Is this helpful? https://www.minuszerodegrees.net/5170/post_er … _post_codes.htm

All three versions of the BIOS are the same here.

09 Refresh bit - Verify that 'refresh bit' is toggling. See note 9.
Note 9 Verify that channel 1 of the 8254 timer chip is periodically generating a 'RAM refresh request'.
So what is being looked at is simply something that triggers a RAM refresh cycle.
That is not the same as verifying that RAM refreshing is actually occuring.

0A 8042/8742 keybd controller - Buffers See note 8
Note 8 1. If output buffer full, flush it. 2. If input buffer full, and 100 ms later is still full, halt the CPU.

Thank you for your quick reply! It is a big help, I missed this post section on the minuszerodegrees.net site.
Checked the timer IC with logic probe. All the three channel's clock, gate and output are active until the post reach the 09. After that all the activity dissapeared.
Checked the DRAM CAS, RAS signals. These show a quick burst of activity until post 09. Tha address and data pins are moving as well, no stuck lines detected.
Checked the 8042 as well. The CE, RD, WR signals are active until the 09 post.
My next try will be a test with Supersoft diag rom, but eeproms shold be sourced for that.

Is this machine still using DMA controller to steal CPU cycles for DRAM refresh? If the mobo has 2 DMACs I would swap them places to see if that does anything useful. Assuming it's not integrated into chipset already, I have very little experience with original IBM PCs.

Deunan wrote on 2025-07-21, 20:26:

Is this machine still using DMA controller to steal CPU cycles for DRAM refresh? If the mobo has 2 DMACs I would swap them places to see if that does anything useful. Assuming it's not integrated into chipset already, I have very little experience with original IBM PCs.

Yes it has two 8237 DMA controllers. Unfortunately they are soldered directly, swapping would be a painfull job.

Waited for the arrival of the eprom eraser, sorry for the delay. Cheap China stuff, lucky to get it without a broken lamp. It works well, but the manual springy timer is killing me.
Landmark diagnostic ROM for the 5170 burned and tried. The test showed RAM refresh error - 2xHI/LObeep + 2xshort beep. Check the user guide of the diag rom for explanation of this error.
Every owner of 5170 is a lucky one because the IBM PC AT reference guide contains the 22 page schematic diagram. https://bitsavers.trailing-edge.com/pdf/ibm/p … rence_Mar84.pdf

The test writes the memory, wait for 10 seconds and read the memory back and check. The RAS was active during the write but was inactive during the idle period. This is the reason the DRAM lost its info, no continuous RAS was present.
Checked how the RAS was generated. My model has only 2 banks therefore only RASO drives the RAS lines of all the memory chips. Focused on RASO, how it is generated.
The 8254 timer IC chanel 1 is producing the necessary timing for the RAS. Checked the signal route from that angle as well.
Finally the two ends come together in a circuit whis is quite a strange one, see sheet 21 of 22.
Sygnal Refresh was stuck during the idle period. Checked U95/74F175 inputs and outputs. Clock was present but found one anomaly. D4 input was high but Q4 was low. Resistance to GND was high, there was no short there. U95/Q4 connected to U68 74F74 CLR inputs. Potential faults: U95/D4 flip flop or U68 CLR inputs pull down the line.
U95/74F175 was removed and replaced with a new 74LS175. Viola! Refresh signal line was active during the idle period of the test. The MB started without any problem after installing the original bios chips back.

One remark: have to be very careful not to stuck in the diagnosing process. I found a strange behaviour and spent a lot of time to try to figure it out but it was a dead end.
During the test the D10 data bit stopped activity, remained low during the idle period of the test. All the other data lines showed continuous activity. It was true for the whole databus, regardless of how many buffer-driver circuit had passed. Accidentally bumped into this measurement and I was not able to determine wether it is the source of the problem or not. Finally I step outside of this hole and walked back to the base: there is no continuous RAS. Lessons learned.

One more remark:
If you get an old machine with 16V tantalum caps on the MB and on the MFM drive's MB than the wise move is to replace these before applying power. It does not worth to risk the health of the MBs.