Specifications Compared
| Spec | L40 | MI250X |
|---|---|---|
| TDP | 300W | 560W |
| VRAM | 48 GB | 128 GB |
| CUDA Cores | 18,176 | Not published |
| FP8 (dense) | 362 TFLOPS | Not published |
| Memory Type | GDDR6 | HBM2e |
| Architecture | Ada Lovelace | CDNA 2 |
| FP16 (dense) | 181.05 TFLOPS | 383 TFLOPS |
| Form Factors | PCIe | OAM |
| INT8 (dense) | 362 TOPS | 383 TOPS |
| Interconnect | PCIe 4.0 | Infinity Fabric, PCIe 4.0 |
| Tensor Cores | 568 | Not published |
| FP32 Performance | 90.5 TFLOPS | 47.9 TFLOPS |
| FP64 Performance | Not published | 47.9 TFLOPS |
| Memory Bandwidth | 864 GB/s | 3,277 GB/s |
| FP8 (with sparsity) | 724 TFLOPS | Not published |
| FP16 (with sparsity) | 362.1 TFLOPS | Not published |
| INT8 (with sparsity) | 724 TOPS | Not published |
Performance Analysis
The dense FP16 rating of 181.05 TFLOPS on the L40 versus 383 TFLOPS on the MI250X indicates the MI250X completes dense FP16 matrix operations more than twice as fast. This gap affects training workloads that rely on dense FP16 precision and inference tasks that stay within dense FP16 limits. The L40 dense FP16 figure of 181.05 TFLOPS remains lower than the MI250X dense FP16 figure of 383 TFLOPS so batch sizes scale with the higher memory bandwidth of 3277 GB per second on the MI250X compared with 864 GB per second on the L40. FP32 workloads favor the L40 at 90.5 TFLOPS over the MI250X at 47.9 TFLOPS because the higher FP32 throughput supports precision sensitive scientific kernels without sparsity.
Current On-Demand Offers
Cheapest secure on-demand offer per provider that is reported in stock right now, per GPU per hour; multi-GPU instances show the whole-instance price alongside. Deploy opens the DeployGPU console for providers on the platform, otherwise the provider's own site. Stock is rechecked every minute.
L40
| Provider | Region | GPUs | Per GPU / hr | Instance / hr | Deploy |
|---|---|---|---|---|---|
| ThunderCompute | USA | 1 | $0.79 | — | Deploy |
| Massed Compute | us-central-3 | 1 | $0.86 | — | Deploy |
| QuantaCloud | us-midwest-2 | 4 | $0.94 | $3.75 | Deploy |
| Hyperstack | CANADA-1 | 1 | $1.00 | — | Deploy |
4 providers in stock, 12 offers (cheapest per provider shown). All L40 offers, price history and alerts
MI250X
MI250X is not offered on-demand by any provider we track right now. See the MI250X rental page for last-seen listed prices and a price alert.
Notify me when L40 drops below a price
One email when the cheapest in-stock on-demand price per GPU-hour falls below your threshold. Cheapest right now: $0.79/GPU-hr.
QuantaCloud
Comparing providers? We broker across all of them.
Stop tab-switching between pricing pages. Tell us what you need, 16+ GPUs reserved or cluster capacity, and we return one quote at partner rates within 24 hours.
When to Choose the L40
The L40 suits environments that require 300 W TDP and PCIe form factor deployment. Its 90.5 TFLOPS FP32 rating exceeds the MI250X 47.9 TFLOPS FP32 rating so FP32 centric workloads gain throughput on the L40.
When to Choose the MI250X
The MI250X suits workloads that need 128 GB memory and 3277 GB per second bandwidth. Its dense FP16 rating of 383 TFLOPS surpasses the L40 dense FP16 rating of 181.05 TFLOPS so large model training and inference benefit from the higher dense FP16 throughput.
Use Cases
The MI250X supplies 128 GB memory and 383 TFLOPS dense FP16 that exceed the L40 48 GB memory and 181.05 TFLOPS dense FP16.
The MI250X 3277 GB per second bandwidth and 383 TFLOPS dense FP16 allow larger concurrent inference batches than the L40 864 GB per second bandwidth and 181.05 TFLOPS dense FP16.
The MI250X 128 GB memory supports fine tuning of models that exceed the L40 48 GB memory capacity.
The L40 90.5 TFLOPS FP32 rating exceeds the MI250X 47.9 TFLOPS FP32 rating for diffusion workloads that stay in FP32.
The L40 90.5 TFLOPS FP32 rating and 300 W TDP suit scientific kernels that require higher FP32 throughput at lower power.
Frequently Asked Questions
What is the memory difference between L40 and MI250X?▾
The L40 contains 48 GB GDDR6 memory while the MI250X contains 128 GB HBM2e memory.
How do the dense FP16 ratings compare?▾
The L40 delivers 181.05 TFLOPS dense FP16 while the MI250X delivers 383 TFLOPS dense FP16.
Which GPU has higher FP32 performance?▾
The L40 provides 90.5 TFLOPS FP32 while the MI250X provides 47.9 TFLOPS FP32.
What are the TDP values for each GPU?▾
The L40 TDP is 300 W and the MI250X TDP is 560 W.
Which interconnect options exist on the MI250X?▾
The MI250X supports Infinity Fabric and PCIe 4.0 interconnects.
Does the L40 list an FP8 rating?▾
The L40 lists FP8 dense performance at 362 TFLOPS.
Which is cheaper to rent, the L40 or the MI250X?▾
Cloud rental prices for both the L40 and MI250X vary by provider, configuration, and availability. This page shows live pricing from 25+ providers updated every 60 seconds. Scroll to the Live Cloud Pricing section to compare current rates.
How much VRAM does the L40 have compared to the MI250X?▾
The L40 has 48 GB of GDDR6 memory. The MI250X has 128 GB of HBM2e memory.
Can I find L40 and MI250X GPUs available to rent right now?▾
Yes. This page shows real-time availability across 25+ cloud GPU providers. The Live Cloud Pricing section displays only in-stock offers with current pricing.
What is the main difference between the L40 and the MI250X?▾
The L40 uses the Ada Lovelace architecture (2023) while the MI250X uses CDNA 2 (2021). The MI250X delivers 2.1x the dense FP16 throughput (383 vs 181.05 TFLOPS, both without sparsity) and 3.8x the memory bandwidth of the L40.
Rent these GPUs
Each GPU page lists every current on-demand offer by provider, per GPU-hour, updated every minute.
Related comparisons
How this page is made
- Specifications come from the NVIDIA, AMD and Intel datasheets for the L40 and the MI250X. Dense and sparse throughput are listed separately.
- Prices and availability are live: every offer shown is an in-stock, on-demand listing from a provider we track, rechecked every minute.
- The written comparison (overview, performance notes, when to choose each card, verdict, use cases and FAQ) was drafted with an AI model from the specification table above and passed an automated check that rejects any figure not in that table. It was last generated on . It contains no prices; those are always read live.
- Read how we collect the data, or report an error on this page.
Next steps
- Rent L40Every current offer by provider, daily price history and a price alert.
- Rent MI250XEvery current offer by provider, daily price history and a price alert.
- GPU price indexHow on-demand prices for the major GPUs have moved, updated daily.
- All GPUsEvery GPU we track, with current lows and provider counts.