Conroe 2.4GHz benchmarked

[FelixDeCat]

OH. MY. GOD!

Good job Intel. :thumbsup:

 

[n19htmare]

Still no word on how well these will OC, just assumptions. can't wait.

 

[Cooler]

That cpu score is very low.
I get 6400 CPU score on much older Intel CPU 840EE. I dont think the dual core +HT is the reason.

 

[Markfw]

Originally posted by: Zap

Originally posted by: Duvie
So in conclusion once those guys stop hand jacking themselves over worthless benches and start running real world apps ...LET ME KNOW

Damnit Duvie, you know what kind of image I have in my head now? :|

I look forward to faster products because it drives down pricing... generally.

BTW, there was a Conroe ES for sale on eBay already, from a supposed legit seller in Malaysia that often sells ES chips.

Well, I can;t find it. "conroe" even in all categories retrieves nothing relevant.

 

[Absolute0]

For all the impatient whiners, keep in mind that Conroe is new, and isn't really supported. These guys have been working for MANY hours with software glitches and problems, you should be thanking them, not complaining about how long it takes to run benchmarks...

To explain why we don't see all the benchmarks, it's not running perfectly because it isn't fully supported.

And to explain the SP32M score, if anyone has run a 32M it takes a lot of memory, Victor has only 512mb of system memory, that slows it down... anyway, 24minutes is faster than an FX60 at 3 Ghz.

 

[FelixDeCat]

Originally posted by: Cooler
That cpu score is very low.
I get 6400 CPU score on much older Intel CPU 840EE. I dont think the dual core +HT is the reason.

I just :heart: the MM # of 120,000!!! :shocked:

vs.

820/x2's 40k (one third Conroes) :shocked:

Im drooling all over my simulated 5cm keys on my $600 Ipaq's virtual keybord.

 

[Absolute0]

Originally posted by: Cooler
That cpu score is very low.
I get 6400 CPU score on much older Intel CPU 840EE. I dont think the dual core +HT is the reason.

Victor fixed the CPU score, it's now 8279, with a 6600LE no less

http://www.xtremesystems.org/forums/showpost.php?p=1376872&postcount=476

 

[TrevorRC]

http://vic.expreview.com/attachment/pi_32m.jpg

23 minutes.

Holy !@#% anyone?

Can only imagine what'll happen with a bit of overclocking

 

[BrownTown]

that 23min time is actually slower then what it sholuld be, the 32M test wants to use more memmory then he has, so its getting bottlenecked, it would likely improve considerably with 1gig of RAM.

 

[Markfw]

Intel has shown in the past to excel at synthetic benchmarks, and this isn;t even supported or hardware you can buy, so again, its only an indicator, that its probably better than the P4.

 

[996GT2]

Hopefully they'll do more real-world tests once they finish the benchmarks, and run them against dual-core AMD processors also.

 

[Faikius]

Originally posted by: Markfw900
Intel has shown in the past to excel at synthetic benchmarks, and this isn;t even supported or hardware you can buy, so again, its only an indicator, that its probably better than anything AMD will field in the next year.

Edited for accuracy.

 

[Cooler]

god i hope there will be bios update for my mobo so I put one in next year. I just love my current mobo.(has most oc options for intel)

 

[Keysplayr]

Originally posted by: Hard Ball

Originally posted by: keysplayr2003
Doesn't the M in SuperPi tests stand for "million decimal places" and not MegaBytes?

1M = 1 million decimal places and not 1 Megabyte?

4M = 4 million decimal places and not 4 MB?

etc.

etc.

???

