Sign in with
Sign up | Sign in
Your question

Large memory pages

Last response: in CPUs
Share
November 29, 2008 7:38:55 PM

While using SiSoft Sandra I attempted to do a benchmark of
the Multi Core Efficiency (Core inter-connects)

The score was pretty low.
At the bottom of the page I got a warning
"Cannot use large memory pages due to lack of privileges"
"Use the local security policy to give yourself Lock Memory Pages privilege"

My first question is what the hell does that mean?
Would enabling large memory pages make a difference in real life and if so how/why?

The comparison results show Intel being very good at this and AMD not so much.
Is this a AMD short coming?

Assuming activating this feature would be a good thing, how would I do it?

System
Biostar TF 520 A2+
AMD 64 X2 5600 (windsor) (2x64kB L1 2x1MB L2)
Nvidia nForce 520 chipset
4x 512mb Corsair memory

Any other info can be provided.



More about : large memory pages

November 29, 2008 8:56:15 PM

If you're measuring raw memory bandwidth then you want to ensure the data is locked into memory and page table accesses are minimised; I guess that's why they want to use large pages and lock them in place.

You may need to run it as an administrator to get those priviledges.
a b à CPUs
a b } Memory
November 30, 2008 2:02:49 AM

I just ran it on my CPU, and it gave the same error, but the numbers were fine. What kind of inter-core bandwidth did it report?
Related resources
December 1, 2008 11:40:46 AM

The numbers I got were
2295 MB/s (very low)
134 NS Latency (very high)

I think I may have a handle on whats going on though.
My computer has been getting pretty unpredictable lately so I am in the process of getting things ready for a reload.
I didnt know before that the NForce drivers need to be deleted before upgrading. (strike one)
I did some updating from Msoft (strike two)
Added Silverlight (possibly strike three)

Any way suffice it to say I did it all to fast and cant really pin point the real cause, so Im slipstreaming Vista & SP1.
Getting the newest drivers from Nvidia and moving everything off to a external.
I also think Im going with V64 this time and see how that goes.
a b à CPUs
December 1, 2008 6:27:43 PM

I just ran it on my Xeon E3110

Benchmark Results
Inter-Core Bandwidth : 9.02GB/s
Results Interpretation : Higher index values are better.
Inter-Core Latency : 29ns
Results Interpretation : Lower index values are better.
!