I have available, and I've got a pair of servos that I can play with, I'll come back with more informed questions next week.

Thanks again for all your advice!

I think there are 2 ways to build a testbed.
1. Use an Ethernet card that clearly separates the PC components from the motion side. My model.
2. Stick a motherboard in your electronics enclosure and treat it as just another building block. Andy's model.

Downside of ethernet is that you have more work to get a working PC system. You also probably will buy different cards for stepper or servo systems but there are ways to support both. (7i80 I think)

The motherboard model is probably the most flexible, simplifies software installation and allows you to start with a parallel port and move on to a PCI Card, adding the daughter cards required to do what you want to do. Some of the motherboards from VIA and others have a guaranteed life which might be of interest to you.

For me, ethernet made sense as the PC side could be separated from the noisy plasma environment. For a mill, that is not so important.

ArcEye wrote: Make sure you choose one with a parallel port, or spend the extra cash you have on a Mesa 5i25 and 7i76 combo and latency becomes irrelevant.

Could you please explain why latency is irrelevant when using the 5i25/7i76 combo?

thanks
Steve

It isn't 100% irrelevant. But the requirements move from <50,000 nS to <500,000 NS

The 7i76 needs an update every 1mS, whereas software step generation requires one every step.