English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The rapid adoption of AI and convenience offered by cloud services haveresulted in the growing demands for GPUs in the cloud. Generally, GPUs arephysically attached to host servers as PCIe devices. However, the fixedassembly combination of host servers and GPUs is extremely inefficient inresource utilization, upgrade, and maintenance. Due to these issues, the GPUdisaggregation technique has been proposed to decouple GPUs from host servers.It aggregates GPUs into a pool, and allocates GPU node(s) according to userdemands. However, existing GPU disaggregation systems have flaws insoftware-hardware compatibility, disaggregation scope, and capacity. In thispaper, we present a new implementation of datacenter-scale GPU disaggregation,named DxPU. DxPU efficiently solves the above problems and can flexiblyallocate as many GPU node(s) as users demand. In order to understand theperformance overhead incurred by DxPU, we build up a performance model for AIspecific workloads. With the guidance of modeling results, we develop aprototype system, which has been deployed into the datacenter of a leadingcloud provider for a test run. We also conduct detailed experiments to evaluatethe performance overhead caused by our system. The results show that theoverhead of DxPU is less than 10%, compared with native GPU servers, in most ofuser scenarios.

DxPU: Large Scale Disaggregated GPU Pools in the Datacenter