L40S vs H100: Specs, VRAM & Price
Short answer: pick the H100 when you're training, fine-tuning, or serving at a scale that's bandwidth-bound. Pick the L40S when you're running inference at moderate scale and want a lower hourly rate with a card still built for AI workloads.
The H100 has 80 GB of VRAM; the L40S has 48 GB.
You can rent both by the hour, no purchase required. H100 from $2.78/GPU/hr; L40S from $0.844/GPU/hr. The L40S is 70% cheaper at the entry rate.
H100
Good for training and serving frontier-scale LLMs
L40S
Good for fine-tuning mid-size models and high-throughput inference
Last updated: 2026-11-03 UTCRefreshes every 6 hours
Specs side by side
| Spec | H100 | L40S |
|---|---|---|
| VRAM | 80 GB | 48 GB |
| Memory bandwidth | 3.35 TB/s | 864 GB/s |
| Architecture | Hopper | Ada Lovelace |
| Compute capability | 9.0 | 8.9 |
| Precisions in hardware | FP32, FP16, BF16, FP8, INT4 | FP32, FP16, BF16, FP8, INT4 |
| Peak FP16/BF16 tensor TFLOPS | 989 TFLOPS | 362 TFLOPS |
| Peak FP8 tensor TFLOPS | 1,979 TFLOPS | 733 TFLOPS |
| NVLink / multi-GPU interconnect | NVLink 4, 900 GB/s bidirectional | – |
| Max board power (TDP) | Up to 700W (configurable) | 350W |
| Form factor | SXM5 | PCIe |
| Regions | 7 | 4 |
On-demand price
| Rate | H100 | L40S |
|---|---|---|
| Lowest per GPU/hr | $2.78/GPU/hr | $1.07/GPU/hr |
Which is better for AI: L40S or H100?
The H100 has roughly 4x the memory bandwidth of the L40S, 3.35 TB/s of HBM3 against 864 GB/s of GDDR6 (NVIDIA product pages, Aug 2026), and is the SXM data-center accelerator, while the L40S is the more affordable data-center-AI card built for inference density rather than peak training throughput. Pick the H100 when you're training, fine-tuning, or serving at a scale that's bandwidth-bound; pick the L40S when you're running inference at moderate scale and want a lower hourly rate with a card still built for AI workloads. On price the L40S undercuts the H100 by 70% at the entry rate ($0.844 vs $2.78/GPU/hr). Worth weighing if the deciding factor above isn't a hard requirement for your job.
Decision criteria
| Memory bandwidth | H100 SXM: 3.35 TB/s HBM3. L40S: 864 GB/s GDDR6 (NVIDIA product pages, Aug 2026). |
| Positioning | H100 is the flagship SXM training/inference accelerator; L40S is a lower-cost, inference-optimized data-center card. |
| VRAM | 80 GB (H100) vs 48 GB (L40S), live from current listings. |
| Availability | 7 region(s) list the H100 vs 4 for the L40S. |
| Entry price | $2.78/GPU/hr (H100) vs $0.844/GPU/hr (L40S). |
Pick the H100 when
- You're training or fine-tuning at scale
- You're bandwidth-bound on large-batch serving
- You need the flagship data-center SXM accelerator
Pick the L40S when
- You're running inference at moderate scale
- You want a lower hourly rate on a card still built for AI
- L40S has better availability where you're deploying
Price-performance
At list rates, 1,000 GPU-hours costs $844 on the L40S against $2,778 on the H100. Neither NVIDIA nor Aquanode publishes a workload-normalized $/token or $/epoch figure for this pair, so $/GPU-hour below is the only apples-to-apples number. A faster card can still cost less per finished job even at a higher hourly rate.
L40S vs H100: common questions
Is the H100 or the L40S better for AI?
Short answer: pick the H100 when you're training, fine-tuning, or serving at a scale that's bandwidth-bound. Pick the L40S when you're running inference at moderate scale and want a lower hourly rate with a card still built for AI workloads.
Is the H100 or the L40S cheaper to rent?
On Aquanode's live marketplace the H100 starts at $2.78/GPU/hr (median $3.70/GPU/hr) and the L40S starts at $0.844/GPU/hr (median $2.36/GPU/hr). The L40S is the cheaper of the two at the entry rate, by 70%.
What is the difference between the H100 and the L40S?
The H100 has 80 GB of VRAM against the L40S's 48 GB; the H100 is a Hopper part and the L40S is Ada Lovelace (compute capability 9.0 vs 8.9); both run BF16, FP8, INT4 workloads in hardware. On price, the H100 lists from $2.78/GPU/hr and the L40S from $0.844/GPU/hr.
How widely available are the H100 and the L40S?
The H100 is listed in 7 regions and the L40S in 4 regions right now.
How much does 1,000 GPU-hours cost on the H100 vs the L40S?
At the lowest rates listed today, 1,000 GPU-hours costs $2,778 on the H100 and $844 on the L40S, a difference of $1,934 for the same runtime. Rates are per GPU per hour and update hourly.
How much VRAM does the H100 have compared to the L40S?
The H100 has 80 GB of VRAM; the L40S has 48 GB.
Is the L40S a good budget alternative to the H100 for inference?
For moderate-scale inference, often yes. It has FP8 support like the H100, just at 864 GB/s of bandwidth against the H100's 3.35 TB/s (NVIDIA, Aug 2026). Today the L40S lists from $0.844/GPU/hr against the H100's $2.78/GPU/hr on Aquanode, so it's worth testing your actual throughput before committing at scale.
Does the H100 or the L40S support NVLink?
The H100 supports NVLink 4, 900 GB/s bidirectional; the L40S has no dedicated GPU-to-GPU interconnect.
How this comparison is calculated
Every price is normalized to a per-GPU hourly rate using the same pipeline as every other pricing surface on Aquanode, and only the cheapest qualifying offer per provider is shown. A dash means that provider does not currently list that GPU.
Architecture, compute capability and supported precisions come from each vendor's published datasheet for that generation, not from the marketplace feed. This page regenerates at most once per hour.
Memory bandwidth, tensor TFLOPS, NVLink/interconnect and max board power are the same verified-from-datasheet figures shown on each model's own H100 page and L40S page, which link the primary vendor source for that model. A dash here means Aquanode hasn't verified that figure from a primary source yet, never a guess.
Go further
Check what fits in each card's memory with the H100 VRAM calculator and the L40S VRAM calculator, see vendor-published throughput in the GPU benchmarks and specs table, or browse every model by generation in the GPU index.
More H100 comparisons
More L40S comparisons
Related reading: H100 pricing and specs, H100 vs H200, A100 vs H100, H100 and H200: SXM vs NVL vs PCIe, and The best GPUs for AI, ranked.
