Skip to main content
← Back to index page

v2.10.0

aaka24
adity1raut
AdityaShome
almazkhalikov
anandj91
aniket866
archlitchi
Aryanbhargava18
asadjan4611
ashergaga

Major features

  • Remove DRA components from the HAMi main Helm chart; Ascend DRA support has moved to the dedicated ascend-dra-driver repository by (@archlitchi) in #2038
  • Add init container support with correct GPU resource accounting by (@maishivamhoo123) in #1773
  • Add PodGroup (gang scheduling) support, including NodeLock retries in Bind and compatibility with the gangScheduling feature gates on Kubernetes v1.36+ by (@lin121291) and (@archlitchi) in #2066 and #2206
  • Add flexible MIG with dynamic MIG instance allocation and deallocation by (@FouoF) in #2378
  • Add a new mutex GPU scheduling policy and support comma-separated policy combinations by (@mesutoezdil) in #2011 and #2621
  • Add opt-in NUMA topology alignment (CPU and GPU) for vGPU replicas by (@saiyam1814) in #2065
  • Add support for Biren devices by (@DSFans2014) in #1711
  • Add AMD MI300x series vGPU support by (@FouoF) in #2290
  • Allow mode-agnostic Ascend pods to schedule across both vNPU-template and HAMi-core nodes by (@ouyangluwei163) in #2035
  • Add Ascend 910C vNPU templates vir05_1c_16g and vir10_3c_32g by (@ouyangluwei163) in #2005
  • Stop trimming device memory to template alignment when using HAMi-core mode for vNPU by (@archlitchi) in #2696
  • Switch the HAMi-core compile image to ubi8 to support all GLIBC versions by (@spencercjh) in #1958

Major bug fixes

What's changed

Other changes

New contributors

Committers: Contributors

Full Changelog: https://github.com/Project-HAMi/HAMi/compare/v2.9.0...v2.10.0

CNCFHAMi is a CNCF Incubating project