Intel Sapphire Rapids Performance With Google Cloud Compute Engine C3

Back in October Google announced their Compute Engine C3 instances in private preview that featured 4th Gen Xeon Scalable "Sapphire Rapids" processors as well as making use of Google's custom Intel Infrastructure Processing Unit (IPU). Since then, back in January, was the big Sapphire Rapids launch with the likes of the Xeon Platinum 8490H being released. Last month meanwhile Google promoted the C3 VMs to public preview state. The Sapphire Rapids C3 VMs remain in "public preview" from Google Cloud during which time there are no charges involved for the CPU costs. For those wondering about the core-for-core performance of Sapphire Rapids in Google Cloud, here are my initial benchmarks of the C3 series.

The C3 VMs on Google Cloud feature Sapphire Rapids processors along with the new custom IPU, Hyperdisk, and other improvements over earlier generations of their virtual machines. For today's article my focus was on kicking the tires with Sapphire Rapids CPUs in Google Cloud. While in public preview and offering them without any charges, only the "high CPU" instance types were made available to my account for testing. Additionally, c3-highcpu-22 and higher (22 vCPUs) were not available... Thus for today's testing is looking at just the Google Cloud VMs with 8 vCPUs given that the c3-highcpu=-8 was the highest Sapphire Rapids configuration I had access to during the Google Cloud benchmarking.

Like with earlier generations, the c3-highcpu-8 configuration features 8 vCPUs that are four physical cores plus Hyper Threading. There was 16GB of system memory available for the c3-highcpu-8 instance.

Comparing Sapphire Rapids in the Google Cloud is a bit peculiar since most of the other Intel VMs are making use of Intel Xeon Scalable "Cascade Lake" and not the prior generation Ice Lake. The Intel Ice Lake instances with Google Cloud are select those within the N2 and M3 series. Additionally, so far no 4th Gen AMD EPYC Genoa VMs are yet available in Google Cloud, thus for this round of benchmarking was seeing how the Sapphire Rapids performance delivers over the vast number of Cascade Lake VMs running on Google Cloud.

Given the Sapphire Rapids vCPU access constraints, the other tested VMs were all 8 vCPUs as well including the:

n2-standard-8: 8 vCPUs from an Intel Cascade Lake processor, 32GB of RAM.

c2-stadard-8: 8 vCPUs from an Intel Cascade Lake processor, 32GB of RAM but this being from Google Cloud's "compute" series.

n2-highcpu-8: 8 vCPUs from an Intel Cascade Lake processor but with 16GB of RAM, matching the "high CPU" vCPU:RAM ratio I had available to with the c3-highcpu testing.

Google Cloud documentation cites the Intel Xeon Gold 6268CL Processor as being used for the N2 machine series while the C2 series makes use of the Intel Xeon Gold 6253CL Processor.

Thee different machine instances and the c3-highcpu-8 VMs were all tested while making use of Ubuntu 22.10: not an LTS release, but in wanting a bleeding edge software stack while pushing Sapphire Rapids in the cloud, Ubuntu 22.10 is a better candidate for GCC 12.2 and other newer software updates over Ubuntu 22.04 LTS. All of these VMs also were using a 300GB SSD disk for running the Ubuntu 22.10 cloud deployment.

So for looking at the Sapphire Rapids performance in Google Cloud were a wide variety of benchmarks conducted that end up showing the core-for-core performance of Sapphire Rapids over the numerous Cascade Lake instances in Google Cloud. In addition, the current performance-per-dollar was looked at during this benchmarking too.