I have never seen an IBM machine 'from the factory' that used an EPROM for the ROM BIOS. They have all been burned ROMs unless it was in a prototype.
IBM sometimes used EPROMs in the 5160 and 5170.
In the 5170, there are PROM and EPROM versions of the final BIOS (11/15/85).
Different part numbers are found on the top of the chips, and IBM changed the contents of the EPROMs to reflect the different part numbers.
They also had to adjust the final byte in one of the EPROMs so that the checksum became 00.
PROMS: 62X0820/62X0821
EPROMS: 61X9266/61X9265
A picture of both sets is at
http://members.dodo.com.au/~slappanel555/misc/5170_ROM.JPG
Later, when the final BIOS (05/09/86) was created for the 5160, both PROM and EPROM versions were created.
The contents are the same but IBM use different part numbers on the top of the chips.
PROMS: 62X0890/68X4370
EPROMS: 59X7268/62X0819
Within the ROM for position U19 is the string, "62X0819=68X4370 59X7268=62X0890".
So why did IBM use EPROMs? Cost? PROM supply shortages? I notice that in the 5160, the EPROMs used by IBM were 100nS. IBM could have used 200nS chips, and so that suggests that cost wasn't the reason.
ROM failure on old motherboards appears to be fairly common. Since becoming a member of these forums a year and a half ago, I've seen about 5 threads where the failure of a 5150/5160/5170 motherboard was due to the ROMs. In fact, only a fortnight ago, the faulty 5170 that dkedrowitsch wrote of (
http://www.vintage-computer.com/vcforum/showthread.php?t=9601) had a faulty EPROM - offline we compared the checksums to known good ROMs and determined that his 61X9265 EPROM has corrupted contents.
In case you don't know, a crude test to check the validity of the ROMs in the 5150/5160/5170 is to calculate an 8 bit checksum.
For the 5150, the 8 bit checksum of U33 should be 00.
For the 5160, the 8 bit checksum of the combined U18/U19 should be 00.
For the 5170, the 8 bit checksum of the combined U27/U47 should be 00.
Eg. 61X9266 (U27) = 10 , 61X9265 (U47) = F0 ---> 10 + F0 = 00