COMPARE / H100 SXM VS L40S
H100 SXM vs L40S
Specs from the chip makers' datasheets; prices from the live Compute Cafe panel, refreshed hourly. Every rate is an on-demand published price with a source receipt.
CHEAPEST H100 SXM
$0.890/hr
SpotGPUs · median $5.49 · 77 providers
CHEAPEST L40S
$0.250/hr
SpotGPUs · median $1.39 · 48 providers
CHEAPER RIGHT NOW
L40S
live panel, 2026-09-27
| H100 SXM | L40S | |
|---|---|---|
| MAKER | NVIDIA | NVIDIA |
| ARCHITECTURE | Hopper | Ada Lovelace |
| LAUNCHED | 2022 | 2023 |
| MEMORY | 80 GB HBM3 | 48 GB GDDR6 ECC |
| MEMORY BANDWIDTH | 3.35 TB/s | 864 GB/s |
| TENSOR COMPUTE | 989 TFLOPS FP16/BF16 tensor dense (1,979 sparse); 1,979 TFLOPS FP8 dense | 362 TFLOPS FP16/BF16 tensor dense (733 sparse); 733 TFLOPS FP8 dense |
| POWER DRAW | 700 W | 350 W |
| INTERCONNECT | NVLink 4.0 900 GB/s | PCIe 4.0 x16 (no NVLink) |
HOW TO CHOOSE
H100 SXM vs L40S
On paper - H100 SXM (Hopper): 80 GB HBM3, 3.35 TB/s bandwidth, 700 W. L40S (Ada Lovelace): 48 GB GDDR6 ECC, 864 GB/s bandwidth, 350 W. The budget-allocation question: one H100-hour versus several L40S-hours - training throughput against inference economics at the same spend.
L40S vs H100 SXM
On paper - H100 SXM (Hopper): 80 GB HBM3, 3.35 TB/s bandwidth, 700 W. L40S (Ada Lovelace): 48 GB GDDR6 ECC, 864 GB/s bandwidth, 350 W. The budget-allocation question: one H100-hour versus several L40S-hours - training throughput against inference economics at the same spend.
Workload fit
H100 SXM: llm pre-training, 70b+ fine-tuning, high-throughput inference. L40S: llm inference up to ~34b, fine-tuning 7b-13b, image and video generation.
H100 SXM VS L40S - FAQ
Which is cheaper to rent, H100 SXM or L40S?
By today's published on-demand rates: H100 SXM starts at $0.890/GPU/hr (median $5.49, 77 providers) and L40S starts at $0.250/GPU/hr (median $1.39, 48 providers). The cheapest live listing is the L40S at SpotGPUs. Figures recompute from the panel on every refresh.
H100 SXM or L40S - which should I rent?
H100 SXM: The H100 SXM is the chip the current generation of frontier models was trained on, and the benchmark every other datacenter GPU is priced against. L40S: The L40S is the workhorse of the inference-and-fine-tuning middle: Ada tensor cores with FP8, 48 GB of memory, and a price well below the H100 line. The side-by-side spec table and live price rows above give the full picture.
Where can I rent H100 SXM and L40S?
77 providers publish H100 SXM rates and 48 publish L40S rates on Compute Cafe. Every row links to the provider's own published price - committed-use and reserved rates are excluded.
Full pricing: H100 SXM · L40S · all comparisons