Asrock Q1900B-ITX

Hope this helps, maybe even with other boards too. Corsair is 1.5V version and the others 1.35V, bios setting for memory voltage was in auto setting.

Yes, that's a very valuable discovery and survey you've made: sensitivity to memory module variety. I had not even considered this as a significant factor with latency.

My "happily performing" systems all employ a single "Kingston Technology 4GB 1600MHz DDR3L PC3-12800 1.35V Non-ECC CL11 SODIMM Intel Laptop Memory KVR16LS11/4" module. This choice was not based on anything but "it's known to work in general applications (i.e., not LinuxCNC) of this motherboard, fairly priced, and available".

However with the BIOS setting for memory I am cautious to set to "manual" to match the hardware, not "auto".

My wish is that linuxcnc.org should inform a reliable baseline reference recipe for a PC parport system from scratch out of commodities you can conveniently "Buy Now" (for Now = today). We might call these live hardware recipes the "warrant" systems. The wiki articles are informative, but there's no definitive reference to current commodities. Identify what available parts you can buy today that require no art, experimentation, or drama on the part of the machine builder to achieve, other than normal PC assembly techniques. This is a matter of responsible curation, not commercial promotion. In 2017, this forum thread was the closest thing to that current information I could find at linuxcnc.org.

Literally any DELL Optiplex series with a Pentium 4 or Pentium D, Core, Core2Duo.
Price from 10 to 100 $. Plenty of them on sale, they never fail.

I should have qualified that "available" means something you can buy new, not scrounging around for used items and uncertain logistics.

A LinuxCNC cookbook should not start with, "first catch your hare."

Is the DELL Optiplex Pentium currently on dell.com a "warrant"? But these cost $700 and are desktop cases with fans.

The ASRock Q1900B-ITX sold for about $80. Assembled into a 1U rackmount case with power supply, HDD, and 4GB memory, the ready system totals under $200.

tommylight wrote:

Literally any DELL Optiplex series with a Pentium 4 or Pentium D, Core, Core2Duo.
Price from 10 to 100 $. Plenty of them on sale, they never fail.

I disagree. I have two Dell 775 Core2Duo 3GHz towers and both are bad performers for linuxcnc. Maybe I need to test different memories for those too...
Dell GX270 P4 is really good, I have two of these in linuxcnc use but they are rather old...

Also it would be good to know which NEW board/models to use, not only some 10 year old models even they are still available somewhere...

Yep. But 5+ years ago is new compared to my Dell GX260 (sorry, typo in previous message saying 270) released in 2002 and rescued from being scrapped from work about 2008 or so.
The Fujitsu D8830 that I'm now replacing is "much" newer, I started emc/linuxcnc with it in 2009.
At least here these are not available and there is not much sense shipping them from other countries.

I'm a good example of "Don't touch if it works" and have problems replacing some 10+ year old hardware after 9 years of own use...
I believe many do the same and not upgrade their systems unless needed.

Anyway now I found a new Q1900B-ITX to replace that Fujitsu, let's see how it goes when the old Dells get broken (I have 1 spare waiting so not very soon I hope)...
One possibility is to buy 2 more Q1900 now and put them to wait to be needed. Or upgrade all three now at the same time. But the old ones still work well so...

It would also be interesting to compile 4.something kernel with rtai and see how it performs on newer hardware like J40xx or J41xx. Just have to get the hardware somewhere first

@Richard J Kinch, have you ever checked cpu temperature when running inside the case ?
I understood you don't have any fans there.

I left the board running latency test overnight and the temp stayed at 54C with about 10% idle on the table in open air, room temp about 22-23C.
Just wondering if a fan is needed with a small case...

ississ wrote:

@Richard J Kinch, have you ever checked cpu temperature when running inside the case ? ... Just wondering if a fan is needed with a small case...

I run them in 1U rackmount cases that are entirely closed other than some perforated vents on the narrow front and back. It's very cramped in there, so I suppose that counts as a small case.

My "temperature check" is to touch the top of the case above the CPU heatsink, or the heatsink itself if the top cover happens to be off. This is never more than comfortably warm to the touch, but it feels like a few watts of warmth, indicating that the heat is dissipated by conduction/radiation and not via the slight ventilation.

My 1U setup is not strictly fanless, because the compact power supply has a tiny fan in it, and that pulls a small flow into the front and out the back. What I mean by "fanless" is no fans in the case itself. The 1U case has mount points for tiny fans, but I don't use them.

You can also run these off 12VDC wall-warts, with those little gadgets that make the various voltages on the connector, and in that case you wouldn't even have the power supply fan. Maybe that would trap heat too much.

One of my dogmas is to manage all the electronics from rackmount and DIN-rail hardware. I find this the most efficient and economical way to keep all the complexity under control, reliable, modifiable/improvable/staged, and maintainable.

Ok. Temps are also easy to check by installing lm-sensors package.
I'm planning to use Chieftec 120W supply. The wall block gets only slightly warm and the pcb part stays cold.

Depends on the wall-wart and mini-psu used, some mini-psus let the incoming 12vdc straight through and then the wall-wart part must be good enough for the mb & co.

Hi again

Unfortunately I counld't get more memory to test.

There is a not-so-nice thing with the parallel port.
The 8 data lines (pins 2-9) give 0/3.3V as they should. But the other outputs (pins 1, 14, 16 and 17) give only ~1.2-1.4V when set to 1 on linuxcnc.
According to the NCT6776D datasheet all these outputs are of same type, 12mA source/sink (page 9, 11,12) here

Also tried to change mode on bios but that has no effect on voltage levels.

I am driving HCPL-2630 optocouplers with 470ohm resistors which results in about 4mA drive current which should be acceptable for any parallel port. Measured on pins 2-9 voltage is 3.1V and drive current 3.3mA.

I also tried with a 2.5k pull-up on pin 1, setting 0 gives ~90mV and high ~2.2V so it's not an open drain.

Maybe the board is broken ?


And yes, my cable is correct, those outputs were not shorted to ground in any moment.

Edit: this was interesting.
Problem went away with this: disable parport in bios, boot (linux does not find parport). Reboot, set parport to manual, ECP1.7, 0x278, all irqs => works as it should.