A100 80GB vs L40S
The L40S rents for 18% less per hour today — $1.47 against $1.79 — with 48 GB against 80 GB of memory.
Prices collected 2026-09-18 · 26 providers price both
Index median per GPU-hour, on-demand, per snapshot.
Side by side
| A100 80GB | L40S | |
|---|---|---|
| Median rental, per GPU-hour | $1.79 | $1.47 |
| Cheapest listing | $0.43 | $0.69 |
| Per month at 720 h | $1,289 | $1,058 |
| Providers pricing it | 37 | 30 |
| Memory | 80 GB HBM2e | 48 GB GDDR6 |
| Memory bandwidth | 2,039 GB/s | 864 GB/s |
| GB of memory per $/hr | 44.7 | 32.7 |
| Architecture | Ampere | Ada Lovelace |
| Class | Datacenter | Datacenter |
| Released | 2020-11 | 2023-08 |
Same provider, both GPUs
The cleanest comparison: 26 providers price both cards, so the difference is the GPU and nothing else. The A100 80GB is cheaper at 0 of them.
| Provider | A100 80GB | L40S | L40S premium |
|---|---|---|---|
 Vast.ai | $0.80 | $0.80 | 0% |
 Seeweb | $1.14 | $0.98 | -14% |
 GPU.ai | $1.21 | $0.80 | -34% |
 RunPod Community Cloud | $1.29 | $0.79 | -39% |
 io.net | $1.31 | $0.89 | -32% |
 Beam | $1.36 | $0.76 | -44% |
| $1.37 | $0.97 | -29% | |
| $1.43 | $0.96 | -33% | |
 Packet.ai | $1.43 | $0.92 | -36% |
 Runcrate | $1.49 | $0.97 | -35% |
 RunPod Secure Cloud | $1.59 | $1.09 | -31% |
 DataCrunch (Verda) | $1.74 | $1.53 | -12% |
 Civo | $1.79 | $1.29 | -28% |
 Koyeb | $1.88 | $1.20 | -36% |
| $2.04 | $1.20 | -41% | |
 Lyceum | $2.05 | $1.44 | -30% |
 Cerebrium | $2.10 | $1.95 | -7% |
 Crusoe | $2.15 | $1.50 | -30% |
 Vultr | $2.40 | $2.30 | -4% |
 Modal | $2.50 | $1.95 | -22% |
 CoreWeave | $2.70 | $2.25 | -17% |
 Lightning AI | $2.71 | $2.14 | -21% |
 OVHcloud | $3.07 | $1.80 | -41% |
 Amazon Web Services | $3.43 | $3.39 | -1% |
 Oracle Cloud | $4.00 | $3.50 | -12% |
 Replicate | $5.04 | $3.51 | -30% |
What each one runs, and what a million tokens costs
Weights + KV cache at 8K context + overhead against 90% of each card's memory. Where a model needs more than one card the cell says how many and prices the set at today's median. $/M tokens is the hourly median divided by estimated single-stream output, from memory bandwidth — an estimate for ranking, not a benchmark.
| Model | Bits | A100 80GB | L40S |
|---|---|---|---|
| Kimi K3 | 4-bit | 22× GPU$39.38/hr | 37× GPU$54.39/hr |
| Kimi K3 | 16-bit | 78× GPU$139.62/hr | 130× GPU$191.10/hr |
| DeepSeek V4.1-Flash | 4-bit | 5× GPU$8.95/hr | 8× GPU$11.76/hr |
| DeepSeek V4.1-Flash | 16-bit | 16× GPU$28.64/hr | 26× GPU$38.22/hr |
| Llama 4 Maverick | 4-bit | 4× GPU$7.16/hr | 6× GPU$8.82/hr |
| Llama 4 Maverick | 16-bit | 12× GPU$21.48/hr | 19× GPU$27.93/hr |
| gpt-oss-120b | 4-bit | ✓ 67.9 GB~569 tok/s · $0.87/M | 2× GPU$2.94/hr |
| gpt-oss-120b | 16-bit | 4× GPU$7.16/hr | 6× GPU$8.82/hr |
| Llama 3.1 70B | 4-bit | ✓ 44.5 GB~41 tok/s · $12.13/M | 2× GPU$2.94/hr |
| Llama 3.1 70B | 16-bit | 3× GPU$5.37/hr | 4× GPU$5.88/hr |
| Gemma 4 31B | 4-bit | ✓ 21.2 GB~94 tok/s · $5.29/M | ✓ 21.2 GB~40 tok/s · $10.21/M |
| Gemma 4 31B | 16-bit | ✓ 65.7 GB~26 tok/s · $19.12/M | 2× GPU$2.94/hr |
| Qwen3 Coder 30B-A3B | 4-bit | ✓ 19 GB~879 tok/s · $0.57/M | ✓ 19 GB~372 tok/s · $1.1/M |
| Qwen3 Coder 30B-A3B | 16-bit | ✓ 62.9 GB~247 tok/s · $2.01/M | 2× GPU$2.94/hr |
| Qwen3.8 27B | 4-bit | ✓ 18.7 GB~107 tok/s · $4.65/M | ✓ 18.7 GB~46 tok/s · $8.88/M |
| Qwen3.8 27B | 16-bit | ✓ 57.6 GB~30 tok/s · $16.57/M | 2× GPU$2.94/hr |
| gpt-oss-20b | 4-bit | ✓ 13.7 GB~806 tok/s · $0.62/M | ✓ 13.7 GB~341 tok/s · $1.2/M |
| gpt-oss-20b | 16-bit | ✓ 43.9 GB~227 tok/s · $2.19/M | 2× GPU$2.94/hr |
| Llama 3.1 8B | 4-bit | ✓ 6.9 GB~362 tok/s · $1.37/M | ✓ 6.9 GB~154 tok/s · $2.65/M |
| Llama 3.1 8B | 16-bit | ✓ 18.4 GB~102 tok/s · $4.87/M | ✓ 18.4 GB~43 tok/s · $9.5/M |
Buy or rent?
Hours of renting at today's median that add up to the card's approximate list price. List prices are typical quoted figures, not tracked, and do not include the server, power or the person to run it — so the break-even is a floor, not a verdict.
Questions
Is the A100 80GB or the L40S cheaper to rent?
On 2026-09-18 the A100 80GB has a market median of $1.79 per GPU-hour across 37 providers and the L40S $1.47 across 30, so the L40S is 18% cheaper per hour. Among the 26 providers that price both, the A100 80GB is cheaper at 0 and the L40S at 26. Per hour is only half the answer — the table below prices a whole model on each.
Which should I rent for a 70B model?
At 4-bit, Llama 3.1 70B needs about 44.5 GB. On the A100 80GB that is one card at $1.79/hr; on the L40S 2 cards at $2.94/hr.
What is the real difference between them?
80 GB HBM2e at 2,039 GB/s (Ampere) against 48 GB GDDR6 at 864 GB/s (Ada Lovelace). Memory decides what fits; bandwidth decides how fast it generates, because each token streams the active weights once. Compute matters for prompt processing and training, not for how fast text appears.