The Q6660 Inside
Honorable
MU_Engineer :
jdwii :
I'm just hoping with Steamroller Amd won't be sacrificing Single threaded performance for multithreaded performance. You people do know that even the Phenom has better performance per clock compared to Piledriver in many cases. If steamroller finally puts Amd around even the first gen I7 in performance per clock i'll be happy.
And keep the high clock rates to.
And keep the high clock rates to.
Single-threaded, per-clock comparisons aren't very useful with today's multi-cored CPUs with grossly different architectures and clock speeds between manufacturers and even different generations from the same maker. The performance of a particular chip in a wide variety of programs is really what counts as that's what people see in real-world usage. That means a mix of increasingly multithreaded programs with very different characteristics. It is very difficult to even get an accurate, repeatable single-threaded benchmark today since it takes specialized tools polling CPU registers to even know how fast the CPU is running at a certain period of time. And then as we have learned, first with Intel's HyperThreading and then later with the original Phenom's per-core frequency scaling and Bulldozer's CMT modular architecture, the OS's scheduler has a big hand to play in how well a chip performs with poorly-threaded tasks as well. The single-threaded, "instructions per clock" type of comparisons are really only for academic interest today. The real useful benchmarks are the "let's run a real program in a manner in which actual users might run it and then take measurements." The 1080p+ gaming benchmarks with high detail and a decent GPU are generally representative of this, and unsurprisingly, don't show a huge amount of difference between analogous CPUs of the two makers.
szatkus :
So they was talking about Opteron parts. According to this roadmap (http://community.amd.com/servlet/JiveServlet/showImage/38-2545-2550/AMD+Server+Roadmap.jpg) apparently they meant that some 65W Berlin will have 2x performance of X2150 (4 1.9GHz Jaguar cores). Nothing suprising and not too accurate.
That roadmap says a few things but doesn't make any mention of the performance of the Steamroller APU vs. Jaguar. All it says is that uniprocessor will move from AM3+/the 4M Piledriver die to FM2+/2M Steamroller, Jaguar SoCs will be replaced with ARM units, and that there will be a Piledriver refresh on C32/G34 but no Steamroller.
I am a little curious what the Piledriver refresh yields on C32/G34. They are keeping the same sockets so I doubt there will be any change in I/O, they are not increasing module count, and I also highly doubt there will be a platform refresh since supposedly Piledriver is the last arch supported on C32/G34 before they move to the new "GC36" socket with on-die PCIe and such. My guess is it's a new C stepping with slightly higher clocks and maybe a few small tweaks to things like caches, FPUs, and branch predictors to subtly improve performance. I suppose they could boost the L3 cache size but I haven't heard of the cache size being a significant issue in these chips.
Perhaps a small clock increase and boost in L2/L3 cache? The boulder GC36 CPUs seem to be the real next step up.