Bobcat core was approximately half a bulldozer module
- 2 ALUs and 2 AGUs per core (ALUs handle majority of instructions)
- 2 64 bit MMX/FP pipes => bulldozers 2 128-bit FMACs (FP) and 2 MMX pipes
Now jaguar is theoretically equal or superior for FPU/SIMD vs. even steamroller
Jaguar - 2 128-bit pipes per core
Steamroller - 1 128-bit pipe per core (FPU pipe 1 or MMX/FPSTO pipe)
Jaguar - 2 128-bit pipes per core (1 pipe is ADD other is MUL though)
Steamroller - 1 128-bit pipes per core
Originally Posted by kevinf
I did the whole Aida/Everest benchmark suite, as well as the full "cache and memory" benchmark. Again, I can't post the results, but here is a summary.
The only non-surprising benchmark was memcopy/read/latency, as the G-series is single-channel vs trinity dual channel.. both systems had CL11 4GB DDR3-1600 ram with all 4 cores enabled.
Some of the CPU benchmarks were +- 10%, AES was 50% faster. FPU on average 30% faster. (in favour of G)
OpenGL/OpenCL/Shader performance on the other hand... the trinity GPU will win, hands down... 8330 vs 7660G
This is the greatest resource of specifications in regards to AMD APUs... (I contribute to it as well).. I advise you to do the same