New Google Cloud Tau VM Benchmarks

Recently DoiT International obtained access to pre-release versions of the newest Tau VMs (T2D instance type) from Google Compute Engine. The T2D family of instances are based on 3rd Gen AMD EPYC processors and offers significant performance improvements over similar instances from other cloud providers. DoiT’s Customer Reliability Engineers conducted a side-by-side comparison testing compute instances on Google Cloud and AWS platforms.

Benchmarking Google Tau instances

Since the tests on Google Cloud Platform were performed with a pre-release compute instance class, PerfKitBenchmarker required several modifications to make it work. The command line below is expected to work when the machines are publicly available. For the AWS test, the PerfKitBenchmarker was used as-is. An example command is shown below:

python3 pkb.py --cloud=GCP --benchmarks=coremark --coremark_parallelism_method=SOCKET --machine_type=t2d-standard-16 --os_type=ubuntu1804 --run_stage_iterations=50 --zone=us-central1-f

python3 pkb.py --cloud=AWS --benchmarks=coremark --coremark_parallelism_method=SOCKET --machine_type=m6g.4xlarge --os_type=ubuntu1804 --run_stage_iterations=50 --region=us-east-1a

GCE Tau t2d-standard-16

CoreMark 1.0 : 469318.315147 / GCC7.5.0 -O2 -g -O2 -DMULTITHREAD=16 -DUSE_SOCKET -DPERFORMANCE_RUN=1 -DPERFORMANCE_RUN=1 -lrt / Heap / 16:Sockets

Amazon EC2 Gravitron m6g.4xlarge instance

CoreMark 1.0 : 311939.483740 / GCC7.5.0 -O2 -g -O2 -DMULTITHREAD=16 -DUSE_SOCKET -DPERFORMANCE_RUN=1 -DPERFORMANCE_RUN=1 -lrt / Heap / 16:Sockets

Given an exact test with equivalent cores of an equivalent system, GCE Tau family instances are 50% better than Amazon Gravitron2 based EC2 M6G instances. This enables customers to optimize their workloads for performance and cost on T2D Google Compute Engine instances.