Originally posted by: Soroush
Originally posted by: MU_Engineer
The architecture of the 6200s is also designed for higher clock speeds and a lot of cores at the expense of per-clock and per-core performance.
I am not expert and usually look at Hz, cache size and number of cores when considering a CPU but you mention something new to me regarding per clock performance; what specifications and terms which describe this, educate me please.
Does your comment hold true in comparison of older 61xx Opteron and new 62xx; I mean do 61xx give better performance than 62xx ones despite having more core, cache and turbo(and I think AMD say they have about 30% more performance)? for example when I check prices for 6172 it is about $1,000 but for 6272 it is less than $550, so why people buy 61xx and not the cheaper and (supposedly better) 62xx?
Thanks for your time and sorry if this is noob question but I am going to build a 4 CPU workstation for video encoding using x264 and seek the best performance for price and have to make educated decision.
The newer CPUs do have about 30% more performance in multithreaded tasks if you compare old vs. new on a dollar for dollar basis using common programs and OSes. The 6172 may perform similarly to the 6172, but the 6272 is half the price. The 6172's dollar for dollar competitor is the 6276 or 6282 SE, which is notably faster. The 6180 SE costs more than any current Opteron 6200 and as a result, no 6200 can outperform it by a huge margin- yet. The 6272's Opteron 6100 competitor would be something like the 6134, which will be notably slower. The newer Opterons really shine in single-threaded tasks compared to the older ones due to the Turbo Core feature.
You also have to remember than the 6200s are brand-new, radically-different CPUs that need the software to catch up to them and the silicon manufacturing process to mature as well. Experimental tests on Linux using bleeding-edge compilers and kernels show the 6200s performing much closer to the 6100s on a clock for clock and core for core basis. The added cores and clocks of the 6200s means that when they are properly optimized for, the 6200s will be considerably higher performers than the 6100s.
I wouldn't hesitate to build a 6200-based 4P machine. Note that x264 doesn't really scale well beyond about 12 threads- you really only get an advantage from a 4P setup in x264 if you want to encode multiple videos at the same time. However, the 6200s have decent single-threaded performance due to Turbo Core, something that those of us who have 6100-based systems we use for general workstation usage really wish we had. My 6128s are good at multithreaded work like encoding, but they are fairly poky in anything that's not very multithreaded. I'll end up replacing the 6128s with some Opteron 6200s eventually for that reason.