17 Sep
2004
17 Sep
'04
9:08 a.m.
Rumor has it that William Maddox may have mentioned these words:
What they are doing is something similar to what Apple
did for the 68K to
PowerPC transition.
Or DEC with their WinNT4 software that allowed DEC Alphas to run 32-bit
Intel WinNT4 binaries. IIRC, it could be done "on the fly" (more overhead)
or "precompiled" and saved back to the hard drive. Pity my memory sucks to
the point I can't remember what it was called...
If you look at Transitive's own website, they
admit that the emulation has overhead.
They'd be fools not to!
I think their claim is that they
are around 80% of native performance on comparably-powerful hardware,
and that in the expected scenario of bringing legacy software forward to
a newer platform, the newer harware would make up the difference.
That's very possible.
With modern dynamic translation techniques, emulation
can be this
efficient, though performance can vary considerably with characteristics
of the workload -- dynamic translation schemes are like caches, and are
sensitive to program locality, among other things. It could very much
be real, in a useful and pragmatic sense, but there's no magic there.
Can you say "Transmeta Crusoe in software?" Under the right conditions
(read: compiled with special optimization flags) it's clock for clock
faster than the x86 it's "emulating."
Laterz,
Roger "Merch" Merchberger
--
Roger "Merch" Merchberger --- sysadmin, Iceberg Computers
Recycling is good, right??? Randomization is better!!!
If at first you don't succeed, nuclear warhead
disarmament should *not* be your first career choice.