ProLiant Servers (ML,DL,SL)
cancel
Showing results for 
Search instead for 
Did you mean: 

DL585 G2 Mem speed 667 -> 533, faster with Shanghai? Is HT faster also?

 
martyscholes
Occasional Advisor

DL585 G2 Mem speed 667 -> 533, faster with Shanghai? Is HT faster also?

Hello,

 

I am trying to understand how the memory speed changes on a G2 with different memory and CPU combinations.

 

Current server is DL585 G2 with 4x Opteron 8216 and works very well running Solaris 11, except a minor USB problem.  Machine has 8x 1GB RAM and 24x 2GB RAM.  This means two of the CPUs have three banks populated (1 bank with 2x1GB and 2 banks with 2x2GB) with two CPUs having two banks populated (1 bank with 2x1GB and 1 bank with 2x2GB).  According to  HP Specs the memory speed drops from 667 to 533 when all four banks are populated, but remains at 667 if one or two banks are populated.  The specs are unclear what happens when when three banks are populated.  What speed does memory run at with three banks populated?  Is there a graceful way to determine memory speed of a particular configuration?

 

We just ordered 16x2GB RAM which will allow us to upgrade to 60GB with 4x1GB left over.  Once upgraded, the memory speed surely will drop to 533, 80% of the 667 it would otherwise run.  We are considering upgrading to quad core Shanghai processors which can operate against 800 MHz RAM.  How does the slowdown on the DL585 work with four banks?  Does it always slow down to 533 or is it 80% of the CPU capability?  In other words, if we install Shanghai CPUs which are capable of 800 and all memory banks are populated, will the speed drop to 533 or stay at 667 (80% of 800)?  The specs for a G5 and G6 are a little more clear.  Will the G2 behave like a G5/G6 with the new CPUs?

 

Along similar lines, the current Santa Rosa CPUs have a HyperTransport speed of 1 GHz but the Shanghai CPUs operate at 1, 2 or 2.2 GHz.  Will the HyperTransport speeds increase with the newer CPUs?

 

I apologize if this has been covered before.

 

Thanks for any insight,

Marty