More Sandy Bridge performance numbers (+Rumoured Pricing)

[Iconyu]

@Calmatory:

I think theres two Sandybridge E's one with Triple channel ram and one with Quad. So the ones benched will be the mainstream Sandybridge, which will probably stay Dual channel.

For the more serious performance heads, it should be worth waiting for the performance range, even if it is a few extra months. If you own a 920 now, would you switch to a i5?

[terrace215]

@Calmatory:

I think theres two Sandybridge E's one with Triple channel ram and one with Quad. So the ones benched will be the mainstream Sandybridge, which will probably stay Dual channel.

For the more serious performance heads, it should be worth waiting for the performance range, even if it is a few extra months. If you own a 920 now, would you switch to a i5?

I think the "triple channel" one was canceled in favor of quad-channel / LGA-2011 on the high-end desktop as well as server/workstation. Perhaps some of the NDA types can comment, in a vague manner, of course.

[kl0012]

is there a simple answer whether or not the benchmarks were both done at exactly 1.6ghz? im still lost about what turbo features might or might now have been on.

Here are the direct links:
http://browse.geekbench.ca/geekbench2/view/273184
http://browse.geekbench.ca/geekbench2/view/274571

SB system has less memory. BTW i7-Q720 gets much higher scores with 64-bit OS.
http://browse.geekbench.ca/geekbench2/view/274468
I would like to see SB runing 64-bit W7 instead of W-XP.

[Manicdan]

Here are the direct links:
http://browse.geekbench.ca/geekbench2/view/273184
http://browse.geekbench.ca/geekbench2/view/274571

SB system has less memory. BTW i7-Q720 gets much higher scores with 64-bit OS.
http://browse.geekbench.ca/geekbench2/view/274468
I would like to see SB runing 64-bit W7 instead of W-XP.

interesting

it looks like almost EVERY single threaded benchmark went way down in perf, but multi is up by a very large amount. anyone have an ideas?

i threw them into excel and heres the results: (results are a percentage of the i7 920)

Int
Single: 93%
Multi: 134%

Float
Single: 95% (one was 160%, most around 80% though)
Multi: 132%

the Memory ones are all single and avearge 98%

[informal]

You can find that Core i7 Q 720 scores better than the 4800 points @ 32bit Win 7:
http://browse.geekbench.ca/geekbench2/view/269110
http://browse.geekbench.ca/geekbench2/view/266512
http://browse.geekbench.ca/geekbench2/view/272074
http://browse.geekbench.ca/geekbench2/view/272601
http://browse.geekbench.ca/geekbench2/view/261386

I'm not sure why would more memory affect the scores if it's clocked lower,but the systems above do have more memory.Problem is that all the above results are done on 32bit OS so it doesn't matter since the OS can't even see the full 4GBs(it will see 4GB, but only address 3.XGB).

[god_43]

Sandy Bridge is a new uarch in fact. The i7 in 45nm are the new arch, the Nehalem family, and the 32nm (Westmere if I'm not mistaking) is a die shrink. Sandy Bridge is a new arch, Ivy Bridge will be a shrink, Haswell will be a new arch and so on

oh thanks.

[Manicdan]

also the SB was on XP instead of Win7

[god_43]

also the SB was on XP instead of Win7

...thats ed up! what kind of moron puts sb on XP......just when i think ive seen it all...

[Manicdan]

...thats ed up! what kind of moron puts sb on XP......just when i think ive seen it all...

i wonder if you can run it on win7 using xp compatibility mode (or that might just make things even worse for comparison)

[Kuroimaho]

Memory timings or something else could cause the difference in results as well not necessarily has to be the OS.
As Francois said these are not tuned intentionally to hide the performance so it is a guesstimate anyway.

[informal]

Actually XP is not that heavy on the system resources as Win 7 or Vista are ,but it would be more useful (for comparison purposes) if they have ran it on Win 7 instead.

[kl0012]

interesting

it looks like almost EVERY single threaded benchmark went way down in perf, but multi is up by a very large amount. anyone have an ideas?

i threw them into excel and heres the results: (results are a percentage of the i7 920)

Int
Single: 93%
Multi: 134%

Float
Single: 95% (one was 160%, most around 80% though)
Multi: 132%

the Memory ones are all single and avearge 98%

There is something interesting here. For some FP test the perf delta between single-threaded to multi-threaded is about 6x-7x. Which mean that, even if we assume an equal turbo for 1 to 4 cores, HT efficiency is reaching 50-85% on a threads with an equal code. Thats very strange.

[informal]

There is something interesting here. For some FP test the perf delta between single-threaded to multi-threaded is about 6x-7x. Which mean that, even if we assume an equal turbo for 1 to 4 cores, HT efficiency is reaching 50-85% on a threads with an equal code. Thats very strange.

