I wouldn't call ARMv7 "bad, slow and limited compared to ARMv8" either. If you look at the recently released instruction timings for the A57 they're not substantially faster than the old A8 timings. The big differences are mainly from hardware division support being mandatory and no need for backwards compatibility with older models which caused unaligned memory access to always be handled by software, even though ARMv6+ hardware can handle it.
(03-07-2015, 05:43 AM)tueidj Wrote: [ -> ]I wouldn't call ARMv7 "bad, slow and limited compared to ARMv8" either. If you look at the recently released instruction timings for the A57 they're not substantially faster than the old A8 timings. The big differences are mainly from hardware division support being mandatory and no need for backwards compatibility with older models which caused unaligned memory access to always be handled by software, even though ARMv6+ hardware can handle it.
I have a feeling even in the face of logic many of these console devs won't touch armv7, either because they've been convinced it's crap, or because they feel it's too "technically last gen" (or some other bullsh*t buzzword phrase). For many of these announced titles I'm sure armv7 would be enough (ignoring actual CPU speed differences). As for Crysis 3, crytek can basically say whatever they want because they are tech gods. Another factor to consider is the X1 in the SHIELD console has basically no ceiling (with a wall plug and a chassis that big it could run full blast without penalty) so that adds to potential performance deltas, making the A57's look even better than they should.
(03-07-2015, 05:43 AM)tueidj Wrote: [ -> ]I wouldn't call ARMv7 "bad, slow and limited compared to ARMv8" either. If you look at the recently released instruction timings for the A57 they're not substantially faster than the old A8 timings. The big differences are mainly from hardware division support being mandatory and no need for backwards compatibility with older models which caused unaligned memory access to always be handled by software, even though ARMv6+ hardware can handle it.
Armv8 support [color=#252525]
AES encrypt/decrypt and SHA-1/SHA-2 hashing instructions, have more general purpose registers(13 vs 31), [color=#252525]
dedicated SP, [color=#252525]Advanced SIMD (NEON) enhanced : [/color][/color][/color]
32× 128-bit registers (up from 16), also accessible via VFPv4 and supports double-precision floating point.
(03-07-2015, 11:10 AM)Kwee Wrote: [ -> ] (03-07-2015, 05:43 AM)tueidj Wrote: [ -> ]I wouldn't call ARMv7 "bad, slow and limited compared to ARMv8" either. If you look at the recently released instruction timings for the A57 they're not substantially faster than the old A8 timings. The big differences are mainly from hardware division support being mandatory and no need for backwards compatibility with older models which caused unaligned memory access to always be handled by software, even though ARMv6+ hardware can handle it.
Armv8 support [color=#252525]AES encrypt/decrypt and SHA-1/SHA-2 hashing instructions, have more general purpose registers(13 vs 31), [color=#252525]dedicated SP, [color=#252525]Advanced SIMD (NEON) enhanced : [/color][/color][/color]32× 128-bit registers (up from 16), also accessible via VFPv4 and supports double-precision floating point.
most of those extras are situational in usage for normal applications. Unless the app is using transcoded x86 libraries, they won't need those extra features for any core functions. I think you're being a bit harsh on the differences.
ARMv7 has conditional execution for almost every instruction, can push/pop large ranges of registers in a single operation, and can separately address the high 64-bits of its NEON vector registers allowing for many horizontal manipulations that ARMv8 doesn't support. It also supports VSWP (which was dropped) and single iteration VTRN, VZIP and VUZP (which only process half the same amount of data on ARMv8).