L40S vs RTX 4090
Manufacturer specifications of the L40S and the RTX 4090, 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 GeForce RTX 4090 24GB
Ada Lovelace
- GPU memory
- 24 GB
- Memory bandwidth
- 1,008 GB/s
Live price on EverXCloud · from $0.33 per GPU-hour
Suited workloads
Key differences
- The L40S has 2.0× the GPU memory of the RTX 4090 (48 GB vs 24 GB).
- The RTX 4090 has 1.2× the memory bandwidth of the L40S (1,008 vs 864 GB/s), which usually sets the pace of LLM token generation.
Specifications
| Specification | L40S | RTX 4090 |
|---|---|---|
| GPU memory | 48 GB (higher) | 24 GB |
| Memory type | GDDR6 | GDDR6X |
| Memory bandwidth | 864 GB/s | 1,008 GB/s (higher) |
| FP16 tensor | 362 TFLOPS | Not published |
| BF16 tensor | 362 TFLOPS | Not published |
| FP8 tensor | 733 TFLOPS | Not published |
| Max power | 350 W | 450 W |
| Architecture | Ada Lovelace | Ada Lovelace |
| Interconnect | PCIe 4.0 | 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 | RTX 4090 |
|---|---|---|
| LLM inference, 16-bit | ≈ 15B parameters | ≈ 9B parameters |
| LLM inference, 4-bit | ≈ 60B parameters | ≈ 30B parameters |
| Full training, mixed precision | ≈ 2B parameters | ≈ 1B 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 RTX 4090: frequently asked questions
Which has more memory, the L40S or the RTX 4090?
The L40S has 48 GB of GPU memory, against 24 GB for the RTX 4090.
Which is faster for LLM inference, the L40S or the RTX 4090?
Generating tokens is usually bound by memory bandwidth, and the RTX 4090 has 1.2× 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 RTX 4090 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.