I am actually asking this question, not stating it as any factoid. Some folks who seemed knowledgable over at XS mentioned this in correcting others who though M stood for Megabyte in SuperPi.

try this:
>>> import sys, math
>>> dir(math)
['__doc__', '__file__', '__name__', 'acos', 'asin', 'atan', 'atan2', 'ceil', 'cos', 'cosh', 'degrees', 'e', 'exp', 'fabs', 'floor', 'fmod', 'frexp', 'hypot', 'ldexp', 'log', 'log10', 'modf', 'pi', 'pow', 'radians', 'sin', 'sinh', 'sqrt', 'tan', 'tanh']
>>> pi = 48 * math.atan(1/float(49)) + 128 * math.atan(1/float(57)) - 20 * math.atan (1/float(239)) + 48 * math.atan(1/float(110443))

Thanks Mr. Wizard, but I have no clue what any of that means, nor does in answer my initial question (in any way that I could understand at least). But I appreciate your enthusiasm.

So, M does indicate "million decimal places" then and not "Megabytes". So then all the arguments about fitting a 1M SuperPi run completely in it's cache is somewhat bogus. As the gent above stated, a 1M SuperPi run requires 8MB of memory to run the calculation to 1 million decimal places. And even the Conroe with 4MB of cache will have to hit the system memory even in the 1M test.

 

[BrownTown]

even a gigabyte cache will have to hit the memmory on compulsory misses, but running an 8MB program in a 4MB cache is gonna cut capacity misses to very very low. Its probably hitting its L2 99+% of the time.

 

[n19htmare]

Originally posted by: Faikius

Originally posted by: Markfw900
Intel has shown in the past to excel at synthetic benchmarks, and this isn;t even supported or hardware you can buy, so again, its only an indicator, that its probably better than anything AMD will field in the next year.

Edited for accuracy.

HaHa...

I still don't know what people are complaining about. He's pretty much running all the benchmarks people are asking him to run. And the Conroe seems to excell in all at it stock speed.
I don't think it'll be disappointing in real world apps (what ever those might be in your definition)

 

[BrownTown]

yeah, people are asking him to run benchmarks and hes doing it. I asked for the CPU-Z latentcy test and he did it like an hour later. (3-cycle L1, 14-cycle L2 exact same as Yonah). He's pretty much posting every benchmark he can. Games would be worthless since he cant get a decent video card to work in the motherboard. Plus, he only has 512MB of RAM, so these benchies arent exactly on a top of the line system anyways.

 

[xenolith]

LOL. I just think it's pretty funny how all the peeps keep intermittently crashing the XS site today.

It's been a while since I've seen so much excitement. And we still have three months to go for Conroe launch, right?

 

[BrownTown]

well, when was the last time a new architecture launched? K8, and that was a few years back. New architectures don't exactly come around very often, and when a new one comes around and its pwning the living daylights out of everything else of course people are gonna be excited.

 

[Cooler]

Originally posted by: BrownTown
well, when was the last time a new architecture launched? K8, and that was a few years back. New architectures don't exactly come around very often, and when a new one comes around and its pwning the living daylights out of everything else of course people are gonna be excited.

Well nothing really new has been out since P6 as they are all based on it with its out of order design.
Netburst added trace cache and more pipes quad pumped bus
AMD added a Memory controler
Conroe added Shared L2 with 4 issue design and less pipes

 

[Hard Ball]

Originally posted by: keysplayr2003

Originally posted by: Hard Ball

Originally posted by: keysplayr2003
Doesn't the M in SuperPi tests stand for "million decimal places" and not MegaBytes?

1M = 1 million decimal places and not 1 Megabyte?

4M = 4 million decimal places and not 4 MB?

etc.

etc.

???

I am actually asking this question, not stating it as any factoid. Some folks who seemed knowledgable over at XS mentioned this in correcting others who though M stood for Megabyte in SuperPi.

try this:
>>> import sys, math
>>> dir(math)
['__doc__', '__file__', '__name__', 'acos', 'asin', 'atan', 'atan2', 'ceil', 'cos', 'cosh', 'degrees', 'e', 'exp', 'fabs', 'floor', 'fmod', 'frexp', 'hypot', 'ldexp', 'log', 'log10', 'modf', 'pi', 'pow', 'radians', 'sin', 'sinh', 'sqrt', 'tan', 'tanh']
>>> pi = 48 * math.atan(1/float(49)) + 128 * math.atan(1/float(57)) - 20 * math.atan (1/float(239)) + 48 * math.atan(1/float(110443))

Thanks Mr. Wizard, but I have no clue what any of that means, nor does in answer my initial question (in any way that I could understand at least). But I appreciate your enthusiasm.

