On Fri, Mar 10, 2017 at 06:34:55PM +0000, Luke Kenneth Casson Leighton wrote:
On Fri, Mar 10, 2017 at 5:19 PM, Paul Boddie paul@boddie.org.uk wrote:
been trying to get their Zen microarchitecture out. However, AMD have a history of underwhelming the market in many ways, so it makes one wonder what
this comes down, unfortunately, to the construction of the x86 instruction set. i've written about this at length in the past, re intel, but of course it applies to amd as well.
to achieve the same performance as pretty much any RISC processor an x86 instruction set has to run at *at least* double that of RISC. the reason is down to the historical compactness of the x86 instructions (using escape-code sequences) which was fine when memory was ultra-expensive over 20-40 years ago.
the specific mstake that AMD made was in not realising this.. and selling their foundry (to become globalfoundries) insead of realising that it was *absolutely essential* to stay ahead of TSMC and keep up with Intel's geometry-ahead-of-the-game plan... which they're still losing btw.
bottom line, if AMD want to stay in business they need to get out of x86. part-hardware-emulated x86 fine (like the Loongson 3H architecture did), non-x86, fine. pure x86: dying and dead very soon.
Intel already tried that a *long* time ago, with the Itanium. It was provided with software that emulated the x86. But AMD made a 64-bit hardware version of the x86 and took over the market because its hardware outran the emulation on the Itanium, forcing Intel to follow suit or lose the Windows market.
Is the situation different now? With an ARM version of Windows, and Microsoft's now proven ability to port Wondows to new architectures, quite possibly.
-- hendrik