MrKermit has brought in 300 years in his first 9 months! In those 9 months MrKermit has hyperjumped to 29th place. A that "speed", in another 9 months he will have put on as much time as ddtung has now in 16th place. ddtung is moving fast too, but not as fast. Well have to wait and see. Maybe MrKermit will comandeer another datacenter.

MrKermit, your awesome!


Cheers

Been here 3 years today:
Total Run Time 18:211:22:08:32 (#1,556)
Points Generated 16,650,241 (#1,030)
Results Returned 28,551 (#1,267)

Astrolab,

Thanks Hypernova,

It is interesting that one single 6 core CPU could now do 18.5years in 1.5years, using HT!
Two i7-980X CPUs would take 9 months, or even 5 atom 330s in 11months, l0l
It's strange to think that in 10years our efforts may look insignificant (MrKermit aside) CPU's will probably have 1024 cores by then, and this could primarily be a GPU project with cards offering up 4K to 16K cores

Perhaps my account will have been tombstoned by then

I think that you are absolutely right. There is no doubt that we are at a turning point or in transition phase between performance increase by frequency increase versus number of cores increase.

It is also not clear what will happen to the CPU versus GPU paradigm. I see in the future a moment where we will have one CPUGPU chip that will have merged everything. If you have hundreds of cores then you allocate the cores according to the needs. That chip will have the additional graphic specific function beside all the cores. The latest Fermi chip may have 3 billion transistors, and the 980X has under 2 billion transistors for 6 cores and 12 threads.
With still shrinking engraving technology we may well be at 10 billion in ten years and this is enough transistors to do everything in a very performant way.

More cores are useless if there's not a fat enough pipe to keep them busy. We routinely have to "undersubscribe" our quad core systems because the memory bandwidth is so bad that involving the addition cores slows things down. With 48 Core Proc's running actual code in labs, I can't even imagine how useless they will be for many of my apps. Pretty soon the notion of, "Flops are Free" will be muted only by the notion of "Unless they need Data".

Building Networks on chips to allow the cores to collaborate directly is interesting, but it's still a network with minimal bandwidth compared to what the flops could consume.

I think I'll go hug my dual core nodes now :)