vCPU speed vs pCPU speed information

[chuckk281]

Customer question from Renjith:

*******

Hi Experts,

Our customer has a requirement of 3.0 GHz of CPU for one of the applications.

ESX hosts is running on BL460C Gen8 with Xeon CPU E5-26700 @ 2.60 GhZ, 2 socket , 8 core , HT active.

So I just want to confirm if we add more vCPU will that deliver more than 3.0 GHz or CPU speed still remains 2.60 GHz similar to physical CPU speed.

**************

Reply from Michael:

**********

If the application is multi-threaded then adding vCPUs will boost the performance of the application and should meet/exceed the applications requirements. 2 x 2.6GHz could provide the application with 2.6GHz per thread = 5.2GHz in total. If it’s single threaded however it can only take advantage of a single core and it won’t benefit from adding more vCPU thereby not meeting the requirements.

The Intel E5-2670 processor supports Intel Turbo Boost technology 2.0 up to 3.3GHz. This feature is described as follows:

“Intel® Turbo Boost Technology 2.0¹ automatically allows processor cores to run faster than the rated operating frequency if they’re operating below power, current, and temperature specification limits.”

http://www.intel.com/content/www/us/en/architecture-and-technology/turbo-boost/turbo-boost-technology.html

So you may get an increase that meets the customers’ requirements but it’s not guaranteed and is dependent on the Blade environment. If you run a heavily utilized system it’s unlikely to hit the higher clock speed during peak workloads. A virtual environment is more likely to be heavily loaded as it typically contains multiple mixed workloads unless you dedicate 1 pCPU each vCPU which is an approach for dense database farms.

You can discuss these options with the customer and application vendor to see what they would support. Failing that you should look at swapping out for faster CPU with 3.0GHz + clock speed….

***********

And from Chris:

**************

All what Michael said is correct.

However, I would remark that the customer requirement of 3.0 GHz core  is totally spurious without further clarification of other parameters that drive performance of the CPU such as:

• Processor architecture – number of functional units, pipeline design, pre-fetch stage, etc.
• Cache sizes and latency at different cache levels – L1, L2, L3 and TLB
• Memory speed

These architectural differences have huge impact on performance, as the following example shows:

DL380 G3 - 3.06GHz (512kB cache)   - performance rating - 1,330

DL380p Gen8 – E5-2690 v2 – 10 core – performance rating per core – 10,770.

So, my feeling is that your customer may care about 12% speed difference in notional clock speed, but ignores architectural differences that may result in 800% performance difference.

In other words the clock frequency is meaningless by itself without specifying the implementation of the architecture.

**************

Comments?