So, M does indicate "million decimal places" then and not "Megabytes". So then all the arguments about fitting a 1M SuperPi run completely in it's cache is somewhat bogus. As the gent above stated, a 1M SuperPi run requires 8MB of memory to run the calculation to 1 million decimal places. And even the Conroe with 4MB of cache will have to hit the system memory even in the 1M test.

Oh, you can download a py interpreter to run this code from:
www.python.org
I'm not sure how many decimal places would windows allow for FP representation, but you should be able to do it correctly under linux.

The point is that calculation of Super PI does not require access to the entire dataset (in this case all of the previous digits), and can be done by shorthand algorithms that could be faithful to over a trillion degits to the right of the decimal. I don't know in which form the result data is stored in SuperPI. And while each digit in the result of the calcuation could be stored as a 32 bit in binary representation, hence a total of 4MB or greater amount has passed through L1 data. There is a difference between the amount of data stored and the data that needs to be accessed.

The common algorithms such as the one above:
pi = 48 * math.atan(1/float(49)) + 128 * math.atan(1/float(57)) - 20 * math.atan (1/float(239)) + 48 * math.atan(1/float(110443))
where atan denotes arctangent in python math library.
And from there you expand each term involving atan according to power series definitions in trigonometry, including:
sin(x) = x - (x^3/3!) + (x^5/5!) - (x^7/7!) + (x^9/9!) + ...
cos(x) = 1 - (x^2/2!) + (x^4/4!) - (x^6/6!) + (x^8/8!) + ...
And then from the fact that tan(x) = sin(x)/cos(x), you can derive the power series representation of arctan value of any rational number. The more terms in the series representing arctan value, the more accuracy you would be able to achieve.

And when 4 or 8MB of memory is used for any program calculating superPI value, it doesn't mean that all of the memory space allocated by the OS will need to be accessed by the algorithm to complete the calculation. Actually, I don't know what method SuperPI used to do the calculation or to store the result digits, but whatever method that they employ, you won't be able to know the amount of memory accessed by its algorithm simply by looking at the amount of address space that Windows is reporting to have been allocated.

Edit: Typos

 

[xenolith]

Originally posted by: BrownTown
well, when was the last time a new architecture launched? K8, and that was a few years back. New architectures don't exactly come around very often, and when a new one comes around and its pwning the living daylights out of everything else of course people are gonna be excited.

And it's about time the cause of this excitement is from an Intel part.

 

[TrevorRC]

Originally posted by: BrownTown
that 23min time is actually slower then what it sholuld be, the 32M test wants to use more memmory then he has, so its getting bottlenecked, it would likely improve considerably with 1gig of RAM.

I know I was taking that into account.

Considering 1M Pi = 8Megs; 32M Pi = 256MB. Nice number, isn't it?

Also the exact amount of RAM the system had.

Count any side loading tasks, and the app itself... and you're drawing on the HDD.

Edit: Someone just said he was running 512. I could've sworm I read 256 somewhere.

 

[BrownTown]

hes running 512, but when i run 32M on my computer its taking more than 512MB to run everything. He has a clean OS install, so that may not be the case on his computer, but certainly 1gig aint gonna do anything but help.

 

[Absolute0]

To end this crap i started running a SP32M on my computer

In task manager, it came up as 265,000 kb, and the page file usage went up by 250,000 kb

So in total, SP32M took about 500 megabytes of system memory for me. So if you have 512mb of ram trying to run a 32M... more will definitely help!