PDA

View Full Version : Equating Flops to Hz. Possible? + home supercomputer news!



Flying_Nutcase
02-16-2007, 06:28 AM
Hi,

Well first, I read in the paper how Intel has figured out how to get 'supercomputer' speeds into our home computers.

"Photo-realistic 3-D games...would also be possible."

"I'm confident [you will be able to buy teraflop computers] in five years or less, maybe significantly less," said Justin Rattner, Intel chief technology officer.

http://www.business-standard.com/iceworld/storypage.php...&autono=274467&tab=r (http://www.business-standard.com/iceworld/storypage.php?leftnm=8&subLeft=3&chklogin=N&autono=274467&tab=r)

I'm wondering how that compares on the GHz scale? Is there a meaningful conversion?

With a bit of luck this will be the quantam leap we've all been hanging out for. Imagine what Oleg could do if the shackles of our current computers could be cast aside. http://forums.ubi.com/images/smilies/16x16_smiley-happy.gif

Flying_Nutcase
02-16-2007, 06:28 AM
Hi,

Well first, I read in the paper how Intel has figured out how to get 'supercomputer' speeds into our home computers.

"Photo-realistic 3-D games...would also be possible."

"I'm confident [you will be able to buy teraflop computers] in five years or less, maybe significantly less," said Justin Rattner, Intel chief technology officer.

http://www.business-standard.com/iceworld/storypage.php...&autono=274467&tab=r (http://www.business-standard.com/iceworld/storypage.php?leftnm=8&subLeft=3&chklogin=N&autono=274467&tab=r)

I'm wondering how that compares on the GHz scale? Is there a meaningful conversion?

With a bit of luck this will be the quantam leap we've all been hanging out for. Imagine what Oleg could do if the shackles of our current computers could be cast aside. http://forums.ubi.com/images/smilies/16x16_smiley-happy.gif

stathem
02-16-2007, 06:33 AM
http://forums.ubi.com/images/smilies/25.gif

FLOPS (http://en.wikipedia.org/wiki/FLOPS)

NonWonderDog
02-16-2007, 07:56 AM
There is absolutely no conversion, but the number of floating point operations per second could be approximated by multiplying the frequency by the number of transistor pathways that can do calculations, and then dividing by the amount of time it takes to do each calculation.

Supercomputers get huge FLOPS ratings because they usually have networks of thousands of processors. The individual processors are usually slower (=cheaper) than the newest consumer processors (although the newest supercomputer actually uses off-the-rack Xeon 5060s http://forums.ubi.com/images/smilies/53.gif), but that really doesn't matter when there are 10,000 of them.

I don't know what the Intel guy was on about. Maybe there's been some breakthrough, maybe not, but "supercomputer performance" is obviously a moving target. Teraflop computers within 5 years would be predicted by Moore's law anyway, and there's ALREADY a petaflop supercompter. Two X1900s in crossfire mode will supposedly do ~1 Teraflop, too--graphics are VERY floating-point intensive, so GPUs can do more floating point operations than CPUs.