Home ยป Optimizing Server Performance: Cloudflare’s Latest Update Features AMD Genoa-X with Emphasis on Cache Efficiency over Core Quantity, 1152MB Cache Size

Optimizing Server Performance: Cloudflare’s Latest Update Features AMD Genoa-X with Emphasis on Cache Efficiency over Core Quantity, 1152MB Cache Size

Cloudflare has announced the upgrade of their servers to version 12 after updating to version 11 since 2021. This time around, they have stuck with AMD EPYC and have opted for AMD EPYC 9684X Genoa-X.

What’s interesting is that Cloudflare has selected a rough specification and narrowed down the models to be all AMD, including AMD EPYC 9654 Genoa, AMD EPYC 9754 Bergamo, and AMD EPYC 9684X Genoa-X. Each model is limited to a heat specification of not exceeding 400W.

The AMD EPYC 9684X Genoa-X is the top model in the lineup. Its standout feature is the 3D V-Cache technology, packing a cache of up to 1152MB. When tested with Cloudflare’s workload, it was found to outperform Bergamo, which has a higher core count. Additionally, the energy efficiency of the Genoa-X is superior, showing that a larger cache size can significantly enhance performance.

Although the number of cores has increased by 50% compared to servers from version 11, Cloudflare has decided not to increase the RAM as the workload has already optimized memory usage. Therefore, it was determined that 4GB of memory per core is sufficient, eliminating the need to increase total memory.

The servers this time around are designed to accommodate full-height, full-length PCIe cards, with a maximum of two cards and a combined heat release of 400W to support graphics cards running AI workloads that customers use in Worker AI applications.

TLDR: Cloudflare has upgraded their servers to version 12, sticking with AMD EPYC and introducing the high-end AMD EPYC 9684X Genoa-X model, featuring impressive 3D V-Cache technology. The servers are optimized for performance with a focus on energy efficiency, and no additional RAM is needed despite a 50% increase in core count. They are designed to support full-height, full-length PCIe cards for AI workloads.

More Reading

Post navigation

Leave a Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Alibaba’s RISC-V Chips Now Available on Scaleway’s Cloud Platform

Initiating Edge Computing Processing for AI Readiness with IBM Power S1012

The Surge of Carbon Emissions Unveiled by Google: A 48% Escalation Over the Past 5 Years Due to AI Expansion