IDF Spring 2007: Benchmarking Penryn

Originally Posted by article
We begged, pleased and argued

You did what now?

Those figures do look nice, but the how much is applicable to the larger L2 and how much to the higher FSB

Originally Posted by article
Since the 45nm process is still immature but these CPUs are offering a performance that will be exactly the same as on desktop.

This doesn't read correctly.

Out of slight curiosity could you overclock a QX6700 to 3.33GHz (on a 1333Mhz FSB) and compare the quad core Penryn numbers clock-for-clock?

Thanks, this was a time sensitive piece and I wasn't available to proof the article before being published. Thus, there were a few errors not picked up by the other proofers.

Originally Posted by JADS
Out of slight curiosity could you overclock a QX6700 to 3.33GHz (on a 1333Mhz FSB) and compare the quad core Penryn numbers clock-for-clock?

It's certainly possible to create a system similar to that in our offices, as our QX6700 was capable of 3.46GHz using multiplier changes. We'll have to see.

yeah that would be excellent.
to try and see what they rate at clock for clock would be great

my question tho is surely with a few OC'ers going upwards of 5Ghz would it be likely we will see a few penryn going upwards of 6Ghz and therefore possibly sub 8 sec super pi runs

Originally Posted by JADS
Out of slight curiosity could you overclock a QX6700 to 3.33GHz (on a 1333Mhz FSB) and compare the quad core Penryn numbers clock-for-clock?

We wanted to but they actually wouldnt let us play with the systems. This was our exact argument but they were having none of it.

Originally Posted by Mary Jo
This doesn't read correctly.

Fixed! The processors are exactly the same as what you can buy but because the 45nm process is so new, the yields are still poor and the cost per chip is still exceptionally high. Therefore supplying the press with samples would cost Intel far too much, instead we'll have to wait a few months.

Originally Posted by DougEdey
You did what now?

Those figures do look nice, but the how much is applicable to the larger L2 and how much to the higher FSB

Subst the s for a d they are next to each other on the kb!

I was quoted "12% to clock speed, 14% to cache" but that was under a certain circumstance I can't seem to find now. It depends on what you're running: if the data can fit in cache then it'll be cache dependent, otherwise it'll be FSB since it'll be accessing memory.

And Intel get the prize for least scientific performance demonstration this year. Wow a new architecture running at higher FSB and higher clock gives a better performance...who would have guessed?! They really should have had Penryn at stock, QX6800 at stock, and then either Penryn at 11 x 266 or QX6800 @ 10 x 333. Without that, the numbers are meaningless!

Still, subject to all these caveats, there is massive potential, particularly in that encoding score!

Originally Posted by Bindibadgi
We wanted to but they actually wouldnt let us play with the systems. This was our exact argument but they were having none of it.

As Tim mentioned you can replicate a 3.33GHz Kentsfield in the lab and do a direct comparison so it should give a better indication of the potential performance improvement.

Penryn good, Nehalem better :)?

Originally Posted by JADS
As Tim mentioned you can replicate a 3.33GHz Kentsfield in the lab and do a direct comparison so it should give a better indication of the potential performance improvement.

Penryn good, Nehalem better :)?

Providing Rich can get the source material used in the benchmarks.

Nice one. Good report - can't wait until someone gets their hands on one of these for some intensive testing. :)

If this is to be believed, AMD better do something fast.

AMD where art thou!

Originally Posted by mclean007
And Intel get the prize for least scientific performance demonstration this year.

Originally Posted by mclean007
And Intel get the prize for least scientific performance demonstration this year. Wow a new architecture running at higher FSB and higher clock gives a better performance...who would have guessed?! They really should have had Penryn at stock, QX6800 at stock, and then either Penryn at 11 x 266 or QX6800 @ 10 x 333. Without that, the numbers are meaningless!

I thought the same, what do they have to worry about if there new design is so much better than a simple die shrink.

Originally Posted by Grinch123456
If this is to be believed, AMD better do something fast.

exactly what i was thinkin.

My AM2 board, which i was under the impression would be future proof for at least 18 months is lookin more and more outdated by the day.

I so wish i'd bought a 775 and C2D... we live and learn.

Originally Posted by JADS
As Tim mentioned you can replicate a 3.33GHz Kentsfield in the lab and do a direct comparison so it should give a better indication of the potential performance improvement.

Penryn good, Nehalem better :)?

To make it properly scientific you'd want to use exactly the same system and benchmarks for both tests, setup by just one person.

Would be good to see some dual core compare's too if possible. Know you stated you'd need exactly the same setup etc which is fair enough, but it'd be nice to get a ballpark figure for the performance improvement (or not).
The clockspeed defecit the Q6800 has over the other quad would certainly be MUCH lower if the clockspeeds were matched, the extra bus speed might well help things to an extent too. (could also be worth changing the labelling here, as the QX6800 isnt labelled as quad core which could cause confusion)
Clock for clock, things don't look a _huge_ amount quicker. From Intels point of view I guess it makes sense for them NOT to overclock one/underclock another as the performance difference between the chips would certainly be less marked than these stock runs with the large clock speed discrepancies