
HAMi is a @CloudNativeFdn open-source GPU virtualization middleware for sharing, isolating, and scheduling heterogeneous accelerators on Kubernetes.
Earth
Joined July 2026
- Tweets70
- Following56
- Followers47
- Likes48
Your Kubernetes cluster has free GPU memory, but new Pods are still stuck in Pending. Why?
GPU fragmentation.
See how it works and try it out 👇
🔗 github.com/Project-HAMi/gpu-…
#HAMi #Kubernetes #GPUSharing
4
🟢 HAMi Community Weekly #01
A big week for HAMi!
Here's a quick recap:
#HAMi #OpenSource #Kubernetes #GPU #CloudNative
26
🎬 HAMi v2.11.0 in 76 seconds!
Remote GPU, AMD GPU sharing, Ascend ENPU, NUMA-aware scheduling & Flexible MIG.
Catch up on what's new in our latest release! 💚
🔗 project-hami.io/blog/hami-v2…
#HAMi #Kubernetes
1
35
🎉 Congratulations to Jimmy Song on receiving the 2026 CNCF Lorem Ipsum Award!
Proud to have you as part of the HAMi community 💚
Live at KubeCon + CloudNativeCon North America: Congratulations to our 2026 Lorem Ipsum Award winners Natali Vlatko, @jimmysongio, and Dylan Reimerink!
Clear documentation across Kubernetes, HAMi, Cilium, and eBPF makes cloud native software usable for everyone.
#KubeCon #CloudNativeCon
1
93
📸 Scenes from #KubeCon NA 2026!
Great conversations, new connections, and lots of interest in GPU sharing and virtualization.
Thanks to everyone who stopped by to connect with the HAMi community! 💚
#HAMi #Kubernetes #OpenSource
65
HAMi joins the Vultr Open Source Alliance as a founding member!
Together with @Vultr, we're advancing open, efficient GPU infrastructure for AI workloads on Kubernetes.
Excited for what's ahead!
🔗 project-hami.io/blog/hami-jo…
#HAMi #OpenSource
3
61
📸 Great day at #KubeCon NA 2026!
Two HAMi sessions in Salt Lake City:
🎤 Maintainer Track — GPU scheduling, hardware support & DRA
🛠️ ContribFest — Hands-on GPU virtualization & device plugins
Thanks to everyone who joined us! 💚
#HAMi #Kubernetes
2
112
Live from #KubeCon NA 2026!
Dynamia's Mengxuan Li and Red Hat's Yuchen Fama are showcasing scheduler-agnostic GPU sharing with DRA + Kueue + HAMi + vLLM/llm-d.
From GPU allocation to dynamic slicing and AI inference, all in one architecture.
#HAMi #Kubernetes #GPUSharing
3
112
🚀 HAMi v2.11.0 is here!
What's new?
🔹 Remote GPU scheduling
🔹 AMD GPU sharing
🔹 Ascend ENPU
🔹 NUMA Refit
🔹 Flexible MIG + CDI
220+ PRs merged from 50+ contributors, including 29 first-timers! 💚
🔗 project-hami.io/blog/hami-v2…
#HAMi #Kubernetes #GPU #OpenSource #CloudNative
🚀 10B+ daily tokens. ~90% GPU utilization.
In a new @CloudNativeFdn case study, Seres shares how SeresCloud uses Kubernetes + HAMi to scale AI inference with fine-grained GPU sharing and scheduling.
📖 cncf.io/case-studies/seres/
#HAMi #Kubernetes #AIInfra
2
46
🚀 HAMi is heading to #KubeCon NA 2026!
Join us on Nov 10 in Salt Lake City for two sessions:
🟩 HAMi Levels up: Smarter Scheduling, Wider Hardware Reach and a DRA-Native Lightweight Future
🟩 GPU Virtualization: First Setup and Understanding the Device Plugin / DRA
1
35
🎥 Missed the latest HAMi Community Meeting?
Last week: AMD GPU sharing benchmarks, documentation updates, PR reviews & energy-aware scheduling discussions.
▶️ Watch the recording: youtu.be/hcWBfDq4kJ8?si=lGMl…
#HAMi #Kubernetes #OpenSource
🤖 Made with AI
36
HAMi is heading to ContainerDays & AI Context Singapore 2026! 🇸🇬
@jimmysongio @fishmanticoder
Join us on Oct 27 :
・12:30 — Hands-on GPU Scheduling in Kubernetes with HAMi
・15:05 — Optimizing Neocloud GPU Utilization with HAMi
Explore GPU sharing & heterogeneous scheduling.
Multi-Tenant GPU Kubernetes on Simulated Hardware with Kamaji and HAMi anishbista.org/blog/multi-te…
Can HAMi and @Linkerd work together to tackle GPU noisy neighbors? Our community contributor Mesut Oezdil explores GPU isolation, latency-aware load balancing, and tenant security in a hands-on lab. Thanks for the great contribution!
project-hami.io/tutorials/la…
49
Project HAMi retweeted
Prefill and Decode Disaggregation with llm-d: A KubeCon China 2026 Keynote Explained
kubezilla.io/llm-d-prefill-d…
Project HAMi retweeted
Most schedulers check GPU limits only at admission. HAMi checks at admission, then HAMi-core enforces them on every allocation at runtime.
That's why NVIDIA's KAI Scheduler adopted HAMi-core.
Check out my new blog:
medium.com/@maishivamhoo/who…
@HAMiProject @CloudNativeFdn
Love to see it! 🙌,
Let's Goooo 😎😎
The best part of the LFX Mentorship is the exposure you get to so many new things.
Currently working on @HAMiProject as an LFX mentee, and I got access to an NVIDIA A16-16Q GPU from @Vultr.
Really enjoying the work!😎😎😎
57
8x more models per GPU without buying a single new card.
That's what fine-grained slicing looks like in production. Congrats to Sangfor on the CNCF case study built on Kubernetes with HAMi + Volcano.
To support 2,000+ engineers running concurrent AI workloads, Sangfor implemented vGPU slicing to run 8x more models per GPU.
By pairing Kubernetes with Volcano and HAMi, they tripled GPU utilization across production.
Read the case study: cncf.io/case-studies/sangfor…
Thanks to the @AI_TCG_CNCF for having us today!
Mengxuan Li, Dynamia Co-founder & CTO and HAMi author, showed how HAMi gets more out of heterogeneous AI clusters: GPU sharing, isolation & topology-aware scheduling on K8s.
Thanks to everyone who joined.
































