NVIDIA A100 · High-Performance Computing
A100 40 GB
- GPUNVIDIA A100 (40 GB) · DedicatedNVIDIA A100 (40 GB) · Shared
- vCPU12 Cores · XEON Gold
- RAM64 GB · DDR4
- Storage500 GB · NVMe SSD
- Bandwidth2 TB p/m
- IP & DDoSIncluded · Shield
Minimum specification. Scale CPU, RAM, and storage in the calculator.
Minimum 12 month commitment
NVIDIA A100 · Largest Models & Datasets
A100 80 GB
- GPUNVIDIA A100 (80 GB) · Dedicated
- vCPU12 Cores · XEON Gold
- RAM64 GB · DDR4
- Storage500 GB · NVMe SSD
- Bandwidth2 TB p/m
- IP & DDoSIncluded · Shield
Doubles GPU memory to 80 GB HBM2e for the largest models and datasets.
Minimum 12 month commitment
Introduction
Unprecedented Acceleration
at Every Scale
We utilise NVIDIA A100 Tensor Core GPUs to empower our users to be able to perform deep learning, HPC, and data analytics tasks with high-performance compute.
Representing the most powerful end-to-end AI and HPC platform for data centers, it allows technologists and researchers to deliver real-world results and deploy solutions into production at scale
With Multi-Instance GPU (MIG), the A100 can scale up efficiently or be divided into seven isolated GPU instances, offering a versatile platform that adapts dynamically to changing workload demands.
Features
Key Features of NVIDIA A100 GPUs
The A100 family shares the same Ampere architecture and third-generation Tensor Cores. The difference between the two cards is memory.
NVIDIA Ampere Architecture
From the smallest job to the biggest multi-node workload, MIG and NVLink let the A100 handle any acceleration need, maximising the utility of every GPU around the clock.
Third-Generation Tensor Cores
Up to 312 TFLOPS of deep-learning performance, 20× the Tensor throughput of NVIDIA Volta for both training and inference.
Next-Generation NVLink
2× the throughput of the previous generation; with NVSwitch, up to 16 A100 GPUs interconnect at up to 600 GB/s for peak application performance.
Multi-Instance GPU (MIG)
Partition one A100 into up to seven hardware-isolated instances, each with its own memory, cache, and cores: right-sized acceleration for every job.
High-Bandwidth Memory (HBM2e)
Up to 80 GB of HBM2e delivers the world's fastest GPU memory bandwidth, over 2 TB/s at 95% DRAM utilisation, 1.7× the previous generation.
Structural Sparsity
Tensor Cores exploit sparsity in AI models for up to 2× higher performance, most notably for inference, and also during training.
Two Cards, One Platform
Choose the Right A100 for Your Workload
Same Ampere compute, different memory. Explore the full specifications and selling points of each card on its dedicated page.
Best for value & density
A100 40 GB
The price-to-performance choice: 40 GB HBM2 with 1,555 GB/s of bandwidth for production inference, fine-tuning, and multi-tenant workloads that fit within 40 GB.
- Memory40 GB HBM2
- Bandwidth1,555 GB/s
- MIG instances7 @ 5 GB
Best for the largest workloads
A100 80 GB
The performance flagship: 80 GB HBM2e and the world's fastest GPU bandwidth (over 2 TB/s) for large-scale training, massive datasets, and memory-bound HPC.
- Memory80 GB HBM2e
- BandwidthOver 2 TB/s
- MIG instances7 @ 10 GB
40 GB vs 80 GB
Compare the Two Models Side by Side
Same Ampere compute; the difference is memory, bandwidth, and power. Here is the head-to-head at a glance.
| Specification | A100 40 GB | A100 80 GB |
|---|---|---|
| GPU memory | 40 GB HBM2 | 80 GB HBM2e |
| Memory bandwidth | 1,555 GB/s | Over 2 TB/s |
| Max TDP (PCIe) | 250 W | 300 W |
| MIG slice size | 7 @ 5 GB | 7 @ 10 GB |
| Best for | Cost-efficiency, inference & MIG multi-tenancy | Largest models, massive datasets & memory-bound HPC |
Cloud GPUaaS
Cost Effective and High Performance
GPU-as-a-Service
Our mCloud platform offers the ability to integrate powerful NVIDIA GPUs directly into your virtual machines through GPU passthrough technology, allowing virtual machines to access the full capabilities of a physical GPU and providing near-native performance
Shared GPU
Through time-sliced access to GPU compute with guaranteed minimums, clients with non-time critical workloads or limited budgets can now get affordable access to cloud-based GPU compute
Dedicated GPU
For those who require their own dedicated GPUs, our platform supports NVIDIA A100 40GB, NVIDIA A100 80GB, NVIDIA RTX A6000, NVIDIA H100, and NVIDIA H200 GPUs, all designed to support any workload you need to run in the cloud
Why Micron21
Why Choose Micron21 for GPU Compute?
As Australia's first Tier IV data centre, our GPU Compute offerings provide reliable, secure, high-calibre performance.
High-Speed Compute & NVMe Storage
Intel XEON Gold CPUs and ultra-fast NVMe SSDs deliver high-performance compute and rapid access to resources for the most demanding applications.
DDoS Protection
Every GPU Cloud Server is protected by our comprehensive DDoS platform, inspecting and filtering traffic across our global scrubbing centres.
Tier IV Data Centre
Tier IV is the highest uptime accreditation a data centre can hold, and we're Australia's first Tier IV-accredited facility.
ISO Certified
ISO 27001, 27002, 27018 and 14520 certified, PCI compliant, and IRAP assessed, up to date with the latest security standards.
Data Sovereignty
Proudly 100% Australian owned and operated. The physical sovereignty of your data is the ultimate peace of mind.
Australian Support
Our dedicated Australian-based technicians are located in the Micron21 data centre, with 24/7 access available.
Take a Virtual Tour
Welcome to our
Tier IV Data Centre
Take a look at the difference a Tier IV Data Centre can make to the reliability and availability of your services.
In terms of data centre infrastructure redundancy, Micron21 is an Australian first and leader in this space, exceeding even the highest accreditations available for the protection of power and cooling.
Ready to deploy A100 GPU compute?
Build and price your exact configuration in minutes with the calculator, or talk to our Australian-based specialists about matching the right A100 to your workload.
