B200 M2 memory inquiry

[russ.givens]

I have 8 B200 M2 blades that are running ESXi.  Presently, each blade has 96GB's of memory installed (12x8GB DIMM's with clock speed of 1333MHz).  I have enough memory to upgrade each blade to 128GB's of memory using 8x16GB DIMM's (1066 MHz clock speed).   There are 3 memory channels per socket in the B200 M2 (cpu 1 A, B, & C and cpu2 D, E, & F).  Is there any negative impact of going from using all slots in every memory channel to only using the full memory channels A & D and half the memory channels for B, C, E, & F?   The documentation states I should populate the DIMM's the following way A1, A2, B1, C1 and D1, D2, E1, F1. 

I could upgrade 6 of my ESXi servers to 192GB's and fill all the channels completely, but I would rather have all 8 ESXi servers will equal memory as they are in the same cluster.

Thanks

[Keny Perez]

Hi Russ,

I would advise to stick to the recommended configuration for the specific server, cause Cisco will only support that configuration. What I mean is, that if you install it and it works for a few days and then fails, you open a TAC case and we see that, we immediately will ask you to arrange the memory sticks as recommended in the data sheets.

With that being said, here you have the possible configurations>

DIMMs per CPU

# CPU 1 installed slots

# CPU 2 installed slots

1                                             A1                                                             D1

2                                             A1, B1                                                        D1, E1,

3                                             A1,B1,C1                                                   D1,E1,F1

4                                             A1, A2, B1,C1                                             D1,D2,E1,F1

5                                             A1,A2B1,B2, C1                                         D1,D2, E1, E2, F1

6                                             A1,A2, B1,B2, C1, C2                                 D1,D2,E1,E2,F1,F2 

http://www.cisco.com/en/US/partner/docs/unified_computing/ucs/hw/chassis/install/blade.html#wp1039261

I hope that helps avoid future inconveniences,

-Kenny

[russ.givens]

Thanks for the reply.  I guess I'm not being clear.  I presently have the configuration where every DIMM slot is full with 8GB DIMMs.  I would like to go to the configuration where I have A1, A2, B1, C1 and D1, D2, E1, F1 populated with 16GB DIMM's.  I looked at the documentation from Cisco and it is indicated that it is supported.  My question is are there any negatives of going to this configuration?

I only have 72 16GB DIMMs.  I have 8 ESXi servers in a cluster.  I could take those 72 DIMMs and upgrade 6 of the ESXi servers to 192GB's, leaving the other 2 ESXi servers at 96GB's.  Or I was thinking of taking all 8 ESXi servers to 128GB's of memory using the DIMM's slots A1, A2, B1, C1, D1, D2, E1, F1.  I would prefer having all ESXi servers in a cluster to have the same amount of memory but just wanted to validate that the memory configuration that I'd be going to is okay are whether there are negatives to it.

[padramas]

Hello Russ,

Installing 16 GB memory modules in DIMM slots A1, A2, B1, C1 and  D1, D2, E1, F1 is supported and valid configuration.

http://www.cisco.com/en/US/docs/unified_computing/ucs/hw/chassis/install/blade.html#wp1035302

A few things to look out for this deployment scenario are

-----------------

Product ID -- ( A02-M316GB2-L )  16 GB DIMMs require Cisco UCS Manager version 1.4(1) and UCS capability catalog version 1.0.16 or later, and the related BIOS package.

Product ID --  ( A02-M316GB1-L )  This DIMM requires UCS capability catalog version 1.0.50.T or 2.0.1nT or later.

Fully populating 1 bank or 2 banks with DIMMs will ensure optimal memory bandwidth running at the 1333 MHz speed. If DIMMs are partially populated in 1 bank (less than 6 DIMMs) or 2 bank patterns (less than 12 but greater than 6 DIMMs) the 1333 MHz speed can be used, but the overall memory bandwidth will not be optimal.

-----------------

If you would like to obtain maximum memory bandwidth, you can populate six blades with 16 GB modules and remaining two blades with 8 GB modules. With different memory capacities, you would also need to consider which hosts can meet the application or guest OS requirements.

HTH

Padma

[Amit Vyas]

Hi,

Just wanted to know from where we can get:

• UCS capability catalog version 1.0.16 or later, and the related BIOS package.
• UCS capability catalog version 1.0.50.T or 2.0.1nT or later

We are also planning to upgrade the memory in same scenario as Russ has shown here.

Apricated you revert for the same

Regards,

Amit Vyas

[Keny Perez]

Amit,

Here you have a link to the latest versions of the capability catalogs available, you may choose the one that helps you most:

http://www.cisco.com/cisco/software/release.html?mdfid=283612660&softwareid=283118767&release=2.0%282f%29&relind=AVAILABLE&rellifecycle=&reltype=latest

-Kenny

[Robert Burns]

Amit,

Did you look on CCO?

The capability catalogue is included with each Infrastructure bundle for UCS. 

The UCS infrastructure bundles can be found here on CCO:

http://www.cisco.com/cisco/software/release.html?mdfid=283612660&flowid=22121&softwareid=283655658&release=2.0%282q%29&relind=AVAILABLE&rellifecycle=&reltype=latest

You can also download the individual Capability Catalogues from CCO here:

http://www.cisco.com/cisco/software/release.html?mdfid=283612660&flowid=22121&softwareid=283655658&release=2.0%282q%29&relind=AVAILABLE&rellifecycle=&reltype=latest

Regards,

Robert