Same CPU single or dual configuration

[JC121]

hello,

i´m living and working in portugal. Start working in IT platforms and virtualization

sorry to start a thread, but could not find in the forum this.

I´m trying to choose some IT platforms to start working. The CPU rating came to the discussion in your group.

We are trying to understand how much a CPU is better then another, and we are using also your CPU rating to evaluate several commercial offers.

Nevertheless, when evaluating for instance an Xeon E5-2680v3 i´ve seen that in a single configuration rating is 18863, but looking at the dual configuration (the one we are trying to buy) we see a rating of 25710.

What does this exactly means?
a) I was expecting to find almost the double rating (2x18863) in the dual configuration of the Xeon.
b) Does this mean that in a dual configuration (assuming the memory is double), the performance of the same CPU as decreased almost by 36%, in relation to single configuration?
c) Or it means that the performance of a single Xeon in a dual configuration as increased 36%?

regards
jc

[Tim (PassMark)]

The implication is that adding a second CPU doesn't double overall performance.

Doubling the CPU count doesn't improve the performance of single threaded tasks, nor improve the speed of the RAM.

You'll probably find going from 1 core to 2 cores does nearly double performance. Even from 4 to 8 cores might come close to double. But when you are talking about going from 24 cores to 48 cores you are at the point of diminishing returns.

For some very specialised software you might get more than 36% improvement for the second CPU. But most software will be bottlenecked by other factors. At lot of software will get zero gain from the second CPU (that is to say it will be limited by disk speed, internet speed, RAM, or being single threaded).

[JC121]

Tim , thank you.

https://www.pugetsystems.com/labs/ar...625/#TestSetup

i found this article about using one or two CPUs. One socket board or two socket boards. Some surprises there (for me) well written and it subscribes our conclusions : for most sw the number ot threads it can use stays within values that are supported in a single processor. adding a a second one to the equation increases very few to none in real world applications.

After this, we are now using affinity to constrain applications to use cores inside a NUMA node to put things in the fast track. we have applications that use 20 cores (HT included).