Slashdot Mirror
A 4.1 GHz Dual Core at $130?
joshmo97 writes "Tom's Hardware has found that the Pentium D 805 runs stable at 4.1 GHz and outperforms Intel and AMD's flagship offerings in many benchmarks. From the article: 'The Pentium D 805 is a budget CPU, but it puts lots of processors from AMD and Intel to shame. Although it is not based on the latest 65 nm core, this CPU remains stable even when operating at amazing 4.1 GHz. The Pentium D 805 ascends to the throne as the new King of overclocking, knocking out the AMD Opteron 144.'"
Let me guess, a hand picked review sample? the best of the best sample chip?
09F91102 no, 455FE104 nope, F190A1E8 uh-uh, 7A5F8A09 that's not it, C87294CE no. Ah! 452F6E403CDF10714E41DFAA257D313F.
It would still be pretty cool if you could use speedstep or whatever to run at a lower speed and lower power most of the time, and then crank it up for a little while when you need the speed, kind of like nitrous injection. This is actually the best automotive metaphor yet because nitrous causes additional heat to be generated :)
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Raise your hand if you've "burned out" a chip that ran stable but was destroyed by overclocking. My Celeron 566 still runs 24/7 at 850 mhz after all these years.
Actually I've been running my old Linux server underclocked for a couple of years now. Athlon 1.4ghz, running at 1ghz. It's much cooler, uses less power, and solid as a rock - even though it lives up in my attic. Apart from a bit of web/database development, and archiving stuff once a week, it really doesn't need much horsepower to serve files and route. Couldn't get it to boot at a lower clock speed though.
Code, Hardware, stuff like that.
You could have just put a 1Ghz Duron in it. They're usually really cool (ironically enough because they're a product of a refined process so they're usually better than the pricier older cores) and super hella cheap, or were.
If you are really concerned try hunting down a 1G duron.
Tom
Someday, I'll have a real sig.
Factor in cooling costs for wherever you put that damn thing.
You can't deduct points for being "elite" though.
Tom
Someday, I'll have a real sig.
Ok, you remember how easy the Celeron 300 overclock to 450 was? Well, this is just as easy. And THAT's why it's news.
You don't have to take it to the M4d extreme and go to 4.1ghz. If you have a good motherboard and some good memory, you can up the FSB on your motherboard and easily get to 3.2ghz (from the stock 2.6ghz). And you don't need to touch the voltage to do it or watercool. Just change the setting and there's a damn good chance you won't have a problem.
It's a $130 processor that, with a minor tweak, can perform like a $1000 processor.
You got lucky, that's all. Some chips barely meet 300mhz and some chips actually rate 500mzh but get stamped as 300mzh because they needed more 300's. There is no way to tell which of these you get when you bought your Celeron 300. If you got the barely-300 chip and overclock it, you fry it. If you got the 500 chip, it runs great.
It's been known for a while that AMD's Opteron 165's ($330) will o/c to at or above the performance of an FX-60 ($1000).
All this article really says is that the ultra-high-end isn't worth it.
I'd like to see a comparison between this thing and the Opty-165 o/c. The Pentium may be a bit cheaper, but factor in the power and cooling bill and (I expect) the higher performance of the Opteron, and it's probably about even.
It's pretty hard to create a problem like this. The manufacturers (with a few notable slipups) test and rate the CPU up to some given temperature. Intel puts a couple of thermal diodes in each chip, to shut the machine down if you exceed the range within which it's reliable. In addition, quite a few parts of the chip have self-checking going on all the time. For an obvious example, cache typically occupies at least half of a modern chip, and a Pentium D's cache error detection/correction coding attached to the data.
In addition, consider what your numbers would mean. An Intel P4 typically executes close to 2 instructions per clock. Just for the sake of argument, let's assume it takes the machine 20 seconds to boot. 4.1 GHz x 2 IPC X 20 seconds gives 16.4 billion (read milliard, if that word's in your vocabulary) instructions just to finish booting.
Since one billion (milliard) is 10^9, that means if your numbers were correct, it would have to survive around 16 incorrectly executed instructions JUST to finish booting. To run a single hour, the number of incorrect instructions would run into the thousands...
The universe is a figment of its own imagination.
You'd be surprised how much "wrong" can happen before you crash. I had faulty ram a while back [well more so it had the wrong timing] and it would boot, run for a while then randomly something would segfault. Then the kernel would panic and lock up, etc.
A simple op like
MOV EAX,[EBX+13]
could excute as
MOV EAX,[EBX+14]
and not result in a significant problem.
As for the self-checks and diodes. You don't have to overheat a circuit to kill it. Over volting a transistor can denature it and you'd never notice. Just like ESD could "partially break" a circuit.
In fact if you looked at a comp lab with open computers chance are at least one IC has some form of ESD damage.
Tom
Someday, I'll have a real sig.
Think about what the whole point of dual core was.
Chip heat output increases exponentially with clock speed. Heat output means power consumption and loud fans and... well, heat.
Everybody was bitching about heat output of chips (OMG Pentium 4 is teh sux0r is sooo h0t)
Take two cores. Essentially underclock them. Now each runs at less than half the heat output (remember that exponential heat curve.... ) and you have two of them so you have more total computing power in a SMP configuration.
Now somebody overclock it and they get really awesome performance... of course, because the cores were essentially underclocked in the first place. But now their heat output is terrible because they abandoned the basic advantage of dual-core: the performance per watt ratio.
"That sort of underlines the problem with the main point though--any other hardware can go anyways, why is an overclocked CPU of particular concern?"
Because you routinely have to overvolt something to run faster.
Here's EE 101 for this topic...
The frequency is limited by the switching frequency and length of the critical path. The longer the path the more time it takes to charge the wires to get stable transitions. So the remedy is to raise the voltage (hint: think VIR triangle).
So if a circuit (or an instance of the circuit) is rated at say 1.3v @ 2.66Ghz then to run it at 4Ghz you probably have to raise the voltage (unless the limitation to 2.66Ghz was artificial).
Now that we are raising the voltage it's not hard to imagine why it could break the processor.
As for your comment between 3 vs 3.2Ghz processors there are important things you're missing. Processor production is NOT exact. Not only can a run of processors on the same wafer behave differently (hint: is yield 100%?) but between runs the actual process (chemical makeup) can change as they optimize the process.
So no, a 3.2Ghz processor even though it's probably made from the same process that makes 3Ghz shouldn't be suspect.
Tom
Someday, I'll have a real sig.
Far more interesting is overclocking a monitor - the increased frequency can make the flyback transformer more efficient, increasing the final anode voltage by lots of kilovolts. I'll leave you to imagine the results. Don't try this at home, kids.
What would Lemmy do?