Slashdot Mirror
AMD Phenom II Overclocked To 6.5GHz
An anonymous reader writes "During CES a group of overclockers with access to liquid nitrogen and liquid helium for the extra boost of coldness cooled an AMD Phenom II X4 chip to -232 degrees Celsius. Once they got the chip cooled to this frigid temperature, they pushed the clock speed all the way up to 6.5GHz, which is a world record for a quad-core CPU, and then dished out an astonishing 45,474 3DMark05 score!"
I was there, too. The coolest it got was approximately -242 degrees C; the warmest was approximately -218 degreesC, at least while I was watching.
The party was the XtremeSystems.org party at its LV headquarters, and it was sponsored primarily by AMD, DFI, Gigabyte, Cooler Master, and Thermaltake. It seems to me that Commodore had a presence there, too.
See ThinkComputers' blog for some more pictures (disclosure: my article).
Colin Dean Go a year without DRM
An slightly overclocked Core i7 965 (Extreme Edition) in a similar rig (in terms of video cards, etc) scored about 26,000 in the same benchmark (3DMark05).
So, no, they didn't have to go to liquid helium to be competetive, but going to liquid helium did allow them to set a world record (although I don't see any Guiness Book or other "official" information about this).
No x86s in this space. IBM has POWER6 running at 5 GHz.
http://www.dieblinkenlights.com
I don't know what weird kind of units you are using in your part of the world. But the rest of the planet is using Celsius for everyday temperature measures and Kelvin for scientific measures (same step size, different zero).
And on our scale, absolute zero (0K) is -273C.
Thus -242C (aka 31K) is pretty legal and possible temperature. (Although maybe not a very common one outside university labs and mad overclocker's basements)
Now please stop using Réaumur scale and start using what everybody else is using around.
--
PS: I checked, -242Ré is indeed impossible on Réaumur scale - 0 K is -218Ré
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
During CES a group of overclocker's
a group of overclocker's what exactly? Is it just me or is the correct use of apostrophe's [sic] starting to become a lost art these days?
Atom itself uses 1 to 4 watts. It's the chipset that sucks.
You're off by two orders of magnitude. 6.5ghz is 153 picoseconds per cycle.
120 characters isn't enough to explain it.
Deep space may be cold, but vacuum is a superb insulator. The chips can't be pushed hard without extensive and expensive heat sinks. Considerations on deep space probes are reliability and low power consumption, and there isn't a lot of need for speed. Reliability, radiation hardness, and low power consumption all have requirements that oppose speed.
Furthermore, since space probes take a long time to develop and use only very well established technology, they are using nearly-obsolete semiconductors by the time they're launched. They're really old when they get where they're going. It's not fast stuff by today's standards.
Contribute to civilization: ari.aynrand.org/donate
I am, my apologies. The end of a long day of physics. That gives a light-distance of 46 mm, or around 5cm. So I got the right final value, just dodgy working.
With pipelining, a signal does not have to travel across the entire length of the chip in one clock cycle. In modern processors, there are always several (usually 10-20) instructions in process at any given moment.
http://valid.canardpc.com/records.php
You got modded funny, but I think you were probably being serious.
I think the reason is that the newer 3DMark suites advanced so much in the realm of GPU-intensiveness, that to overclock a CPU and get a higher score without being GPU-bound, you have to go back to 2005.
Convert FLACs to a portable format with FlacSquisher
Ummm, if you do the rest of the math, I think you'll see that the electrical signals propagate sufficiently slower than the speed of light so that they in fact can't cross the chip in a single cycle.
Also, I think if you run all the numbers, you'll find that in 45nm chips at 3GHz, a signal can't cross the entire chip in a single clock cycle. Or if it can, it can't actually be used (ie. it can't participate in logic or be stored in a flip-flop or latch).
This has been a well-known problem in micro-architecture for a few years now. And is a contributing factor for why we've gone to multiple cores instead of higher performing single cores.
Strangely, the result states that the CPU was running at 4481 MHz.
As of this afternoon, they /do/ make chips that expensive, and more:
http://techreport.com/discussions.x/16298
Their new top-of-the-line chip:
Opteron 8386 SE 8 sockets max 2.8GHz 105W $2,649
Hail Eris, full of mischief...
E pluribus sanguinem