Cirrascale
USA 4 configs 4x
Previous-gen AMD data center GPU for HPC and scientific computing.
Pricing for the MI250 is consistent across 1 cloud provider, averaging around $1.50/hr per GPU. The range is tight, but the lowest current rate is $1.30/hr.
USA 4 configs 4x
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.
128GB HBM2e across dual GCDs. FP64 performance makes it strong for scientific HPC workloads. Competitive pricing as a previous-generation AMD data center GPU.
Older CDNA 2 architecture with no FP8 support. ROCm maturity for AI workloads, while improving, lags behind MI300X. Limited cloud provider availability.
The median on-demand price has held steady around $2.10/hr per GPU across providers.
With 128GB of VRAM, the MI250 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.
The MI250 has 128GB of VRAM. Multi-GPU setups increase total memory, but that memory is not automatically pooled across GPUs.
The MI250 has 3,200 GB/s of memory bandwidth. Higher bandwidth helps with faster data transfer between GPU memory and compute cores.
The MI250 supports 6 precision formats. Training: BF16, FP16, FP32. Inference: INT4, INT8. Scientific: FP64.
Yes. The MI250 supports Infinity Fabric with 400 GB/s of bidirectional bandwidth. This helps accelerate multi-GPU communication.
MI250 pricing currently ranges from $1.30/hr to $1.62/hr per GPU, depending on the provider, instance type, and billing model.
At 720 hours per month, one MI250 can cost between $936.76 to $1,169.75 per month, depending on the provider. Reserved and spot pricing can lower that further.
The MI250 is available from 1 cloud provider: Cirrascale.
Yes. We currently track 4 MI250 listings across 1 cloud providers:
| Billing type | Listings | Avg $/GPU/hr |
|---|---|---|
| Reserved | 4 | $1.48/hr |
| GPU Architecture | CDNA2 |
| Lithography | TSMC 6nm FinFET |
| Stream Processors | 13,312 |
| Compute Units | 208 |
| Peak Engine Clock | 1700 MHz |
| Peak FP16 Performance | 362.1 TFLOPs |
| Peak FP32 Matrix Performance | 90.5 TFLOPs |
| Peak FP64 Matrix Performance | 90.5 TFLOPs |
| Peak FP32 Performance | 45.3 TFLOPs |
| Peak FP64 Performance | 45.3 TFLOPs |
| Peak INT4 Performance | 362.1 TOPs |
| Peak INT8 Performance | 362.1 TOPs |
| Peak BF16 Performance | 362.1 TFLOPs |
Source: official AMD MI250 datasheet.
Last updated