B230 BIOS Setting for Intel Speedstep

[Lewis Benton]

I was curious if anyone has been able to disable Speedstep on a B230 blade?  I've set a custom BIOS policy with it disabled, assigned it to the Service Profile for the blade and then rebooted the blade, but it still shows as enabled inside the BIOS screen.

Speedstep is not an option on the BIOS Defaults (under Servers->Policies-BIOS Defaults in UCSM) and it is greyed out inside the BIOS screen.

I'm assuming that is what VMWare noticed while looking at another issue as throttling my CPU performance by 50%.

Thanks in advance,

Lewis Benton

[Robert Burns · Accepted solution]

I discussed this with our Dev team.  This is expected behavior.

With our Intel EX Chipset blades (B230 & B440) there are implicit levels of power capping enforced.  UCS manages the Power Capping of blades using the P-states features of the chipsets.  By disabling Speedstep this essentially disables the P-State functionality and UCSM power capping can not be enforced.  This is all done to protect the Chassis and blades against power related issues in extreme corner cases.  It's quite rare to ever see any UCS deployment where EX blades have been capped because of this.

We have an open bug to have the BIOS token completely removed from the B230/B440 so it will no longer be visible in the BIOS.  As you can see we've already removed the "BIOS Defaults" option for these two platorms.  This will be corrected in a future 1.4 patch release.

Regards,

Robert

[Lewis Benton]

Thanks Robert,

I'll have to figure out what to tell VMWare then since they want me to turn it off.  Does that mean that the capabilities of the processor are being restricted, for example clock-speed?

Thanks again,

Lewis Benton

[Robert Burns]

We have your case (TAC) with VMware.  We'll inform them direclty.

Regards,

Robert

[Robert Burns]

Lewis,

Just to be clear, performance of the EX blades are NOT inhibited by power capping under normal or even heavy utilisation.  It's simply a defense mechanism to prevent any problems occuring in extreme cases (such as running burn-in or benchmarking applications which do not simulate BAU operation).  It's just ceiling protection for the system.

Regards,

Robert

[Robert Burns]

Just a notice to advise that a new BIOS has been released - 1.4(3) which addresses a few bugs including one involving VMware performance on Intel Nahalem - EX blades including the B230.

Resolved Caveat snippet from Release Notes:

"A Nehalem -EX based blade running VMware  hypervisor no longer sees lower VM efficiency than expected when  compared to other blades using a different Intel architecture.  (CSCtq00382)"

FYI - The bug above is not related to the inability to disable speedstep in the BIOS of EX blades as this thread was originally created.  As explained before, Speedstep is disabled as a Power Capping requirement.  There has been some confusion of these two issues being related, but they are separate issues.

Full release notes available here:

http://www.cisco.com/en/US/partner/docs/unified_computing/ucs/release/notes/OL_24086.html

Regards,

Robert