8/16 may appear when ivy bridge e is introduced, but at a cost of $1000. Server chips are out of price range for most crunchers. With Intel wasting die space with IGP, I doubt well see even 6 core non enthusiast within that 10 year window.

While a teraflop is alot, its really not in the sense for real molecular dynamics. Exaflop is needed for.real time atom scale physics.

With 200 euro i can buy a Bulldozer X8 or a Sandy Bridge 8 core. They are great with HCC!!!


You're right, but some considerations:
1) More brute force calculation, more precision/quality of the software is needed. Cuda and OpenCl are young languages...
2) Real molecular dynamics are not the only simulations that can run on gpu....x-ray cristallography for example :-)
www.gpuscience.com is a great resource to see the state of the art on gpu scientific use....

Bulldozer isn't a power hog (I own one), power draw is slightly less than PHII series (I own some of them), even pushed to 4GHz. And to above they shouldn't be 200 Euros since they cost 175$ on the market over here, which should be ~135 Euros.

If you run a project that is integer heavy the AMD chip will completely overrun i7 processors, but around 30% slower with floating point projects.

CPU load comparisons:

http://www.techspot.com/review/452-amd-bulldozer-fx-cpus/page11.html

http://www.tomshardware.com/reviews/fx-power-...n-efficiency,3060-11.html

Efficiency for single-threaded apps:

http://www.tomshardware.com/reviews/fx-power-...n-efficiency,3060-12.html

My 3820@4.3 w/ 680, CPU fully loaded on WCG w/ one Einstein app uses 290W at socket. So 9 tasks total. Now the GPU is not being fully strained (basically at 60%TDP with one app).

With a 3820@ 4.3, I will be able to run 2 680's (second one on way) using a 750W PSU,with plenty of "safe" headroom.

Not trying to be a "Intel fanboy". I still buy cheap Phenom II's when they're around, like the 830 I picked up for $50. Just posting graphs. Compared to Intel, they are power hogs.

If you know which projects you like run are mainly/only integer, than it's not a bad idea since yes, they do have more integer cores. However, that shows in it's power consumption. No extra power is free.

try taking a core i7 and benching multicore efficieny alongside FX.

Take i7 and see how fast it is at 1 thread, then compare how well it scales to all threads.
Take FX and see how fast it is at 1 thread, then compare how well it scales to all threads.

You'll find FX scales significantly better than i7 to 8 threads, hyperthreading results in lost performance.

Individually, all of the bulldozers ALUs are incredibly fast, en masse of 8 ALUs it is well double that of i7 ALU performance, however the problem is that it only is capable of this if the program is properly coded and compiled with XOP instructions. Similarly, the FPUs in the FX are faster than all of the FPUs in the Phenom II X6 by around 50%, but only if the program is compiled with support for FMA4, in any other case only half the FPU is utlized at any given time. Since Intel owns most market share and many programmers employ the Intel compiler, not many people take into account that Intel compiler discriminates against AMD processors, and doesnt include support for AMDs instruction sets.

I also run a really old 750W PSU with 2 Radeon HD 7950s OCd by 50% with a FX-8120 using stock voltage OCd to 4.4Ghz, power consumption is perfectly fine, 10 less watts than my Phenom II 1090T at full load. Power consumption isn't that bad, after taking out the GPU power draw the system only uses around 180 watts. With the GPUs it is only an additional ~400 watts. Plenty of safe headroom there.

Point is, power consumption doesn't matter, it's a cheap CPU with a cheap motherboard, with an option to easily upgrade to a Piledriver if I have reason to.