Google Cloud Tau T2A Ampere Altra vs. T2D AMD EPYC Performance

Written by Michael Larabel in Processors on 21 September 2022. Page 2 of 5. 8 Comments

For OpenJDK Java workloads such as those run with Renaissance, the Tau T2A series performed competitively with AMD. While at the lower vCPU counts the T2A VMs offered slightly better performance, at 48 vCPUs it was a tie but the T2D AMD VMs have the benefit of scaling up to 60 vCPUs.

Meanwhile with the demanding SPECjbb benchmark, the estimated performance had the AMD EPYC powered T2D VMs delivering much better performance over the T2A series. At 48 vCPUs, the estimated composite max-jOPS was 64% faster with the EPYC T2D instance over T2A.

The Apache Spark data analytics engine performance was competitive between both Tau VM classes. This has come after significant investment by the Java/OpenJDK community in squeezing out optimized support on AArch64 and ensuring its software ecosystem support is in good shape.

For SHA-512 operations within Apache Spark, the Tau T2D instances were leading over the Arm T2A instances.

The OpenSSL 3.0 SHA256 performance was closely matched between the T2A and T2D instances across the tested vCPU sizes. Again, with the AMD advantage of being able to offer up to 60 vCPUs over the T2A series currently taping out at 48 vCPUs -- we'll see though if that changes in the future such as if Google begins offering any Ampere Altra Max powered instances.

While the SHA256 performance was in good shape for T2A, with OpenSSL 3.0's RSA4096 benchmark out-of-the-box on Ubuntu 22.04 LTS, the Tau T2D series was scaling much better than the T2A VMs.


Related Articles