best CPU for crunching? Q6600? E8400? Others?

[Foxery]

wired: The Core architecture crunches Folding@Home about 2.5x faster than a P4 at the same clock speed. My home machine puts my 3 borg boxes at the office to shame, even before overclocking.

 

[Amaroque]

Originally posted by: wired247
Purely for the sake of argument....

what would be faster at crunching?

4x P4 3.0GHz with HT (4 computers)
1x Q6600 running at stock clock

?

Even though I dont particulary like the P4, I'd have to say four P4's would most likely win.

 

[imported_wired247]

I don't have any spare CPUs lying around that aren't folding right now , but I am clueless when it comes to how different architectures "get more done" per clock cycle, which seem to be the whole argument about AMD versus intel thing from before C2D's became mainstream...


Is it proportional to # of transistors? Or just how intelligently the architecture is designed?

 

[Insidious]

I think it's about how they're designed.

I know for a long time AMD could do floating point calculations much more effeciently than intel and that tended to make AMD better for certain jobs (like gaming).

I know that for folding, there was a big difference as Intel introduced new instruction sets (SSE, SSE1, etc...) and the folding folks wrote software specifically designed to take advantage of those new instructions.

So maybe part of the answer is that people can identify the differences in processors and then write code designed to bring out those strengths (or weaknesses) as it suits their adjenda.

Right now, Core2 is the design that everyone seems to be able to make look good with their software. I know my new PCs game a lot better than my old AMD dual cores, they overclock like crazy (even better than my old Opterons) and they are relatively cheap for a "new" processor.

It's always changing (thank goodness.. that helps keep prices down) and you will only have the "best" PC for a few weeks or months at best no matter what you build.

-Sid

 

[PCTC2]

Originally posted by: Fullmetal Chocobo
I'm not sure about the E8400, but I stress tested (crunched) a Q6600 that I built for IT, and that Q6600 destroyed my E6400. And if there are any specific projects that won't take advantage of a quad-core, you can usually run multiple instances to accomplish the same thing.

I'm not so sure, Fullmetal..

I have my E6400 @ 3.2GHz and my Q6600 @ 3.0GHz both running the same project on SMP.
E6400: 1760ppd, 1400min project.
Q6600: 1760ppd, 1100min project.

Sure, it's 5 hours difference, but dual to quad, that's not bad.

 

[Insidious]

Originally posted by: PCTC2

Originally posted by: Fullmetal Chocobo
I'm not sure about the E8400, but I stress tested (crunched) a Q6600 that I built for IT, and that Q6600 destroyed my E6400. And if there are any specific projects that won't take advantage of a quad-core, you can usually run multiple instances to accomplish the same thing.

I'm not so sure, Fullmetal..

I have my E6400 @ 3.2GHz and my Q6600 @ 3.0GHz both running the same project on SMP.
E6400: 1760ppd, 1400min project.
Q6600: 1760ppd, 1100min project.

Sure, it's 5 hours difference, but dual to quad, that's not bad.

That 11 minutes is what I get on each of TWO SMP work units on my Q6600 @ 3.45HGz

Correct my math, but that's a smidge more than TWICE as much. not my definition of close.

-Sid

 

[dajeepster]

Originally posted by: Insidious

edit:
There is NOTHING that the P4s can do better than the Core2 architecture as far as I know.

WRONG... the P4s make a much better paper weight than the Core2s

 

[TallBill]

Originally posted by: VirtualLarry
I bought two Q6600s, I came to the conclusion myself that quad-core would be better (more brute horsepower) for DC apps.

I agree, my two Q6600's are beasts. Unfortunately at this time I can't buy any more horsepower. It'll be a few years.

 

[Strebor]

Purely for the sake of argument....

what would be faster at crunching?

4x P4 3.0GHz with HT (4 computers)
1x Q6600 running at stock clock

I've got 2 P4 630s at 3.0GHz, and a Q6600 at 3.0GHz.

PPH is 16 on each of the P4s, and 80 on the quad.

Put the quad at stock and double the P4's and you get 64 PPH on each...

 

[Assimilator1]

Only 80 ppd on the quad? ,that's way under ,it should be in the thousands! ,or have I mis-read something?

Originally posted by: Insidious
I think the P4s would be MUCH better.........

So, can I have your Quad?

-Sid

(The P4s are pittiful for the client I like, so I guess it's the Quad for me)

edit:
There is NOTHING that the P4s can do better than the Core2 architecture as far as I know

Yes their is ,use more electricity:laugh:

Amaroque
Re 4xP4 vs Q6600 ,no the Q6600 would toast the P4s.
Even in single threaded apps clock for clock Core 2 is anywhere from 50-100% faster than a P4.

 

[Strebor]

Sad isn't it , meant pph.

 

[Fardringle]

Originally posted by: dajeepster

Originally posted by: Insidious

edit:
There is NOTHING that the P4s can do better than the Core2 architecture as far as I know.

WRONG... the P4s make a much better paper weight than the Core2s

They also make better space heaters.

 

[biodoc]

Originally posted by: PCTC2

Originally posted by: Fullmetal Chocobo
I'm not sure about the E8400, but I stress tested (crunched) a Q6600 that I built for IT, and that Q6600 destroyed my E6400. And if there are any specific projects that won't take advantage of a quad-core, you can usually run multiple instances to accomplish the same thing.

I'm not so sure, Fullmetal..

I have my E6400 @ 3.2GHz and my Q6600 @ 3.0GHz both running the same project on SMP.
E6400: 1760ppd, 1400min project.
Q6600: 1760ppd, 1100min project.

Sure, it's 5 hours difference, but dual to quad, that's not bad.

My q6600@3.2 GHz will do 4500 PPD (2 x SMP, 2653 Wus, 64 bit ubuntu linux).

A fair comparison would be 1 x smp for the core 2 duo and 2 x smp with the q6600 using 64 bit linux. Stanford's implementation of windows version of SMP is pathetic. The later will not take full advantage of the q6600. The affinity changer does help quite a bit though.

 

[Markfw]

Originally posted by: LOUISSSSS
my Q6600 @ 3.6 gets me 3520PPD in F@H with dual SMP clients. what does everyone's e8400 get?

Mine Q6600's@3.5 get 4500 ppd Are you running affinitychanger ?