L40S vs A100 SXM
Manufacturer specifications of the L40S and the A100 SXM, side by side, per GPU. A value the manufacturer does not publish for the exact model is marked as such, never estimated.
NVIDIA L40S 48GB
Ada Lovelace
- GPU memory
- 48 GB
- Memory bandwidth
- 864 GB/s
Live price on EverXCloud · from $0.79 per GPU-hour
NVIDIA A100 SXM 80GB
Ampere
- GPU memory
- 80 GB
- Memory bandwidth
- 2,039 GB/s
Live price on EverXCloud · from $1.79 per GPU-hour
Suited workloads
Key differences
- The A100 SXM has 1.7× the GPU memory of the L40S (80 GB vs 48 GB).
- The A100 SXM has 2.4× the memory bandwidth of the L40S (2,039 vs 864 GB/s), which usually sets the pace of LLM token generation.
- The L40S has 1.2× the dense BF16 tensor throughput of the A100 SXM (362 vs 312 TFLOPS), the precision of most training runs.
- The L40S has ray-tracing cores for 3D rendering; the A100 SXM does not.
Specifications
| Specification | L40S | A100 SXM |
|---|---|---|
| GPU memory | 48 GB | 80 GB (higher) |
| Memory type | GDDR6 | HBM2e |
| Memory bandwidth | 864 GB/s | 2,039 GB/s (higher) |
| FP16 tensor | 362 TFLOPS (higher) | 312 TFLOPS |
| BF16 tensor | 362 TFLOPS (higher) | 312 TFLOPS |
| FP8 tensor | 733 TFLOPS | Not published |
| Max power | 350 W | 400 W |
| Architecture | Ada Lovelace | Ampere |
| Interconnect | PCIe 4.0 | NVLink 3 (600 GB/s) · PCIe 4.0 |
The higher value of each measured row is shown in bold. Manufacturer specifications, per GPU. Tensor throughput is shown without sparsity. Values the manufacturer does not publish for this exact model are left out.
What fits in memory, on one GPU
| Workload | L40S | A100 SXM |
|---|---|---|
| LLM inference, 16-bit | ≈ 15B parameters | ≈ 30B parameters |
| LLM inference, 4-bit | ≈ 60B parameters | ≈ 100B parameters |
| Full training, mixed precision | ≈ 2B parameters | ≈ 4B parameters |
Cautious estimates counting 80% of memory for model weights; context length, batch size and framework change the real limit. Several GPUs together hold larger models.
L40S vs A100 SXM: frequently asked questions
Which has more memory, the L40S or the A100 SXM?
The A100 SXM has 80 GB of GPU memory, against 48 GB for the L40S.
Which is faster for LLM inference, the L40S or the A100 SXM?
Generating tokens is usually bound by memory bandwidth, and the A100 SXM has 2.4× the bandwidth of the L40S. Memory size then decides which models fit. Measure your own model before a large run.
Can I rent the L40S and the A100 SXM by the hour?
Yes, when sellers list them on the EverXCloud spot exchange: you buy GPU hours at the live price from a prepaid balance in euros, US dollars or pounds. Prices are shown excluding tax and EverXCloud is the merchant of record for every rental.
Be ready on launch day
Create your account now: you can top up your balance and rent the GPU that fits your workload as soon as the market opens.
Have GPU servers?
Businesses can join the EverXCloud Network and sell hours of their GPU servers on the spot exchange.