Overclockers Top 6GHz With A 3.6GHz-Rated P4

sH4RD writes "The 6GHz barrier has been broken by two guys, a little LN2 (liquid nitrogen for those not as chemistry inclined), and an Intel Pentium 4 (Prescott) 3.60GHz. Check out some icing and some proof of speed. Better yet take a look at how fast it calculates pi. Also be sure to check out the original announcement."

Comments

Erm...

[__aavhli5779]

I sure hope they were using Gentoo, because if not they couldn't take advantage of those incredible speeds with some hot -O3 -funroll-loops action :P

In all seriousness, this is pretty amazing, but I can't really see the usefulness. For sheer geek pride, sure, why not? But as far as I can tell the expense involved outweighs any gain in performance; for probably half of what these poor folks spent getting a P4 to run stably at 6 ghz (and it doesn't even sound super-stable from what I've read) they could've probably bought a couple more CPUs and had a proper SMP system instead. Regardless, I admire their tenacity and mourn for the warranty on their poor CPU :P

Re:Erm...

[bizpile]

But as far as I can tell the expense involved outweighs any gain in performance;

Isn't that what being a geek is all about? It's a "becuase it's there"-type of thing.

Re:Erm...

[Anonymous Coward]

In all seriousness, this is pretty amazing, but I can't really see the usefulness. For sheer geek pride, sure, why not?

Oh for crying out loud. Nearly every article on Slashdot falls into that category. As does Slashdot itself. It's Saturday night, lighten up.

Re:Erm...

[julesh]

You need to learn a bit more about why caching works before making statements like that.

Let's assume, for the sake of argument, that main memory is 10 times slower than the slower cache, and that half the cache size means twice as many cache misses (the real figures are actually worse than this, but I don't know what they are exactly).

So, we have:

Main memory access time: 20 units
Small cache access time: 1 unit
Large cache access time: 2 units

Over the course of 100 memory access, lets say 10 of these are misses with the large cache, and based on our assumptions above 20 are misses with the small cache.

Small cache total time: 20*20 + 80*1 = 480
Large cache total time: 10*20 + 90*2 = 380

So, in this situation the large slow cache clearly performs better. In other situations, you might be able to make the small fast cache perform better (e.g. lower number of cache misses, presumably due to applications with smaller working set sizes, or faster main memory might help here).

Tops 6ghz?

[Anonymous Coward]

So it tops 6 GHz, but they only calculate pi at 5.4 GHz? Sounds like the only thing it can run at 6 GHz without crashing is CPU-Z...

Re:Updating Wikipedia Now

[Anonymous Coward]

You know what? people who say "they have way too much time on their hands" or "how do they have the time to do that" .. REALLY piss me off

Cause just about always, people who say that never did a damn progressive thing ever in their life. It's just a lame excuse underacheivers make trying to pull down those who are out there making stuff happen.

Look at yourself first .. Browsing slashdot .. how come you have time to do that? Like what exactly do you do with YOUR time? I dont see you being the CEO of a major corp or contributing anything to society.

Re:Cold!

[_Pablo]

If Intel hadn't decided to kill P4 in favour of PM then we may have had to do it sooner rather than later!

It would be amazing to have to use LN2...but then again since I first stuck my finger on top of my 68000 and realising it was a lot hotter than my 6502 i'm constantly amazed how hot these things are getting.

Sure.. that's cool.. what about...

[Anonymous Coward]

the warranty? I'm sure the heat and added radiation just nuked their ability to have kids. Guess we can't pass on great OC'ing skills through the gene pool 8-)

Re:Throughput not clock speed

[PureCreditor]

the physical speed limitations of hard drives (sustained read/write, not SATA) is the real wimpy part of the bandwidth....

when running multi-gigabyte SQL queries (at work, our entire RDBMS is about 1TB), the crawling speeds of the hard drive is evident. the time it takes to develop the SQL query and the time it takes to run it are comparable (btw, the queries are okay optimized)

6GHz might be useful for 3D rendering jobs or obsessive gamers, but for the bulk of the business world, the HDD is still the pain in the a**

This is the type of thing.....

[Sean Johnson]

that helps to perpetuate the "MHz myth". If MHz don't matter, why are these guys doing these crazy things to increase the MHz? I know it is just for the "fun of it", "to see if it can be done". However,(tongue-in-cheek) this stuff just influences people to rely on MHz numbers more and more. It teaches young-ins that more MHz is better whatever the cost. What we need is a great story about how Bill Buxley and his pal Jan Hammy had strung 32 CPU's together with chicken wire in thier garage. This would be the parallelism hack equivalent to overclocking. Pretty soon though we would have to contend with the "parallelism myth" and the industry would in turn be trying to deemphasis parallelism for Mhz. It would be a cycle in that manner until finally one day the industry hits it big with the "quantumn computing myth". Stay tuned if your still alive by then. LOL!

Re:This is the type of thing.....

[RzUpAnmsCwrds]

MHz *does* matter. A 3GHz Opteron should be 2x as fast as a 1.5GHz Opteron (of course, that doesn't take into account the rest of the system - memory bandwidth, disk bandwidth, etc.)

The "MHz Myth" referrs to the fact that MHz is a poor metric to compare CPUs with. It's fair to compare a 3.2GHz P4 Prescott to a 3.6GHz P4 Prescott and expect that the 3.6GHz chip will be faster. What doesn't make sense is to comare a 3GHz P4 to a 2.4GHz Opteron and claim that the P4 is faster.

Re:6 GHz is not that impressive.

[PureCreditor]

nearly any other CPU architecture running at 6GHz will muder the P4's performance in either integer or float-point, partially due to Prescott's insanely long 31-stage pipeline and relatively week parallelism

also, the P4 excels only in programs so small it can fit in the really small L1 cache. AMD's L1 cache is the really juicy one. =)

Intel designed the Pentium 4 solely around marketing's requirements of consumer hype instead of sound technical choices.

Re:Yksi Kaksi Kolme (or however you spell it)

[Anonymous Coward]

"Laskenta Alkaa"="Calculation Begins"

Laskenta=Calculation
Alkaa=To begin, He/she/it begins

Airport has nothing to do with it.
Airport=Lentoasema
Lento=Flight
Asema=Stati on
So in a way Airport in Finnish is "Flight station"

So now you know :)

Re:6 GHz is not that impressive.

[Wtcher]

The DRAM may not be able to fetch that fast, but an increased bus rate will still provide some mediocum of bonus performance. Still, I notice now that the FSB clock was not increased at all.

So long as some effective pre-fetching is in place, there's going to be some asymmetric performance gains. It won't be 1:1 since the memory isn't actually getting any faster, but it doesn't matter so long as the data is where it needs to be when it's needed. In any case, the fact is having the CPU far outstrip the performance of the rest of the system isn't a particularly unusual occurrence.

You silly mods...

[Anonymous Coward]

...at least RTFPP before modding the child Insightful.

When the child post reiterates the caveat of the parent post it's not insightful, it's redundant!

Re:Hang one outside the ISS !!

[evilviper]

A vaccuum is a very good insulator... Fortunately, space isn't even remotely a perfect vaccuum.

Re:Don't forget the dual clocked ALU

[Bender_]

Interesting, how is it synchronized with the rest of the pipeline? Is there any publication about this?

Re:calculate pi...

[Stuart Gibson]

Memorising pi is easy, provided you count in base pi.

Stuart