Chinese 3040 machine

Andy Peters said:

Can you do direct parallel-port access with a modern version of Linux?

-a

Click to expand...

Not really: nothing like what could be done with OUTB in DOS days.

Most of the 'parallel port' cards are just bit-banged interfaces sitting behind PCIe bridge chips like those from PLX Tech.  Programming them involves a complicated interrupt model and you can't guarantee latencies (don't ask me how I learned this).    In addition, most force you to go through an extensive FIFO (as they are assuming you'll want to burst many megabytes of data at maximum speed), so the programming model looks something like a) send a short packet, b) ask the PCIe chip to flush it's FIFO, c) spin waiting for that to happen (or register a callback for when it's done), d) weep into your coffee mug.

Matador said:

Andy Peters said:

Can you do direct parallel-port access with a modern version of Linux?

-a

Click to expand...

Not really: nothing like what could be done with OUTB in DOS days.

Most of the 'parallel port' cards are just bit-banged interfaces sitting behind PCIe bridge chips like those from PLX Tech.

Click to expand...

It's actually worse that that, because at some point, the interface went from being the Centronics Parallel Printer interface to the Enhanced Parallel Port or some such. Part of the "upgrade" was to make the port truly bidirectional. And that's when it went from being something entirely controlled by the host processor to an offloaded peripheral.

Programming them involves a complicated interrupt model and you can't guarantee latencies (don't ask me how I learned this).    In addition, most force you to go through an extensive FIFO (as they are assuming you'll want to burst many megabytes of data at maximum speed), so the programming model looks something like a) send a short packet, b) ask the PCIe chip to flush it's FIFO, c) spin waiting for that to happen (or register a callback for when it's done), d) weep into your coffee mug.

Click to expand...

You need the FIFO because PCI is a bursty interface. It hates single-beat transfers. Plus you usually have a DMA engine in charge of the data transfers, not the host CPU. 

The PLX parts are weird. Their PCIe-to-local-bus part (PEX8311) is actually TWO of their chips in one package:  the PEX8112 PCIe-to-PCI bridge, and the PCI9656 PCI-to-local-bus bridge. And there's a PCI bus between the two chips, and both chips require their own configuration EPROM, and it's all really silly.

If I had to start a PCIe design today, I'd use an FPGA with the PCIe hardware built in.

-a

Due to famliy obligations, porches to be built etc. the diy-time has been reduced to a minimum. But some cnc-ing have been done.

Initially I had trouble with scaling. But after browsing the applied home-burnt cdr:s that came with the machine I found the right xml and problem solved.

I´ve worked on acrylic for some panels. No problems here besides calculating and experimenting feeds and speeds for the best cut. Plastic can be a PITA but material is easy to get hold of. Machinewise it works great.

Yesterday was alu-day. 6mm pockethole and slot was cut for trials. Feeds: 200mm/m, 0.3 mm/pass. With these conservative settings it worked fine althou rigidity comes a bit short in the plungemove. The spindle/structure throws a bit when plunging leaving a small "dogbone" shape when entering the material. Laborating with bit and plungespeed may overcome this. And working with roughing/finishing. Bits used: 6mm 3 flute HSS Tin and singeflute 1/8" cheapo Carbide. 600w dc spindle at 14000 rpm.

General issues is still that the bed is not flat. For high accuracy work, PCBs: or alike, a facemilled fixture will be needed.

following.

How much is the variation from flat?

Jakob E.

In this picture you can see how the table deviates from course. This may be overcome by changing the alu-profile or working with a plained spillboard. As I see it, this is the weakest point of the machine. But in next post you se that it is possible to do detailed work anyhow.

This is done by attaching laminate with doublesided tape to non-plained mdf spillboard. 0.2 mm -z with cheapo v-bit 30 degree that came with the machine.

wow. that is excellent. I had nowhere near that level of precision with my little zen toolworks 7x7 machine.

Yes. More boards have been produced with great results. Upcoming is a big tube-pre board 160x100 mm. The bed uneveness may show when working in bigger format. Does anyone have exeperience of autolevelling software? It does indeed look like a good way to overcome shortcomings of materials and machines.

Do you all work with v-bits on pcb:s? In delicate situations it is needed no doubt. I did a tube PSU-board the other day using a 1,5 mm chipbreaker end mill with good results. Using end mill will also reduce need for exact flatness as you can go deeper if you have air cuts without compromising trace width.

Ps. Limitswitches now work after previous uploading of correct xmls.