Help with Turbo Boost (E5-2687W)

[idata]

HI all,

Lately we have purchased new HP workstation (Z820) equipped with dual E5-2687W. The turbo multiplier for this particular processor is 3/3/3/4/4/5/5/7, which means the maximum clock speed is 3.8 GHz with only 1 core active. We installed RHEL 5.8, enabled turbo boost, and Intel Enhanced Speed Step in the BIOS. What I found out when I am trying to run a single threaded program, it is actually slower than our 3 years old workstation with dual W5590 by 10 - 20%. Upon further investigation, it seems like even when I load just one of the core, all the other core in that particular socket turbo up, hence the maximum turbo boost clock speed is just a meager 3.4 GHz . My question is why would all the cores turbo up even when I am just running a single threaded application and I used taskset to make sure that the application wasn't bouncing among the cores. Anyone experience the same problem? I am using i7z to monitor the clock frequency and also the C state of the processor.

Chase

[idata]

There must be something that is requesting computer resources and getting assigned to the other cores.

You may want to check if the software application in use is optimized for the W5590 (pre configured micro code) and not for the newer E5-2687W. If the software application in use is not optimized for the current processor in use, this would be an expected behavior.

Intel® Enhanced Speedstep Technology activates when the processor is under a low utilization while the Intel® Turbo Boost Technology comes in when there is high processor utilization.

[idata]

Thanks. I now find out the problem is partly due to the CPU not enter C3 and C6 sleep state. We are using RHEL 5.8 and found that the CPUs only go to C1 but not higher C state even the workstation is idling, which explains why all the cores are turbo up even when I only loaded 1 core because from Intel's documentation, C1 core is considered an active core. Requested help from HP and Redhat, but nobody seems to be able to figure out why they cannot entered higher C state. Most the time they will cite that "Intel® Turbo Boost technology core frequency upside availability is ultimately constrained by power delivery limits, but within those constraints, it is limited by the following factors: • The estimated current consumption of the processor • The estimated power consumption of the processor • The temperature of the processor" and conclude that the CPU has exceeded those limit and hence not be able to reach the maximum performance level. I disagree that this is the cause because the E5-2687W is suppose to be a 150-W part and the temperature is just slightly above 40 C.

[OGodi]

Agree!

I am having a similar issue on Dell M620 with e5-2680, I am able to get 3.2Ghz boost at most even if I disable all cores from being active and have just the single core to run, Temp is 44-51C at most,

[OGodi]

I have an update,

on our Dell M620 with e5-2680 having disabled CPU cores in BIOS and having the BIOS set to max performance!

echo 1 > /sys/devices/system/cpu/cpuX/online ;will NOT do the job

• 2 cores enabled = 3.5Ghz turbo boost (max spec of the e5-2680)
• 4 cores enabled = 3.2Ghz turbo boost
• 6-8 cores enabled = 3.1Ghz turbo boost per core

so my conclusion is that turbo boost MAX frequency is determined by the machine at boot time according to the amount of cores active and NOT by the OS (at lease when BIOS is controlling MAX performance),

the OS will effect frequency between C1-C6 and the MAX turbo of all cores on a socket, ~1.5Ghz-MAX (3.1-3.5Ghz on e5-2680)

I will further play with passing the control to the OS and see if the Dynamic turbo boost can get to higher Ghz when controlled by the OS,

[OGodi]

update,

setting the BIOS to OS controlled instead of performance forced by BIOS (last reply) allowed my system to go up to the CPU max spec (3.5Ghz) on active cores that perform benchmark while keeping in active core on 1.2Ghz

so I am now able to get both max performance when job runs and max power save when idle...