Rent MI355X
Compare live on-demand and spot rental prices across 130+ cloud providers. AMD's CDNA 4 flagship. 288GB HBM3e with 8 TB/s bandwidth — competitive with NVIDIA B200 on memory-bound workloads. Full ROCm support.
| Provider | Region | On-Demand /hr | Spot /hr | Monthly est. | Avail. |
|---|---|---|---|---|---|
Choosing the right billing model for MI355X
Provision and terminate at any time. Ideal for development, short experiments, and workloads with unpredictable duration.
Instances can be reclaimed when demand spikes. Best for fault-tolerant batch jobs, training with checkpointing, and preprocessing.
Lock in a rate for 1–3 months. Right for sustained production inference or long training runs where cost predictability matters.
Best use cases
- Large-scale LLM training
- Memory-bound inference
- AMD ROCm workloads
MI355X — Specs & Benchmarks
Performance bars, compute tiers (FP32/FP16/BF16/FP8/INT8), memory specs, LLM model size guidance, and related GPU comparisons.
MI355X Rental Guide
The MI355X is AMD's CDNA 4 flagship, built on TSMC 3nm with 288 GB of HBM3e memory and 8 TB/s of memory bandwidth. That bandwidth figure is the defining characteristic: for workloads where throughput is constrained by how fast data moves between memory and compute — large transformer inference, memory-bound fine-tuning, and multi-tenant serving of 70B+ parameter models — the MI355X sits at the top of the market alongside NVIDIA's Blackwell generation. If your bottleneck is memory bandwidth rather than raw FLOPS, this GPU is worth evaluating seriously.
Cloud availability is limited and concentrated. The MI355X is currently offered by a small number of providers, primarily those with direct AMD partnerships or early-access agreements. Check the live pricing table on this page for current listings — availability changes as AMD's supply ramp progresses. Because supply is constrained, on-demand spot pricing may be less predictable than for more widely available GPUs like the H100 or A100. Reserved or committed-use contracts, where available, are worth exploring if you need guaranteed access.
The software story requires attention before committing. AMD ROCm is the open-source compute stack that replaces CUDA on AMD hardware. PyTorch, JAX, and most major frameworks have ROCm support, but the ecosystem is less mature than CUDA: some libraries require patching, certain CUDA-specific kernels have no direct ROCm equivalent, and debugging tooling is less developed. Teams migrating from NVIDIA hardware should budget time for porting and testing. That said, AMD has invested heavily in ROCm since 2023, and major frontier labs now run production workloads on MI300X and MI355X — the gap is narrowing.
The MI355X makes the most sense for teams running memory-bound workloads at scale who are either already on ROCm, willing to invest in the migration, or using a managed inference service that abstracts the hardware layer. For training runs where NVLink-equivalent all-reduce bandwidth matters, check whether your target provider offers multi-GPU MI355X nodes with Infinity Fabric interconnect — not all cloud configurations expose this. For inference-only workloads on smaller models (under 30B parameters) that fit comfortably in less VRAM, the MI355X's cost-per-token advantage over an H100 narrows considerably, and a more available GPU may be the pragmatic choice.
Frequently Asked Questions
How much does it cost to rent a MI355X?
MI355X on-demand rental prices vary by provider and region. On-demand rates typically range based on availability and provider margins — use the comparison table above to see current live rates across all providers. Spot instances are generally 40–70% cheaper than on-demand but can be interrupted. Monthly cost estimates (hourly rate × 730 hours) are shown in the table for sustained workloads.
Which cloud provider has the cheapest MI355X?
The cheapest MI355X provider changes as providers update their pricing. The comparison table above shows live rates sorted by price, so the cheapest option is always at the top. Factors beyond headline price include region (latency to your users), availability (high/medium/low), and billing granularity (per-second vs per-hour minimums).
What can I run on a MI355X?
With 288GB of HBM3e, the MI355X can run LLM models up to approximately 144B parameters at FP16, 288B at INT8, or 576B at INT4/GGUF quantization. Common workloads include: Large-scale LLM training, Memory-bound inference, AMD ROCm workloads. AMD's CDNA 4 flagship. 288GB HBM3e with 8 TB/s bandwidth — competitive with NVIDIA B200 on memory-bound workloads. Full ROCm support.
Should I use on-demand or spot pricing for MI355X?
Spot instances save 40–70% vs on-demand but can be interrupted when the provider needs capacity back. Use spot for: batch inference jobs, training runs with checkpointing, preprocessing pipelines, and any workload that can tolerate interruption and restart. Use on-demand for: production inference serving, interactive workloads, and jobs that cannot be interrupted. Most providers bill per second, so short on-demand jobs are not penalized by hourly minimums.
How does the MI355X compare to the H100 for cloud rental?
The H100 80GB delivers 1,979 TFLOPS FP16 with 3,350 GB/s HBM3 bandwidth, compared to the MI355X's 1300 TFLOPS FP16 and 8000 GB/s bandwidth. The H100 is significantly more expensive — typically $2.50–$5.00/hr vs lower rates for the MI355X. For workloads that fit within 288GB and don't require FP8 precision, the MI355X often delivers better cost-per-token than the H100.
What is the memory bandwidth of the MI355X and why does it matter?
The MI355X has 8000 GB/s of memory bandwidth. For LLM inference, memory bandwidth is often more important than raw TFLOPS — each autoregressive token generation reads the full model weight matrix from VRAM, so bandwidth directly determines tokens-per-second throughput. Higher bandwidth means faster inference for the same model at the same batch size. For batch inference (processing many requests simultaneously), compute throughput becomes more important.
Can I use the MI355X for Stable Diffusion or image generation?
Yes — the MI355X is well-suited for Stable Diffusion and image generation workloads. Image generation is primarily FP32 and FP16 compute-bound, and the MI355X's 163.4 TFLOPS FP32 throughput determines images-per-second. The 288GB VRAM fits SDXL (requires ~6GB) and most ControlNet pipelines. For high-throughput image generation at scale, compare cost-per-image across providers using the GPU cost calculator.
Other CDNA 4 GPUs to compare
MI355X vs Alternatives — Spec & Rental Comparison
| Spec | MI355X this | MI355X 288GB |
|---|---|---|
| VRAM | 288GB HBM3e▲ | 288GB HBM3e▲ |
| Memory Bandwidth | 8000 GB/s▲ | 8000 GB/s▲ |
| FP16 TFLOPS | 1300▲ | 1300▲ |
| INT8 TOPS | 2600▲ | 2600▲ |
| TDP | 750W▲ | 750W▲ |
| Architecture | Antares | Antares |
| Process Node | TSMC 3nm▲ | TSMC 3nm▲ |
| Max model (FP16) | ~144B params▲ | ~144B params▲ |
| Max model (INT4) | ~576B params▲ | ~576B params▲ |