keenerb wrote:

http://wiki.osdev.org/ISA_DMA

According to this, DMA will run at 4.77mhz, "NO EXCEPTIONS" on an ISA system, so this was probably a waste of $16, but we'll try anyway.

I'm not convinced that link is correct. We looked at the Xi 8088 schematic, and DMACLK is set to the same as the CPU CLK.

Yep, I dont even KNOW what's true or not!!!

my undestanding is the dma speed is the least of either the ram speed or the dma controller speed.

directly from the intersil 15mhz 82c37 data sheet;

CLOCK INPUT: The Clock Input is used to generate the timing signals which control 82C37A operations. This input may be driven fr […]
Show full quote

CLOCK INPUT: The Clock Input is used to generate the timing signals which control 82C37A
operations. This input may be driven from DC to 12.5MHz for the 82C37A-12, from DC to 8MHz for
the 82C37A, or from DC to 5MHz for the 82C37A-5. The Clock may be stopped in either state for
standby operation.

now the original 8237 chip or the -5 chips can only have a 5mhz max CLK (so turbo on those would be broken maybe?)

My first set of memory adapter boards arrived. These are the ones without the buffering:

Unfortunately, no improvement in stability. These are 10ns chips with built-in ECC. I figured it was a longshot that a 10ns chip would be any more stable than a 55ns chip at the speeds we're functioning at, but it's an interesting datapoint regardless.

I'd expect my second set of memory adapter boards to arrive maybe by the end of the week. Those will be the ones with the onboard buffering.

Scott

Well I had zero success over the weekend.

I lost the 12mhz DMA chips somehow, they've vanished.

My 8-bit memory expansion board won't work in a 16-bit slot due to clearance issues.

I didn't have time to breadboard the memory either, with it being my wife's Master's Degree graduation and Mother's Day...

keenerb wrote:

Well I had zero success over the weekend. […]
Show full quote

Well I had zero success over the weekend.

I lost the 12mhz DMA chips somehow, they've vanished.

My 8-bit memory expansion board won't work in a 16-bit slot due to clearance issues.

I didn't have time to breadboard the memory either, with it being my wife's Master's Degree graduation and Mother's Day...

I've now gotten my replacement interrupt controllers and DMA controllers, all rated at > 10mhz. I'll be replacing them tonight hopefully, and trying out the standalone XT memory card.

Replaced the remaining LS04 chip with ACT04.

Replaced DMA controller and both PICs with 10mhz and 12mhz arts

Running memory test. Checkit benchmark reports identical speed (2.00 times an IBM XT).

My buffered RAM boards should be here tomorrow. Ought to be able to try them out by the weekend.

Scott

replaced my 2x 82c59's, my 82c37 and for kicks put in a D8087-1, wow that thing gets hot doing nothing! Ill probably heatsink it just as a matter of course... waiting for mouser now. got like 100 different IC's coming (gotta fix some arcade PCB's which need replacing lots of 74F/ls/als/etc chips too)

smbaker wrote:

My buffered RAM boards should be here tomorrow. Ought to be able to try them out by the weekend.

Scott

Mine as well, but I don't have the RAM I need.

As an aside, the memory test I started late last night was still running with no errors this morning, and this was running with two normally "problematic" cards.

Perhaps replacing the dma/pic with faster parts did help, but I'm pretty pessimistic. I've gotten this far before with no issues, so only time will tell.

Yes the 8087 gets REALLY hot. I think it nearly doubled the power draw of the board if my memory is correct.

Tried out my buffered RAM boards... They're even less stable.

The Xi 8088 will error out the buffered RAM on post in my 7-card configuration. If it's in the first RAM socket, it won't boot. If it's in the second socket, it'll just declare it bad and ignore it. Removing the CP/M coprocessor (certainly the most troublesome card I have), it will boot without error but checkit shows a high rate of memory errors.

This is a surprising result. I figured in the worst case, the buffered RAM boards would accomplish absolutely nothing. I didn't think they'd be worse than unbuffered RAM.

Or it's not doing the buffer thing corectly.

Checkit is still running, no errors evident. This is unusually stable for a five-card setup.

I would think if it wasn't buffering correctly, then it wouldn't work at all. I'm tempted to build a second board, just in case I cooked this one with the hot air or something else weird happened.

Another thing to look at, the buffer is one more load attached to MEMR, and MEMR is that oscilloscope trace I had that showed large overshoot and undershoot.

I tossed two more cards into the mix, my adlib and my soundblaster. Restarted memory check in FAST mode.

Does checkit alert you when memory errors happen, or does it wai till the end to alert you?

I ran 99 quick memory tests; received an error on one, execution 73.

ERROR at Address 097c30h, bits .EDCB..8........ Parity

Seemed MUCH more stable than normal, with 7 cards installed.

Very interesting. Is there any way you can narrow it down to which of the changes improved your stability? I wonder if it's some unexpected side effect of the 8087.

Scott

I haven't re-installed my 8087. It just locked up running Champions of Krynn demo after about three hours, so it's not THAT much better.

I've filled up my 10 slot ISA backplane. Re-running memory tests.

checkit running memory tests wont really exercise the bus, so its interesting that the memtest ran for hours with but one error, but when you start using cards on the bus playing CoK (disk reads, etc), it locks up much sooner.

I completed around 700 memory tests with no failures this afternoon, with seven cards installed.

That's a massive improvement over my previous attempts.

Is there a possibility that the ISA signals degrade over time; has anyone left their xi8088 running for a while and then inspected the signals? I think the answer is yes, but just verifying...

I haven't performed a comparison of stability over time, though I would not be surprised if there may be some temperature-related component drift (though I would expect the computer to reach a steady operating temperature faily quickly)