Again in August Google Cloud introduced the C3D cases powered by AMD EPYC 9004 “Genoa” processors whereas solely final week was C3D promoted to basic availability. Curious concerning the efficiency of C3D after being impressed by AMD EPYC Genoa bare-metal server efficiency at Phoronix in addition to what I’ve seen with Genoa within the cloud at Microsoft Azure and Amazon EC2 / AWS, listed here are some benchmarks of the brand new C3D up towards different GCE cases.
In Google’s announcement final week they famous the overall objective C3D VMs can supply as much as 45% efficiency will increase over prior era VMs. Given what I’ve seen out of Genoa and the nice Zen 4 advantages like AVX-512 and transitioning to DDR5 system reminiscence, these claims aren’t all that shocking. Among the specifics Google shared of their announcement was round 54% higher NGINX efficiency, MySQL and PostgreSQL database servers as much as 62%, and in-memory databases like Redis as much as 60%.
Google Cloud presents C3D VMs as much as 360 vCPUs (the C3D vCPUs are made up of a mixture of bodily cores and SMT sibling threads) and as much as 2.8TB of DDR5 reminiscence. For the needs of this preliminary Google Cloud C3D benchmarking on Phoronix, I used to be specializing in the 60 vCPU C3D VM occasion. For this text the next occasion varieties had been in contrast:
c3d-standard-60: The brand new C3D AMD EPYC Genoa occasion with 60 vCPUs, powered by AMD EPYC 9B14.
c2d-standard-56: The prior era AMD EPYC Milan occasion with EPYC 7B13 processor. This occasion was at 56 vCPUs with not having a 60 vCPU possibility.
n2d-standard-64: The AMD EPYC Rome based mostly occasion with EPYC 7B12 processors. Right here the closest sized occasion was 64 vCPUs.
c2-standard-60: The Intel Xeon Cascade Lake competitors at 60 vCPUs for reference.
Sadly for Google’s new C3 machine varieties which are powered by the latest-generation Sapphire Rapids processors, they’re sized at 22 / 44 / 88 vCPUs (amongst different smaller and bigger sizes). So sadly nothing within the 56~64 vary as examined with the opposite machine varieties, which is why there wasn’t any C3 occasion examined for this text because of the starkly completely different sizing. In any case this text is usually targeted on the AMD EPYC generational efficiency in Google Cloud.
All the examined cases had been on Ubuntu 22.04 LTS. Along with uncooked efficiency the performance-per-dollar based mostly on the present hourly pricing was additionally tabulated.
