Nehalem Overclocking @ XS

[massman]

Ehrm ... you're kidding me, right?

E8400 3.6G: 225021
QX9650 3.6G: 226854

Difference: 1833 points, which makes the quadcore exactly 0,8145% faster, not even ONE whole percent. Seriously, re-run Aquamark and you'll be losing more than that due to the variance in the score.

[massman]

You'd better switch motherboards then, the gain is much bigger:

[BenchZowner]

Ehrm ... you're kidding me, right?

E8400 3.6G: 225021
QX9650 3.6G: 226854

Difference: 1833 points, which makes the quadcore exactly 0,8145% faster, not even ONE whole percent. Seriously, re-run Aquamark and you'll be losing more than that due to the variance in the score.

Those were the averages of 5 runs m8.
I know it's weird, and I've tried it again and again and again, with the same results.

[massman]

Ever heard of the law of large numbers?

"The law of large numbers (LLN) is a theorem in probability that describes the long-term stability of the mean of a random variable. Given a random variable with a finite expected value, if its values are repeatedly sampled, as the number of these observations increases, their mean will tend to approach and stay close to the expected value."

Basicly: the lower the number of tests you run, the more likely your results are flawed. 5 runs and an average gain of 0,8145% does not prove anything. Come back if you've done 5000 runs and still have a 1% gain, then you can (maybe) call this an anomality.

[BenchZowner]

Pieter, I can't remember AquaMark 3 giving me a 1500+ pts difference between any runs at the same settings.
The average variation is around 300pts, and the max I've seen in my testing is about 800-900pts.

Once I get done with the P5Q3 Deluxe review I'll set up the system again and have it run AQ3 50 times with the QX9650 and 50 times with a E8600 at the same settings and tabulate the data.
TBH though, I don't think that the E8600 will get a higher score than the QX9650 at all.

[massman]

Please run the Lavalys Everest benchmark as well, just to exclude the bandwidth/latency theory.

[Johnny Bravo]

we we can all accept that AM3 is a single threaded app right?

So therefore the cores themselves aren't making the difference so maybe its more to do with how the cores access the memory? Perhaps bandwidht is better shared between the two. I'm a bit excited as it means there is something new to discover

[Oliver]

thx for posting Charles

keep pushing it

[BenchZowner]

Please run the Lavalys Everest benchmark as well, just to exclude the bandwidth/latency theory.

Will do mate

we we can all accept that AM3 is a single threaded app right?

Of course we can accept that, because it is true

So therefore the cores themselves aren't making the difference so maybe its more to do with how the cores access the memory? Perhaps bandwidht is better shared between the two. I'm a bit excited as it means there is something new to discover

If AQ3 could take advantage of two cores and use them wisely when there are more than 2 available, we could've an explanation ( using Core 0 and Core 3 to give each core the whole L2 cache of the die )... but it's not.

But since AQ3 is single-threaded... well... we'll have to find out

[massman]

I'm with Johnny on this one. The actual number of cores doesn't matter, it's probably the bandwidth balance over the cores. Maybe because a quadcore requires more resources, it uses more memory actively and creates a minimal extra on bandwidth?

Just to make it complete, you should run the E8400 at 1 core and the QX9650 at 2 cores and 1 core :p

//EDIT: Just to be clear: I expect to see less difference when you ran the benchmark 50 times, this theory is only when the the conclusion of the tests show that the QX9650 is >= 0,8145% faster than the E8400.

[SoF]

guys keep on-topic and pm yourself about that am3 thing...

charles that's a great show-off that new chip, seems like you are one step further with nehalem oc than everybody else can't wait for some 3D results - numbers are looking way better than I expected so far. Thanks for sharing and keep going interesting to see about the voltages, hope it can take 2V on ram under cold for a longer time...

[[XC] Oj101]

If AQ3 could take advantage of two cores and use them wisely when there are more than 2 available, we could've an explanation ( using Core 0 and Core 3 to give each core the whole L2 cache of the die )... but it's not.

But since AQ3 is single-threaded... well... we'll have to find out

You may be onto something there. Windows might use one core from a pair and AM3 a core from the other pair.

[Gautam]

I agree with Bill and Dino regarding AM3. It is simply faster clock for clock with 4 cores. I didn't believe it for a sec when some people started hinting at it, but I found it to be true...

No idea why.

[Fr3ak]

Max bclk of ~220, expect way lower though for most chips.

Ye, expect around 180 MHz for stable results on air.

I agree with Bill and Dino regarding AM3. It is simply faster clock for clock with 4 cores. I didn't believe it for a sec when some people started hinting at it, but I found it to be true...

No idea why.

That might be true, but for now dual cores overclock much higher than quads, so even if there is a 1% difference lock for clock, who cares when dual cores can be clocked 8 % higher (Wolfdale/Yorkfield wise)?

[Pt1t]

Ye, expect around 180 MHz for stable results on air.

Yes , what i can with i7 920 and EX58-Extreme on aircooling. That can prime to ~190mhz Bclk and bench to 200.

[c1rcu1tburn3r]

Mmm are you happy about those i7 guys? I'm not so enthusiast atm...we'll see

[RoyaL]

wtf ..not even 300fsb?

ok,i'll wait for the next step

[PytonOrm]

Get some Ln2 nitrogen on it ? =)

[Gautam]

wtf ..not even 300fsb?

ok,i'll wait for the next step

The base clock is NOT a datapath. It's just a clock which the QPI bus, CPU and other clocks are derived off of. What the base clock is at has no direct bearing on performance.

[wagih]

Great Numbers
Just with the lunch

[Fitseries3]

Yes , what i can with i7 920 and EX58-Extreme on aircooling. That can prime to ~190mhz Bclk and bench to 200.

i thought the i7 chips have 2 multis up from what they run stock?

cant you run 22x with the 920?

on TEC would i be able to get 4.2ghz benchable? higher?

[dinos22]

That might be true, but for now dual cores overclock much higher than quads, so even if there is a 1% difference lock for clock, who cares when dual cores can be clocked 8 % higher (Wolfdale/Yorkfield wise)?

Oli you are right that's why all current records are on dual core

my point wasnt to argue that
i simply said that we'll see some nice AM3 scores because of monster bandwidth and extra cores........i know it sounds weird and i didnt believe it at first but it just seemed to score better clock for clock so all i am saying is that these new chips should make AM3 look good...just a general comment

massman i total get what you are saying and i dont know if its the extra cores or whatever but that is the main difference between dual core and quad core CPUs which is why i assumed that it may be those extra cores that boost it somehow....NFI how it works with AM3 lol

half to one percent on 350K score is a pretty big figure

[G H Z]

Nice results FUGGER, looks like we'll be seeing some big gains in Vantage and 3DMark06. Have you tested wPrime yet?

[Pt1t]

i thought the i7 chips have 2 multis up from what they run stock?

cant you run 22x with the 920?

on TEC would i be able to get 4.2ghz benchable? higher?

yes off course you can run @ 22X in idle or sometimes with pi 1M but when you launch multithread aplication the multiplier goes down to 20X.

[pxhx]

Nice results FUGGER, looks like we'll be seeing some big gains in Vantage and 3DMark06. Have you tested wPrime yet?

congrats FUGGER.
yeah... i'm curious too.
Run wPrime, please.