One of the cool things about it (and this is not really about Memtest, even though it factors into it) is that there exists a Linux kernel patch that makes it reliably use bad memory by avoiding the bad spots. Oftentimes, memory goes bad in such a way as to cause only particular offsets to fail, and you can use Memtest to find those bad spots and produce the correct configuration information to then tell Linux to not use those areas of memory. And it's a pretty slick solution, too; it just allocates those areas of memory in the kernel as if they were being used as data and then never uses or releases them.
That's definitely cool, but you've got to reckon that nowadays it's a technology whose time has come and gone.
(Oooh, can we have a "when I were a lad" thread? Can we? Can we?)
When
I were a lad, RAM upgrades for A540s cost UKP400 for 4Mb, because the memory architecture was such that a new memory controller chip was needed for each 4Mb. We Archimedes users jealously eyed PC users whose RAM came in commodity SIMMs, and for the sizes anyone would actually need (4Mb, 8Mb) cost about UKP25 per Mb. Then the RiscPC came out, and it took two standard SIMMs up to 128Mb each. Atomwide published SIMM prices all the way up to 64Mb, and also offered 128Mb SIMMs. They didn't publish the price for 128Mb SIMMs, but did say that if you bought two 128Mb SIMMs they'd throw in a free RiscPC worth UKP1500 to put them in.
And nowadays Scan are flogging off 64Mb SIMMs at UKP9.50 and 128Mb SIMMs at UKP10. Mapping out bad RAM blocks just isn't worth the candle...
Peter