The AMD CPU provided just over a 33% improvement vs the Intel CPU’s for single threaded workloads. For multi-threaded workloads we would expect improvements far beyond that.
Averaging the i7 results gives a computation time of 6.55ns in this task, with the GMC machine coming in slowest (notably, with Hyper Threading disabled on this machine) While on AMD’s side we saw a completion time of 4.32ns
Obviously these are just small tests, and in the real world these improvements would be amplified by more than a few nanoseconds, which is why we’re excited to start rolling out AMD 3800x’s globally!
For team blue we used three overclocked intel i7 7700k machines to get our base average. Two of the machines were provided by OVH, and one of the machines was provided by GMC. The GMC machine had Hyper Threading disabled.
For team red we used a 3800x provided by OVH USA. We’re going to add onto this post with additional 3800x results in the UK when we get our hands on one.
All machines were running Ubuntu 16.04, and all machines had no load other than the stress test on them. All machines had the same performance CPU frequency scaling settings.
Because Garry’s Mod is single threaded, we’re only testing the performance with 1 thread. For the test we used sysbench with the following arguments:
sysbench --num-threads=1 --test=cpu --cpu-max-prime=10000 run
We decided to use this benchmark method as we knew it was also favoured by other players in the market such as GMC.
The results (lower is better)
Team Blue (Intel i7 7700k OC)
OVH (i7 7700k OC)
GMC (i7 7700k OC)
Team Red (AMD 3800x)
OVH (AMD 3800x)