AMD MI455X
AMD's CDNA 5 flagship for frontier-scale training and inference in Helios rack systems.
Compare vs other GPUs →Key Specifications
Compare Cloud Provider Prices
The MI455X is listed by 4 cloud providers, but none offer public on-demand pricing.
How we sort results
Default ranking
Our algorithm weighs five factors to find the most relevant matches for you:
- Price: We blend total hourly price with price-per-GPU to balance affordability and value.
- Specs: We favor offers with higher CPU, RAM, and GPU Memory.
- Billing: We favor on-demand billing for simplicity and flexibility over spot instances, reservations, and custom quotes.
- Location: We blend both datacenter proximity and provider HQ location. Datacenter location matters for latency, while HQ location matters for compliance and support.
- Provider diversity: Each time a provider appears in the list, their subsequent offerings are ranked a little lower, so one provider's offerings don't crowd out the top positions.
Sorting and filtering
Click any column header to sort by that column. Use the filters above the table to narrow results by billing type, GPU count, vCPUs, or RAM. Custom sorting resets the default relevance ranking.
Transparency and funding
Ads and sponsors: Any paid placements are fixed at the top and clearly labeled as sponsored content.
Affiliates: Any affiliate links will be indicated to you as well. We may earn a commission if you click them, but this never influences the ranking order.
| Provider | GPUs | Total VRAM | vCPUs | RAM | Billing | $/GPU/h | Total/h | Availability | |
|---|---|---|---|---|---|---|---|---|---|
TensorWave
|
1x MI455X 1x MI455X 432GB | 432GB | -- | -- | Custom Accepting reservations, pricing TBD | Custom | Custom | Waitlist | View |
Cirrascale
|
1x MI455X 1x MI455X 432GB | 432GB | -- | -- | Custom Announced for the AI Innovation Cloud, pricing TBD | Custom | Custom | Waitlist Last checked <24h ago | View |
Azure
|
1x MI455X 1x MI455X 432GB | 432GB | -- | -- | Custom Custom terms negotiated with the provider. | Custom | Custom | Waitlist Last checked <1h ago | View |
Vultr
|
1x MI455X 1x MI455X 432GB | 432GB | -- | -- | Custom Custom terms negotiated with the provider. | Custom | Custom | Waitlist Last checked <15m ago | View |
| No offerings matching your filters. | |||||||||
Heads up: We do our best to keep these specs & prices accurate. However, cloud costs may fluctuate based on region, usage, and other factors not listed here. These are estimates based on common setups and are for informational purposes only. Always verify current rates & exact specs with the provider before provisioning.
Frequently Asked Questions
Why choose the MI455X?
432 GB of HBM4 per GPU, the largest memory capacity announced for a data center accelerator, with 23.3 TB/s of bandwidth. Designed for the 72-GPU Helios rack architecture with UALink scale-up networking, targeting frontier model training and large-scale inference.
When is the MI455X not a good fit?
Shipments are ramping from late 2026 into the first half of 2027, so cloud availability is minimal and pricing is not yet published. For AMD workloads today, the MI355X is in production clouds now. ROCm, while rapidly improving, is narrower than CUDA, so evaluate software compatibility before committing.
What size AI models can the MI455X run?
With 432GB of VRAM, the MI455X can usually run 70B-class models with headroom, and may handle much larger models in 4-bit quantized form depending on runtime overhead, KV cache, and context length.
How much VRAM does the MI455X have?
The MI455X has 432GB of VRAM. Multi-GPU setups increase total memory, but that memory is not automatically pooled across GPUs.
What is the MI455X's memory bandwidth?
The MI455X has 23,300 GB/s of memory bandwidth. Higher bandwidth helps with faster data transfer between GPU memory and compute cores.
What data types does the MI455X support?
The MI455X supports 6 precision formats. Training: BF16, FP16, FP32. Inference: FP4, FP6, FP8.
Does the MI455X support Infinity Fabric?
Yes. The MI455X supports Infinity Fabric with 3600 GB/s of bidirectional bandwidth. This helps accelerate multi-GPU communication.
Which cloud providers offer the MI455X?
The MI455X is available from 4 cloud providers, including Cirrascale, TensorWave, Vultr. Pricing and availability vary by region and billing model.
Can I rent the MI455X in the cloud?
Yes. We currently track 4 MI455X listings across 4 cloud providers:
Billing type Listings Avg $/GPU/hr Custom contract 4 Custom
Technical Specifications
| MXFP4 Performance | 40.26 PFLOPS |
| MXFP6 Performance | 20.13 PFLOPS |
| MXFP8 Performance | 20.13 PFLOPS |
| FP16/BF16 Matrix Performance | 5.03 PFLOPS |
| FP32 Vector Performance | 315 TFLOPS |
| Memory Capacity | 432 GB HBM4 (12 stacks) |
| Memory Bandwidth | 23.3 TB/s |
| Scale-up Bandwidth | 3.6 TB/s bidirectional (UALink over Ethernet) |
| Host Interface | 256 GB/s bidirectional Infinity Fabric |
| Compute Units | 256 WGPs across 8 XCDs |
| Peak Engine Clock | 2.4 GHz |
| Transistor Count | 320 billion |
| Lithography | TSMC N2 (compute) / N3 (I/O), CoWoS-L packaging |
| Cooling | Direct liquid cooling |
| Rack Configuration (Helios) | 72 GPUs, 31 TB HBM4, 2.9 EFLOPS MXFP4 |
Source: official AMD MI455X datasheet.
Alternatives to AMD MI455X
AMD MI355X
AMD's current CDNA 4 flagship with 288GB HBM3E, available in production clouds. From $8.60/hr per GPU across 4 providers.
Nvidia B300
Nvidia's Blackwell Ultra accelerator with 288GB HBM3E and CUDA support. From $0.98/hr per GPU across 17 providers.
Nvidia GB300
Nvidia's rack-scale system pairing 72 Blackwell Ultra GPUs with Grace CPUs. From $4.26/hr per GPU across 5 providers.
Last updated