NVIDIA L40SvsAMD MI250X
Detailed specifications, performance benchmarks, and pricing comparison to help you choose the right GPU for your AI workloads.
NVIDIA L40S
Ada Lovelace · 2023
AMD MI250X
CDNA 2 · 2021
Specifications · Comparison
Side by side specs.
| Spec | NVIDIA L40S | AMD MI250X |
|---|---|---|
| Architecture | Ada Lovelace | CDNA 2 |
| VRAM | 48 GB | 128 GB+62.5% |
| Memory Type | GDDR6 | HBM2e |
| Memory Bandwidth | 864 GB/s | 3277 GB/s+73.6% |
| FP32 Performance | 91.6 TFLOPS+91.2% | 47.9 TFLOPS |
| FP16 Performance | 183 TFLOPS+91.2% | 95.7 TFLOPS |
| INT8 Performance | 733 TOPS+91.4% | 383 TOPS |
| TDP | 350W | 560W |
| Form Factor | PCIe | OAM |
| Price (avg/hr) | $0.85 | $1.35 |
Performance · Analysis
Performance breakdown.
Compute (FP32)
Raw single-precision floating point throughput
NVIDIA L40S is 91.2% faster
Training (FP16)
Half-precision performance for deep learning training
NVIDIA L40S is 91.2% faster
Inference (INT8)
Integer performance for model inference workloads
NVIDIA L40S is 91.4% faster
Memory Bandwidth
Data transfer rate between memory and compute units
AMD MI250X is 73.6% faster
Best Compute
NVIDIA L40S
Most Memory
AMD MI250X
Best Training
NVIDIA L40S
Best Value
NVIDIA L40S
Pricing · Cost
Cost comparison.
Hourly
Save $0.50 with NVIDIA L40S
Daily
Save $12.00 with NVIDIA L40S
Monthly
Save $360.00 with NVIDIA L40S
Use Cases · Workloads
Best for your workload.
NVIDIA L40S
AMD MI250X
Platform · Benefits
Why Runcrate.
Instant Deployment
Get your GPU instance running in minutes with pre-configured AI environments. No setup complexity.
Pay Per Hour
Only pay for the compute you actually use. Prepaid credits with transparent, per-hour billing.
Reliable Infrastructure
Enterprise-grade reliability with automatic failover and data persistence across sessions.
Related · Comparisons
Compare other GPUs.
FAQ · Questions
Common questions.
Deploy NVIDIA L40S or AMD MI250X
Get started with GPU cloud computing in minutes. No setup complexity, no long-term commitments.