Ok heres a post from the Yahoo forum By Art fennerty the genius who invented Mach3 that may may explain something of what I mean and also clear up why Mach likes 25K best.

It's part of a reply over along running issue a guy had and so some of it may not make sense but it's interesting reading and should shed some light.!

Here's the deal with kernal speed.

Lets say you select 25Khz, and then tune your motors. While tuning, you find you have the velocity slider up all the way, but would like to go faster than you can tune.. you then must select the next higher kernal speed, and retune all motors. If while tuning, you find you cannot go fast enough, repeat till the kernal speed is such that you CAN select the speed you need.

The reason is this..

In 25Khz mode, the computer interrupts every 40us.. in 35Khz, every 28.5us , in 65Khz, every 15us , and for 100khz, every 10us.

Now, the time-in-int is the amount of time spent servicing that interrupt. One of the secrets of the printer port driver, one Ive never discussed , is that
all drivers in windows are told to do their thing fast, and get out. My driver breaks that rule..heck it breaks most rules. :) . SO it takes the time it needs to
do 1 step pulse, read input, set outputs, unset the step pulse, jog if necessary, etc.. and then exits to wait for the next interrupt. The time-in-int is the time it takes to do all that.

So lets say your in 25khz, every 40us an interrupt will come along, and your machin has a time-in-int of 15, that means your cpu has 40-15=25us to do its
normal windows activity. Your computer is now 37% a pulse driver, and 63% a windows OS..

Time-in-int doesnt vary from kernal speed to kernal speed, only computer to computer. Lets say yours is 11us. ( a bit high ), if your in 65Khz mode, your now interrupting every 15us.. , so Windows now has 3us every 15 to do its thing, its now a cpu that is 20%windows, and 80% pulse engine.

The windows component runs Mach3, so you now have 20%Mach3 ( plus windows fucntions) , and 80% pulse engine. As the pulse engine % goes higher, the chance of lockup goes higher.

So higher kernal speeds give you higher criticality, which can be defined as your sensitivity to random bad events affecting the computer. The lower the kernal speed, the higher your immunity. As youve noticed, your 45Khz machine doesnt lock up as much as your 65Khz machine. That follows the logic Im explaining. (Im almost willing to bet your machine at 25Khz will never lock up , or would be very very rare, (Im assuming this from your statement about the two machines correlated to hundreds of very similar conversations over the years with lockup victims. Fully 80% of them didnt realize the connection, and tuning lower in kernal speed stopped the problem.

Now sometimes you NEED higher kernal speeds, ( high step count machines typically), but fully 90% of people who have set high kernal speeds
dont really need them, they do so simply as a result of the very human instinct to have the highest possible speed selected. ( "Yeah boys, my engine has triple hedenstock carbs with dual-hemi semi-octagonal rebuf cyclinder hose accessory packs!".. :)

You may wonder..why the heck dont I tell people this more forcefully, since it will stop most lockups. Fact is, higher kernal speeds DO have a smoother
motion just as a result of granularity, and Ive found the vast majority CAN use higher speeds with no problem. ( I am limited here to 65K ), I use 25K though for reliability. With over 25 thousand machines out there, the number of overall lockups is extremely small, and thats including the tendancy people have to select the higher speeds,so as a result, I try to keep my nose out of their selections, unless they have trouble.

The final question you should be thinking about now is.. "Why didnt you ask me this when I complained about lockups.. :) ), the answer is that the
subject comes up so infrequenctly now that even I forgot to ask this basic question about the kernal speed. I used to post it periodically on this group,
but stopped quite a while back. In retirement I guess Im getting forgetful is my only excuse. Hopefully, youll find 25K never locks up.. Im suspecting this is true in your case. I hope so.

For those that want maximum reliability, I suggest this, use as low a kernal frequency as you can live with. On many machines maximum cutting speed is easily achievable in 25Khz, the only sacrifice in using 25Khz is perhaps a slower rapid speed. Rapids are nice, but not as nice as perfect stability, so I tend to give up some rapids to put their power into stability. Look at kernal speed as a trade-off, you can go real fast, or real stable.. (This is only if you have a lockup problem, many can have both and use higher kernal speeds. )

My own suspicion, again based on time and numbers, is that many people would have lockups, but most leave the kernal at the install speed of 25Khz.. I rarely hear form them. I hear most from 65Khz, and Ive heard quite a bit from 100Khz.. this is because those that can run 100khz are a rare breed, very fast clean machines with extremely stable operating system installs. For 65Khz, you better have a nice fast computer. 45Khz isnt bad, lockups on those indicate a computer suffering some periodic problem that slows the time-in-int too much..

None of this incidentally explains a random move. The driver is pretty much incapable of it, it canot move unless commanded, and it takes quite a sequence normally to command it. Youd have a better chance of winning a lottery than getting uncommanded motion.....unexpected..yes. :), that happens to all of us.. usually at our unknown request, but uncommanded...near impossible in the drivers context.

Let us know how 25Khz works.. do you have to sacrifice much speed ( or any ) to use it? And what IS your time-in-int number?

My average time-in-int is 5-7us by the way, which is pretty normal for a 2Ghz machines, by dual core is only 3-4us, meaning it will take much higher
kernal speeds without reaching any high level of criticality. My 1.2Ghz was around 15us.. ( pretty bad, but ran fine..)

Sorry for the ramble, I guess it was time for one anyway, for some reason there are some that actually like my rambles. lol

Thanks,
Art