Some Thoughts On Single Core Performance – Notebook vs. Datacenter

I guess you have noticed the one or other of my recent articles over the past few weeks about the experience I have gained with my new ‘refurbished’ workstation. While I previously thought that single core performance of  x86 processors in high end workstations and particularly in data centers must be so much more powerful than the CPU in my notebook, my recent experience tells me otherwise.

This whole investigation started when the videos I needed to encode for the Vintage Computing Festival Berlin this year took a long time to complete on my notebook. To speed things up, I rented a high end virtual cloud server, pushed the raw video material there but had to find out that I got a ‘meager’ speed-up of a factor of 2. I also got the same result when encoding videos on the CPU on my 6-core Xeon workstation. A factor of 2 over the notebook CPU was all I could reach (*). Also, when looking at the Geekbench results for various processors, even the most recent high end CPUs only reach a speed-up of 2-3x for single core performance compared to my 5 year old ultra low power notebook CPU. The Xeon CPU in my workstation is one generation ahead of my mobile CPU (i5-5300U vs. a Xeon E5-1650 v4) and ‘only’ delivers roughly 50% (0.5x) more single thread speed. So applications that are unable to split their work into threads that can run concurrently, high end workstations and data center blades do little for you. I find that a rather surprising result and something to be aware of when you think of high end hardware as far superior that what is in a notebook!

So where is the difference then? As I hinted above, workstations, blade server and data centers far surpass my notebook processing powers for applications that can split their work into concurrent threads and can thus use many CPU cores at the same time. In my tests, my Z440 workstation beats my notebook with speed-ups of 6x to 8x in that category. In data centers, servers with a massive number of CPU cores can obviously do a lot more things at the same time than my notebook can do and one application is to have many customers share the same server. But again, when it comes to single core performance, that data center resource is not much faster than that notebook on your desk.

(*) Note: When running the H.264 encoding on the GPU of the workstation I got a speed-up of 6x)