Up nextNext guide
Hugging Face Accelerate for Multi-GPU Training
Technical
Technical GUIDE
Multi-Instance GPU Partitioning
NVIDIA Multi-Instance GPU partitions supported GPUs into instances with assigned compute and memory resources.
Listen to this guideAI-narrated · ~2 min
On this page2 min read
Overview
It can improve sharing for smaller workloads, but profile support and isolation boundaries depend on the GPU and configuration.
Deep Dive
MIG was introduced with the NVIDIA A100 and is supported on selected later GPUs. A GPU Instance groups compute and dedicated memory resources, including assigned paths through cache and memory hardware. On an A100 40 GB, supported layouts can provide up to seven small GPU Instances or fewer larger ones. Compute Instances can further divide compute resources inside a GPU Instance while sharing its memory resources. This differs from ordinary software time slicing, but does not imply immunity from every device-level fault or interference in the wider host and service. Workload capacity, supported profiles, and operational constraints should be tested on the actual GPU.
Real-World Implementation
A cloud provider splits one A100 into seven instances so seven customers each get a guaranteed, isolated GPU slice for inference.
A university research cluster gives each PhD student a 10GB MIG instance for prototyping instead of monopolizing whole cards.
An inference service packs several small language and vision models onto one H100, each in its own partition with predictable latency.
A Kubernetes cluster advertises MIG instances as schedulable resources so pods request 'nvidia.com/mig-1g.5gb' like any other resource.
The Future of Multi-Instance GPU Partitioning
As GPUs grow to 80GB, 141GB, and beyond, partitioning becomes more attractive because individual models rarely need a whole card for inference. Expect tighter Kubernetes and cloud integration, dynamic repartitioning without draining the node, and finer-grained profiles. Competing vendors are pursuing similar SR-IOV-style GPU virtualization, and serverless inference platforms increasingly rely on partitioning to pack many models densely and cut idle waste.
Putting it into practice
Why it matters
- Cost and budget. Architecture decisions drive performance and operating cost for years.
- Clearer decisions. Technical education helps teams choose the right stack, not just the newest one.
- Quality control. Better engineering choices reduce reliability incidents in production.
Watch out for
- Optimizing one benchmark can hide broader system weaknesses.
- Infrastructure and maintenance costs are often underestimated.
- Security and observability gaps can grow as systems become more complex.
Next steps
- Define latency, quality, and cost targets before implementation.
- Benchmark under realistic load and data conditions.
- Instrument monitoring for errors, drift, and user impact.
- Prepare rollback and incident response paths before scaling.
Keep Exploring
Free newsletter
Keep up with AI in 3 minutes a day
One short email each weekday with the three AI stories that actually matter. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Multi-Instance GPU Partitioning quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Frequently asked questions
What is Multi-Instance GPU Partitioning?
NVIDIA Multi-Instance GPU partitions supported GPUs into instances with assigned compute and memory resources. It can improve sharing for smaller workloads, but profile support and isolation boundaries depend on the GPU and configuration.
Which resource boundary distinguishes separate MIG GPU Instances from software time slicing?
MIG assigns hardware resource slices. Compute Instances within a shared GPU Instance have a different boundary, and device-level faults are not universally eliminated.
On which NVIDIA architecture was MIG first introduced?
MIG debuted with the Ampere-based A100 and continued on H100 and later data-center GPUs.
On an A100 40 GB using supported small-instance profiles, how many GPU Instances can a layout provide?
The A100 supports layouts of up to seven GPU Instances. Maximum counts and profiles should not be generalized to all GPU models.
In a MIG profile like '1g.5gb', what does the '5gb' represent?
The profile encodes compute slices (1g) and the dedicated memory (5GB) given to the instance.
Which workload is MIG especially well suited for?
MIG shines when many small, isolated workloads (like inference) share one card with guaranteed quality of service.
Keep learning
Related guides
More guides picked for this topic