Benchmarking an inexpensive 16-CPU 128GB home server
This is the final post on the 16-CPU 128GB server I built earlier this year. I had the idea when I saw Intel released a new 10-core Core i7–6950X, which weighs in at $1749.99 on Newegg for the CPU alone. After reading up about inexpensive Xeon E-2670 v1’s which were less than $70 second-hand from decommissioned servers, I built an entire machine for around $500 less than Intel’s 6950X. For more information on the build, see Part 1 and Part 2.
Here’s a recap of the machine specs.
- 2 x Intel Xeon E5–2670 SR0KX (20MB cache and 8 CPU cores per socket)
- Hynix 128GB 16x8GB PC3–12800R RAM
- Samsung 850 EVO 2.5" SSD
The benchmark I’m using for the comparison is linux-bench. This is a great script which runs on Linux. It downloads and compiles a set of benchmarks, running them in sequence and recording all the performance values to a log file.
The tests I used are:
- 7zip: A popular tool to compress and decompress files.
- c-ray 1.1: A ray-tracing benchmark.
- NAMD: A molecular modelling benchmark
- OpenSSL: Open-source benchmark to sign and verify files using RSA 4096 bit keys.
These benchmarks are all designed to use all the available cores on the machine, and stress the CPU. The rest of the tests in linux-bench either didn’t compile and run on my machine, or I couldn’t find i7–6950X performance figures to compare to.
To compare the results with Intel’s i7 6950X, I used Anandtech’s CPU comparison page. Anandtech use the same linux-bench script to test Linux performance so we can make a direct comparison.
The 7zip compress benchmark showed a clear advantage to the 16-CPU server, with a performance of 63383 MIPS compared to Intel’s i7 6950X 49766. That’s 25% more performance!
The 7zip decompress benchmark also shows a similar performance gap of around 25% in favour of the 16-CPU server.
The c-ray 1.1 is a ray tracing benchmark, and is designed to be small enough to fit into a CPU’s cache. This means it’s a great test of raw CPU performance. The 16-CPU server completed the ‘Hard’ benchmark 6 seconds faster than the i7-6950X.
Results for the NAMD benchmark were closer, there was less than a second between the two systems.
The OpenSSL benchmarks were also very close, with less than 5% separating the performance of the two in both signing per second, and verifies per second.
Based on all the benchmarks I ran, the 16-CPU home server outperformed the Intel Core i7-6950X in each test (!). These benchmarks focused on raw CPU performance, and didn’t stress either RAM or storage. As the Core i7–6950X has a faster 4-channel DDR4 2400 memory interface, compared to the DDR3 1600 memory used in the home server, it’s likely to win any memory-intensive benchmark.