That or in single thread mode the Turbo on the old i7 is ridiculously high while in MT mode it is only one bin.With SB,the Turbo in single thread mode is maybe lower but it kicks in aggressively in MT mode providing the performance gains of 6x or 7x. Also you can see in my previous post that there are numerous 5100+ results on 32bit win7 OS for the i7 Q720. Q720 Turbo with 3-4 loaded cores is 1.73Ghz while for 1 core it's 2.8Ghz.

[Manicdan]

i just dont like this geek bench, seems to be way to picky on so many settings, that we want apples vs oranges, but are getting a comparison like eggs to potatoes

[Iconyu]

i just dont like this geek bench, seems to be way to picky on so many settings, that we want apples vs oranges, but are getting a comparison like eggs to potatoes

Also to make things worse Intel could have gimped the ES chips, as they don't have to prove anything.

[Chumbucket843]

There is something interesting here. For some FP test the perf delta between single-threaded to multi-threaded is about 6x-7x. Which mean that, even if we assume an equal turbo for 1 to 4 cores, HT efficiency is reaching 50-85% on a threads with an equal code. Thats very strange.

i wouldnt be surprised if HT could reach that efficiency. one bottleneck in nehalem is that it cant do enough load/stores per cycle to keep two memory bound threads fully used. this is the most likely reason why 50% tends to be the upper bound for HT speed ups. doubling load/store capabilities should put the bottleneck elsewhere.

[terrace215]

That or in single thread mode the Turbo on the old i7 is ridiculously high while in MT mode it is only one bin.With SB,the Turbo in single thread mode is maybe lower but it kicks in aggressively in MT mode providing the performance gains of 6x or 7x. Also you can see in my previous post that there are numerous 5100+ results on 32bit win7 OS for the i7 Q720. Q720 Turbo with 3-4 loaded cores is 1.73Ghz while for 1 core it's 2.8Ghz.

It could be that the single-thread SB turbo is not yet working to its ideal aggressive setting, true. Doubt that there is much turbo in the multithreaded case for either part, I expect the gains have to do with the extra load port given to SB. They did promise HT improvements.

[-Sweeper_]

no server chips are not q3/2010 that is few desktop and mobile , servers is q1/2011.

uuuh, no, they are not.
exactly the opposite...

[-Sweeper_]

or:
- the cooling was not sufficient (CPU, VR) to allow more headroom
- the PCU was not tuned (early ES)

so the turbo mode could be off or not working normally?

[informal]

It could be that the single-thread SB turbo is not yet working to its ideal aggressive setting, true. Doubt that there is much turbo in the multithreaded case for either part, I expect the gains have to do with the extra load port given to SB. They did promise HT improvements.

I think that you will see some very aggressive Turbo that will kick in with even all 4 cores loaded.Sandy Bridge has a GPU that sits idle in most non graphic intensive tests,so there is a solid power saving potential to tap into.

so the turbo mode could be off or not working normally?

I think he is referring to single thread results that are lower than with the i7,while the MT results are much higher.This looks like the situation where Turbo in "old"
i7 kicks properly to 2.8Ghz but sinks to lowly 1.73Ghz with all the cores loaded,while SB Turbo mode doesn't kick in properly with 25% load on the chip but clocks up way more with 3-4 cores loaded.

[Kuroimaho]

Also to make things worse Intel could have gimped the ES chips, as they don't have to prove anything.

Read Francois posts, he wrote that they actually did that so actual performance can't be squeezed out of these.

[-Sweeper_]

Read Francois posts, he wrote that they actually did that so actual performance can't be squeezed out of these.

by the way, he could give us some light about those numbers...

where are you dr.who?

[duploxxx]

uuuh, no, they are not.
exactly the opposite...

get lost, there is not a single server ship in a new arch to be released in Q3/2010 then I would have it already before summer on ES. Again estimation is q1/2010 early or late I will give that the benefit of the doubt, will be sure about that in September.

[-Sweeper_]

get lost, there is not a single server ship in a new arch to be released in Q3/2010 then I would have it already before summer on ES. Again estimation is q1/2010 early or late I will give that the benefit of the doubt, will be sure about that in September.

They shall arrive in Q3/2011 (along with socket LGA2011), maybe a little bit earlier in late Q2.
There will be no SB for servers in Q1/2011.
Sorry, you are wrong.

Edit: Just saw I posted 2010 in the first page, my distraction.

[terrace215]

I think that you will see some very aggressive Turbo that will kick in with even all 4 cores loaded.Sandy Bridge has a GPU that sits idle in most non graphic intensive tests,so there is a solid power saving potential to tap into.

That's a good point. I still think that memory and/or turbo is not properly tuned with those results. Both should be better than the older model.