Hello Brandon Murray,
The last 10 years have been complicated as far as floating point goes. The modern floating point format was formally proposed at Kansas City in the mid-1970's and adopted by a chip company named Intel who electrified the community with the 8087 math coprocessor. I seem to recall it did about 40 Kflops double precision. Eventually the format was adopted as IEEE 754-1985. It takes a long time to adopt a standard.

Having started as a hero, Intel became everybody's favorite whipping boy when they never updated the 8087. They produced variations that would cooperate with newer CPU chips (80287 with 80286; 80387 with 80386) but were easily out-computed by every competing math coprocessor. This finally began to change with the Intel 80486 DX which combined the CPU with the FPU (Floating Point Unit) on the same chip. With the Pentium progress sped up and by the turn of the century, modern CPUs were delivering supercomputer speeds, beating the early CDC supercomputers of the early 1970's. Once computers started to be measured in Gflops, we had reached the modern era that WCG is flourishing in, running science programs on home computers rather than on giant computers at national laboratories.

Progress since then has been - - peculiar. Standard CPUs have slowly doubled and in some cases redoubled their speed while also adding cores per chip. Since most people are oblivious to FPU speeds, I have noticed that floating point is slower on some chips produced since 2011. It has become a commodity that sometimes plays second-fiddle in CPU designs.

The joker is the GPU. Since 2007, general-purpose GPUs (GPGPUs) have been built which stream floating-point on the GPU board. The problem is the storage and retrieval of the floating point results, which require peculiarly limited algorithms although they can run superfast. There is a rainbow evolutionary development of different types of chips and boards, to see which can find or develop a profitable market. I do not doubt that some architectures will be successful and it will be interesting to see what sorts of algorithms and programs they can run.

Lawrence

The interesting thing is that our (WCG crunchers) current performance is 700 TeraFLOPS and the peak (during HCC1 GPGPU phase) was 1.5 PetaFLOPS

Cheers

I remember the days I've read about the pentium 60 and 66 transistor count in 1994- 95. it was 3,3 millions which fasinated me. this summer I came accoss an artical about the Nvidia Titan. I remember something like 7 billions. along with the stuctural changes it's not suprising to see such leaps in the GFlops numbers..
I am chrunching with an 8 core I7 laptop( on ubuntu linux), The performance is out of my finest fantasies when I started crunching back in 2007..
It2s a wonderful feeling inside knowing that we let all this power to eat away the epidemic problems of the humanity..
Let's crunch on :)

[ot]You're reacting to someone who returned 2 results, the last 38300+ hours ago [shortly after joining].

Yes, I checked it prior to my reaction, but posted it anyway

Cheers