Nehalem only supports DDR3 800 or 1066

[Shtal]

We were surprised to see that the Bloomfield generation of CPUs will officially support DDR3 800 or 1066. Even the extreme edition 3.2GHz Core i7 officially only supports DDR3 1066 and not faster. There won't even be an official support for DDR3 1333 not to mention 1600 or faster.


Fudzilla has already mentioned the fact that for some reason you need to run memory CPU voltage synchronously which might be an overclocker's worst nightmare. At this current stage you cannot go over 1.65V and some memories such as DDR3 1600 or DDR3 2000 might need a bit more than that. With more than 1.65 you will kill any current Nehalem CPU.

Bloomfield Nehalem doesn’t run with DDR2 memory as the memory controller is inside of a CPU and so far we are not aware of any plans to release Bloomfield Core i7 CPU that would support DDR2.

DDR3 1066 doesn’t really sound like a state of the art speed but this is the reality. http://www.fudzilla.com/index.php?option=com_content&task=view&id=9214&Itemid=1

Model Memory
Brand G.SKILL
Model F3-8500CL6D-2GBHK
Type 240-Pin DDR3 SDRAM
Tech Spec
Capacity 2GB (2 x 1GB)
Speed DDR3 1066 (PC3 8500)
Cas Latency 6
Timing 6-6-6-15
Voltage 1.65V
Heat Spreader Yes
http://www.newegg.com/Product/Product.aspx?Item=N82E16820231162

Looks like we have a limit; but still, triple channel DDR3 1066MHz @ 6-6-6-15 Timing is very good
Install using intel-X58 MB of 6 sticks G.SKILL 1GB DDR3 = Total 6GB memory for running vista64!

 

[homerdog]

That's odd, but I think this is the wrong section for a Nehalem discussion.

 

[Shtal]

http://www.fudzilla.com/index.php?option=com_content&task=view&id=9247&Itemid=1

Nehalem 4.1GHz pictured Print E-mail
Written by Fuad Abazovic
Thursday, 04 September 2008 10:16

Image

Super Pi under 10 sec

We have already told you that Nehalem is highly overclockable and today we have a screenshot of a successful overclock on a Bloomfield, Core i7 CPU.


The chap managed to overclock 2.93GHz Engineering samples to a whopping 4.11GHz on air and he increased the voltage to 1.576V.

They set a multiplier to 31 and he got the final score of 4112.7 MHz and this is probably not the last stop when you overclock by air. The normal multiplier for 2.93GHz Nehalem is 22 times and you can see that the CPU has four cores and eight threads, 4x256KB L2 cache and 8MB L3 shared cache.

The CPU was stable at 4.11GHz and it finishes super Pi in just under 10 seconds, 9.969 to be exact.

 

[Blakhart]

This thing is going to put the hurt on my brand new e8500 isn't it?

 

[ShaidarHaran]

1.576V on air is INSANE!

Then again, so is that SuperPi time!

 

[Albuquerque]

Well, in the grand scheme of things, even DDR3-800 in triple-channel with the new integrated memory controller is going to yield memory bandwidth far in excess to just about anything we currently have available. And even if you could overclock your rig to the max and match Neha's bandwidth figure on your current hardware, you still wouldn't get the far lower latency that's available with Nehalem either.

 

[Shtal]

Intel has told that the upcomming Processors Core i7 are faster than first-tests have shown. The cause is that all tests were made with the B0-Stepping, while the final processors will have the B2-Stepping or even higher.

http://forums.vr-zone.com/showthread.php?t=323664

 

[Thorburn]

Fudzilla has already mentioned the fact that for some reason you need to run memory CPU voltage synchronously which might be an overclocker's worst nightmare. At this current stage you cannot go over 1.65V and some memories such as DDR3 1600 or DDR3 2000 might need a bit more than that. With more than 1.65 you will kill any current Nehalem CPU.

Total toss, the VCore and VDIMM are not linked.

Anyone can ignore this, 1.65V will kill a 45nm Core 2 chip, hell 1.5V will kill it over time and I can't imagine Bloomfield would be too happy either (plus the TDPs would go up massively).
So unless we are getting DDR3 DIMMs that can run at about 1.1-1.2V (hint - we aren't) I think we can safely say this is Fuad being an idiot as usual.

 

[pjbliverpool]

1.576V on air is INSANE!

Then again, so is that SuperPi time!

Do you have any comparison SuperPi times? I don't kow whats considered normal for this benchmark.

 

[ShaidarHaran]

Do you have any comparison SuperPi times? I don't kow whats considered normal for this benchmark.

You may want to check the forums @ xtremesystems for the answer to that. I recall getting into the low 11's on my E8400 @ similar clocks (4-4.2) but I don't run her that high anymore, least not in the summer. Relatively untweaked time on my part. I'm sure with better memory timings 10's would be easy.

I guess what impresses me is seeing Nehalem perform so well in Super Pi with just the 256KB L2 cache per core. Looks like the L3 + IMC combo is going to work just fine.

 

[pjbliverpool]

You may want to check the forums @ xtremesystems for the answer to that. I recall getting into the low 11's on my E8400 @ similar clocks (4-4.2) but I don't run her that high anymore, least not in the summer. Relatively untweaked time on my part. I'm sure with better memory timings 10's would be easy.

I guess what impresses me is seeing Nehalem perform so well in Super Pi with just the 256KB L2 cache per core. Looks like the L3 + IMC combo is going to work just fine.

Cheers, i'm guessing core count doesn't make a difference then?

 

[ShaidarHaran]

SPi is single-threaded. You can run multiple instances if you have some other academic curiosity about bandwidth bottlenecks, perhaps. Say testing in single vs. dual channel mode with RAM @ 800MHz and 1GHz.

SPi seems to really like cache, so Conroe/Penryn have always done extremely well in small working set sizes like the 1M test. Hence my L2 comments WRT Nehalem.