<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi" -->

---
title: Memory-Enforced GPU Sharing in Kubernetes with NVIDIA...
description: CUDA 13.4 adds a memory boundary that enforces exact or ranged GPU memory limits per workload, transparent to the application. A new open-source Kubernetes...
canonical: https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4 | daily.dev
og:description: CUDA 13.4 adds a memory boundary that enforces exact or ranged GPU memory limits per workload, transparent to the application. A new open-source Kubernetes...
og:url: https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi
og:image: https://api.daily.dev/og/posts/MZnSzRFHI.png
og:image:alt: Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4

**[NVIDIA Developer](https://daily.dev/sources/nvidiadev)** · 7 min read · 0 upvotes · 0 comments

## Summary

CUDA 13.4 adds a memory boundary that enforces exact or ranged GPU memory limits per workload, transparent to the application. A new open-source Kubernetes operator called gpu-fractioning, hosted under the KAI Scheduler GitHub organization, applies these limits on each node. Starting with v0.18.0, KAI Scheduler integrates with it through a GPU-sharing mode called NvFractions, letting developers set per-container memory requests, burst limits, and compute mode (time-slicing or sm-sharing) via pod annotations. A workload can request, for example, 40Gi guaranteed with a burst to 60Gi across four GPUs.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://developer.nvidia.com/blog/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4>

## Questions this post answers

### How do I enforce a hard memory limit per workload when sharing a GPU in Kubernetes?

CUDA 13.4 adds a memory boundary that caps each workload's GPU memory, enforced transparently so an allocation above the configured size fails just as it would on a smaller GPU. The open-source gpu-fractioning operator, run alongside the NVIDIA GPU Operator, applies per-container memory request and limit annotations on each node to enforce this.

_Teams building multi-tenant GPU clusters can track CUDA and Kubernetes GPU-sharing updates like this on daily.dev._

### What's the difference between time-slicing and sm-sharing compute modes in CUDA MPS?

Time-slicing is serial: workloads take turns on the whole GPU, each getting roughly 1/n of the time, with full error isolation per workload. SM-sharing is parallel: the GPU's streaming multiprocessors are split and all workloads run continuously, with the split settling around 1/n under load, but all workloads share a single error domain.

_Engineers choosing a GPU compute-sharing mode for trusted vs. untrusted workloads can follow this tradeoff on daily.dev._

### Which Kubernetes scheduler version adds fractional GPU memory sharing with CUDA 13.4?

KAI Scheduler adds a GPU-sharing mode called NvFractions starting with version v0.18.0, which lets it pick a GPU slice for each workload and write per-container annotations that the gpu-fractioning operator then enforces on the node. It can be enabled via a Helm install flag, global.nvFractions.set=true.

_Platform teams rolling out GPU-sharing schedulers can keep up with releases like this on daily.dev._

## Similar posts on daily.dev

- [GPUs on Kubernetes are bring-your-own-everything, starting with the bill](https://daily.dev/posts/gpus-on-kubernetes-are-bring-your-own-everything-starting-with-the-bill-7wb27xeuf) · The New Stack · 0 upvotes · 0 comments
- [How to Run Isolated Tenant Kubernetes Clusters on Shared GPU Infrastructure](https://daily.dev/posts/how-to-run-isolated-tenant-kubernetes-clusters-on-shared-gpu-infrastructure-g9yhk1w1z) · NVIDIA Developer · 0 upvotes · 0 comments
- [GPU Scheduling and Bin-Packing in Kubernetes: Pack More AI onto Every GPU](https://daily.dev/posts/gpu-scheduling-and-bin-packing-in-kubernetes-pack-more-ai-onto-every-gpu-6nuaafnff) · Cast AI · 0 upvotes · 0 comments
- [Ensuring Balanced GPU Allocation in Kubernetes Clusters with Time-Based Fairshare](https://daily.dev/posts/ensuring-balanced-gpu-allocation-in-kubernetes-clusters-with-time-based-fairshare-t7sb4ftnr) · NVIDIA Developer · 0 upvotes · 0 comments

---

Tags: [#kubernetes](https://daily.dev/tags/kubernetes), [#nvidia](https://daily.dev/tags/nvidia), [#mlops](https://daily.dev/tags/mlops), [#cuda](https://daily.dev/tags/cuda)

[View this post on daily.dev](https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4","url":"https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi"},"datePublished":"2026-10-27T20:52:23.311Z","dateModified":"2026-10-27T21:14:51.751Z","description":"CUDA 13.4 adds a memory boundary that enforces exact or ranged GPU memory limits per workload, transparent to the application. A new open-source Kubernetes...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/bb54411844a6a0079001524d7b120b39?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/bb54411844a6a0079001524d7b120b39?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"NVIDIA Developer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"NVIDIA Developer","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/86e45aab42ba48ce83103d01b1119910","url":"https://daily.dev/sources/nvidiadev"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"kubernetes,nvidia,mlops,cuda","timeRequired":"PT7M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"NVIDIA Developer","item":"https://daily.dev/sources/nvidiadev"},{"@type":"ListItem","position":3,"name":"Memory-Enforced GPU Sharing in Kubernetes with NVIDIA CUDA 13.4"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/memory-enforced-gpu-sharing-in-kubernetes-with-nvidia-cuda-13-4-mznszrfhi#faq","mainEntity":[{"@type":"Question","name":"How do I enforce a hard memory limit per workload when sharing a GPU in Kubernetes?","acceptedAnswer":{"@type":"Answer","text":"CUDA 13.4 adds a memory boundary that caps each workload's GPU memory, enforced transparently so an allocation above the configured size fails just as it would on a smaller GPU. The open-source gpu-fractioning operator, run alongside the NVIDIA GPU Operator, applies per-container memory request and limit annotations on each node to enforce this. Teams building multi-tenant GPU clusters can track CUDA and Kubernetes GPU-sharing updates like this on daily.dev."}},{"@type":"Question","name":"What's the difference between time-slicing and sm-sharing compute modes in CUDA MPS?","acceptedAnswer":{"@type":"Answer","text":"Time-slicing is serial: workloads take turns on the whole GPU, each getting roughly 1/n of the time, with full error isolation per workload. SM-sharing is parallel: the GPU's streaming multiprocessors are split and all workloads run continuously, with the split settling around 1/n under load, but all workloads share a single error domain. Engineers choosing a GPU compute-sharing mode for trusted vs. untrusted workloads can follow this tradeoff on daily.dev."}},{"@type":"Question","name":"Which Kubernetes scheduler version adds fractional GPU memory sharing with CUDA 13.4?","acceptedAnswer":{"@type":"Answer","text":"KAI Scheduler adds a GPU-sharing mode called NvFractions starting with version v0.18.0, which lets it pick a GPU slice for each workload and write per-container annotations that the gpu-fractioning operator then enforces on the node. It can be enabled via a Helm install flag, global.nvFractions.set=true. Platform teams rolling out GPU-sharing schedulers can keep up with releases like this on daily.dev."}}]}
```