I’ve been quite impressed with the performance of the Chromebook that has become my primary portable system. It is an Acer Chromebook 11 C740. The CPU is a dual-core Celeron 3205U running at a nominal 1.5 GHz. It has 4 GB of RAM and a 256 GB SSD drive. I’m running this system in Developer mode which gives shell access. I’ve created a chroot that is running a current version of Ubuntu. This is running within the ChromeOS system. I decided to do some testing to see what the real values were. This is the new system.

The big system was custom built a few years ago and has multiple spinning disks, the smallest of which is 400 GB, and 16 GB of RAM. The CPU is an eight-core AMD FX-8120.

The old system is a Dell Insprion E1405 purchsed nearly 9 years ago. The CPU is a dual-core Core 2 Duo T7200 running at a nominal 2.0 GHz. It has 3.2 GB of RAM and a 320 GB spinning disk. This has become my primary desktop as its hinge has broken and it is no longer usable as a laptop.

Michi-Go

The process that got my attention is a program I translated from Python to the Go language that plays the Go game. This program reads a couple of large files once at the start, but is processor-bound for the rest of execution.

There are a couple benchmark test built into the program. The first, mcbenchmark, reads the pattern files and then does 20 playouts of the initial, empty, board. This benchmark is single-threaded. The second, tsbenchmark, starts with the initial, empty, board and generates a move. This requires 1400 playouts in the current version of the program. This benchmark is multi-threaded.

The results for this set of test are:

There is some indication that the Chromebook, new system, may be faster per thread in the tsbenchmark than the big system. I’ll need to look at this in more detail.

Unix Bench

This is an old series of benchmarks used to evaluate the relative performance of servers. This is version 5.1.3 of the Byte Unix Benchmarks. I’m only reporting the overall score for each system. The default is to do a single-threaded run and a multi-threaded run with the number of parallel processes equal to the number of cores available.

These are the results:

I should probably try to fill in some of the gaps in this data to better understand the differences in the operation of these three systems.

Conclusions

Not too much to draw at present, but one thing is pretty clear: The modern Celeron is better than the old mobile Core 2 Duo, even at a slower clock speed.