Latency issues related to nvidia

Hi,
i've been working on the creation of a 3d printer using linuxcnc for quite a few time now, i've created the base configuration that i will use with the machine once it'll be ready, but now it's time to fix a problem.
I still have a few issues with latency on my pc (an HP XW4300, with pentium4 3.6 ghz, 1GB of RAM, NVIDIA GTX240).
This pc was born with a quadro graphics card in it, but i've never managed to boot a linux distro with it.

Latency looks quite acceptable when in idle or doing non-graphic-intensive tasks (hardly over 40 000, i think that when i reach that limit is still due to the graphic card), but as soon as i run some of the glx tests from the globs package (the first one in particular) i get really unacceptable peeks (sometimes over 300 000 ns). GLXgears for example does not generate peeks like that, but it stillmesses up a little with the jitter.

For this reason after only a few seconds from the start of linuxcnc i get an error message saying "Unexpected realtime delay on task 1. This Message will only display once per session. Run the Latency Test and resolve", and for this reason i'm quite concerned that the machine will eventually stall or do wierd things.

I read that i should switch to vesa drivers in xorg.conf, but doing so will leave me with the magical blinking cursor at boot (only on the graphical terminal, ctrl+alt+1-6 would still open a working terminal).

What can i do to fix this problem?

Thank you very much

I did more or less what is written in that article, except for the fact that i already had a xorg.conf file, but editing it as said (by replacing nvidia with vesa) corrupts the graphic shell
is there a way to boot in some kind of verbose mode to see what is going on?

ps: i forgot to mention that i've installed linuxcnc using the standard debian iso provided on linuxcnc website.

A couple of things about Ethernet latency:

You wont get sserial errors unless you have a sserial link running

Ethernet latency is not checked by the latency test, it can be checked with ping and chrt or you can use linuxCNCs
tmax values for read tmax to check this (these are in CPU clock cycles), for example:

halcmd show param hm2_7i76e.read.tmax

Will give you the maximum time from read-request to data returned from the 7I76E (in CPU clocks)

Also, for most Intel Ethernet chips, you _must_ disable interrupt coalescing in order to get decent Ethernet latency

Hi, sorry for the delay.
I think i actually tested both of them, because now i'm using proprietary drivers (i've already made a few experiments to try to fix this problem when i first started to use this pc for linuxcnc, but quite a lot of time have passed since that).

Hi, i resume this topic to ask another question related to this issue.
Considering that my problems look to be entirely related to the graphics card, i decided to make a test, and i switched from Gmoccapy to Axis.
Well, before i've never been able to run an entire print gcode without experiencing a "following error", now the machine has been running for at least 6-7 hours without any issues.

Considering that the appearence of the 3d preview on the 2 interfaces look identical i used to think that they were exactly the same component. What can cause that behaviour on Gmoccapy?
Do you have any idea of how i could fix those issues so that i can switch back to it?
If the only solution is a new graphic card (it could be worth it) do you have any advice?
Take into account that on glxgears my GT240 doesn't show any issues, but if (for example) i run the first test in the "globs" package i get a spike of 300'000+ on the base thread, so if you have any suggestion for a working graphics card, it would be REALLY appreciated if you could do that test too with it.

Thank you!