Originally Posted by sdlvx
No, it's a significantly more amount of instructions.
You are also forgetting about AVX. I've seen custom compiled Blender end up more than twice as fast with AVX and other optimizations enabled when comparing Gentoo to Windows.
Phoronix has also seen massive gains in some things (but not always) by enabling AVX.
However what I'm getting at is that if a game can use AVX, FMA, etc very well (which are a part of Jaguar) it would be unsurprising to me if Jaguar in PS4 ended up twice as fast as a Windows version running the same code but in legacy SSE mode.
This has been a major point that people discussing PS4 seem to constantly miss. As someone who has cut their render times in half by switching to AVX instead of a standard 'runs for everyone" Windows exe, I think that to assume that Jaguar will be comparable to desktop CPU performance because they are both x86 is a horribly short sighted. I wouldn't be surprised if a properly optimized for AVX program was twice as fast on PS4 as it was on the PS4 version of the APU AMD is going to sell running SSE code. And if it's x87 (like Skyrim), the AVX version would humiliate it. All on the same silicon.
I think that for a lot of people who are writing off PS4 because it's a 1.6ghz 8 core Jaguar are going to get a very harsh lesson in x86 software optimization and how CISC architectures work where instructions are meant to give more performance, not efficiency or tweaks to the microarchitecture. It should wake a lot of people up to the fact that a bunch of CPU graphs are completely useless if you don't know what kind of software optimizations and compiler settings were used. Needless to say I'm looking forward to the day when people realize running 4 games for a CPU review and declaring a winner is virtually meaningless given how much software choices can make.
EDIT: I'm just going to make a raw speculation based on what I've seen with compiler optimizations and playing with AVX in Gentoo, but in Blender I matched a 3930k clock for clock with my FX 8350 when 3930k had default blender.org exe and I had a fully optimized set up. That, quite honestly, should never happen and it's a massive outlier, but it's proof that software optimization can make a huge difference.
If people are calling Jaguar a little behind IB in IPC (but it can't clock nearly as well, so I'm not saying Jaguar is close to i3 performance in general use cases), leaving it at an 8 core 1.6ghz IB, I wouldn't be surprised that if you were to compare SSE version of a game to AVX version of a game that a single Jaguar core running AVX code would be close to a 3ghz range IB core running SSE (< version 3).
I know this is a sensitive subject and it makes people heated but it's using different instructions so it's not exactly fair to say one is better than the other overall. So don't misunderstand and assume I'm saying Jaguar is better than IB or something, it's not. However if Jaguar is in an environment where it can have highly tuned code for AVX, FMA, etc and IB is in an environment where it's running code that's designed to also run on ancient hardware with a small instruction set, the Jaguar is at a massive advantage and people aren't giving it enough credit, mainly because they see both are x86 CPUs and they assume they are the same.