This discussion intersects with one which occurred mere weeks ago.
I cannot confirm the certainty of what I say here, but the indications have left me with comfortable conclusions -- even if tentative -- about the latest version (3.34? 3.4?) of MEMTEST86.
If you find the MEMTEST86 web-site, it immediately presents some sales-hype to you about the convenience of charging $10 to your credit-card so that a working bootable CD of the program can be sent to you. For the money, this is not unreasonable, and not out of line with what MS and other software-publishers charge for copies of service-packs and other free downloads.
However, $10 is $10. A CD costs pennies. And a pile of $10 checks buys a lot of kiddie-porn and junk-food. Bad joke. Buys a lot of anything . . . .
More than one of us determined that neither the floppy-disk version nor the CD "ISO" version create workable copies of the program on the respective mediums.
If you look further at the MEMTEST86 web-site, you see that important aspects of MEMTEST86+ v1.70 were incorporated in MEMTEST86 v.3.4(?) -- addressing newer processors and chipsets.
Go to the web-site for MEMTEST86+, download v.1.70, and try making self-booting disks (or discs) from it instead.
Also, Microsoft provides a memory-testing program called WINDIAG Memory Diagnostic which will test up to 4GB of socketed memory.