Discussion [WikiChip Fuse]The x86 Advanced Matrix Extension (AMX) Brings Matrix Operations; To Debut with Sapphire Rapids

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

tamz_msc

Diamond Member
Jan 5, 2017
3,710
3,554
136
  • Like
Reactions: Tlh97 and Burpo

Doug S

Platinum Member
Feb 8, 2020
2,202
3,405
136
Intel never deprecates anything, they still support booting in 8086 real mode even though it no longer serves a purpose in modern 64 bit CPUs.
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
I wonder what the microcode bloat is these days to support all those legacy instructions and modes? I can’t imagine anything that isn’t used commonly still exists on any sort of “hardwired” fashion.
 

Doug S

Platinum Member
Feb 8, 2020
2,202
3,405
136
I wonder what the microcode bloat is these days to support all those legacy instructions and modes? I can’t imagine anything that isn’t used commonly still exists on any sort of “hardwired” fashion.

Consider how tiny the 8086 and even 80386 were in terms of number of transistors. Chip area to support that legacy is probably not even measurable. Which is probably why Intel hasn't ever made a move to try to drop it, even though there would be no possible objection.

When Apple dropped its 32 bit legacy it had a goal in mind - yes being able to design ARMv8 only support helped because they could drop the ARMv7 ISA (I think it even had to support Thumb to meet the condition of the ARM license despite Apple never using it) which I'm sure was a bigger percentage of core area than stuff like x86 real mode on the Intel side. We can see now the real goal has been to get everything on both the x86 and ARM sides to 64 bit only, to minimize the complexity of the transition as much as possible.

There's perhaps also some value to Intel to keeping that legacy around - for instance they could deprecate 387's 80 bit FP but that gives them a hidden advantage against someone trying to port that code to another architecture. Nothing else supports Intel's 80 bit FP format, and IEEE 128 bit double precision FP requires some complex rounding if you need bit for bit compatibility with Intel's 80 bit format. So it may act as a lock-in of sorts for certain legacy software.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
Intels 3DNow! moment? How things change and stay the same...
Hardly, 3DNow was really getting on when it was deprecated by AMD.

AVX512 is much more recent, but a complete disaster in terms of how it has been handled compared to SSE2, SSSE3, AVX and AVX2.