L40S vs RTX PRO 6000 Blackwell
The L40S rents for 28% less per hour today — $1.47 against $2.05 — with 48 GB against 96 GB of memory.
Prices collected 2026-09-18 · 17 providers price both
Index median per GPU-hour, on-demand, per snapshot.
Side by side
| L40S | RTX PRO 6000 Blackwell | |
|---|---|---|
| Median rental, per GPU-hour | $1.47 | $2.05 |
| Cheapest listing | $0.69 | $0.80 |
| Per month at 720 h | $1,058 | $1,472 |
| Providers pricing it | 30 | 24 |
| Memory | 48 GB GDDR6 | 96 GB GDDR7 |
| Memory bandwidth | 864 GB/s | 1,792 GB/s |
| GB of memory per $/hr | 32.7 | 46.9 |
| Architecture | Ada Lovelace | Blackwell |
| Class | Datacenter | Workstation |
| Released | 2023-08 | 2025-04 |
Same provider, both GPUs
The cleanest comparison: 17 providers price both cards, so the difference is the GPU and nothing else. The L40S is cheaper at 17 of them.
| Provider | L40S | RTX PRO 6000 Blackwell | RTX PRO 6000 Blackwell premium |
|---|---|---|---|
 Beam | $0.76 | $1.09 | +43% |
 RunPod Community Cloud | $0.79 | $1.69 | +114% |
 GPU.ai | $0.80 | $1.84 | +131% |
 Vast.ai | $0.80 | $1.47 | +84% |
 io.net | $0.89 | $2.24 | +152% |
| $0.96 | $2.27 | +136% | |
| $0.97 | $2.19 | +126% | |
 Runcrate | $0.97 | $2.41 | +148% |
 Seeweb | $0.98 | $1.44 | +47% |
 RunPod Secure Cloud | $1.09 | $2.14 | +96% |
| $1.20 | $1.90 | +58% | |
 DataCrunch (Verda) | $1.53 | $1.90 | +25% |
 Nebius | $1.69 | $1.80 | +7% |
 Modal | $1.95 | $3.03 | +55% |
 Cerebrium | $1.95 | $2.50 | +28% |
 CoreWeave | $2.25 | $2.50 | +11% |
 Oracle Cloud | $3.50 | $4.50 | +29% |
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 | L40S | RTX PRO 6000 Blackwell |
|---|---|---|---|
| Kimi K3 | 4-bit | 37× GPU$54.39/hr | 19× GPU$38.85/hr |
| Kimi K3 | 16-bit | 130× GPU$191.10/hr | 65× GPU$132.92/hr |
| DeepSeek V4.1-Flash | 4-bit | 8× GPU$11.76/hr | 4× GPU$8.18/hr |
| DeepSeek V4.1-Flash | 16-bit | 26× GPU$38.22/hr | 13× GPU$26.59/hr |
| Llama 4 Maverick | 4-bit | 6× GPU$8.82/hr | 3× GPU$6.14/hr |
| Llama 4 Maverick | 16-bit | 19× GPU$27.93/hr | 10× GPU$20.45/hr |
| gpt-oss-120b | 4-bit | 2× GPU$2.94/hr | ✓ 67.9 GB~500 tok/s · $1.14/M |
| gpt-oss-120b | 16-bit | 6× GPU$8.82/hr | 3× GPU$6.14/hr |
| Llama 3.1 70B | 4-bit | 2× GPU$2.94/hr | ✓ 44.5 GB~36 tok/s · $15.78/M |
| Llama 3.1 70B | 16-bit | 4× GPU$5.88/hr | 2× GPU$4.09/hr |
| Gemma 4 31B | 4-bit | ✓ 21.2 GB~40 tok/s · $10.21/M | ✓ 21.2 GB~82 tok/s · $6.93/M |
| Gemma 4 31B | 16-bit | 2× GPU$2.94/hr | ✓ 65.7 GB~23 tok/s · $24.7/M |
| Qwen3 Coder 30B-A3B | 4-bit | ✓ 19 GB~372 tok/s · $1.1/M | ✓ 19 GB~772 tok/s · $0.74/M |
| Qwen3 Coder 30B-A3B | 16-bit | 2× GPU$2.94/hr | ✓ 62.9 GB~217 tok/s · $2.62/M |
| Qwen3.8 27B | 4-bit | ✓ 18.7 GB~46 tok/s · $8.88/M | ✓ 18.7 GB~94 tok/s · $6.04/M |
| Qwen3.8 27B | 16-bit | 2× GPU$2.94/hr | ✓ 57.6 GB~27 tok/s · $21.04/M |
| gpt-oss-20b | 4-bit | ✓ 13.7 GB~341 tok/s · $1.2/M | ✓ 13.7 GB~708 tok/s · $0.8/M |
| gpt-oss-20b | 16-bit | 2× GPU$2.94/hr | ✓ 43.9 GB~199 tok/s · $2.85/M |
| Llama 3.1 8B | 4-bit | ✓ 6.9 GB~154 tok/s · $2.65/M | ✓ 6.9 GB~319 tok/s · $1.78/M |
| Llama 3.1 8B | 16-bit | ✓ 18.4 GB~43 tok/s · $9.5/M | ✓ 18.4 GB~90 tok/s · $6.31/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 L40S or the RTX PRO 6000 Blackwell cheaper to rent?
On 2026-09-18 the L40S has a market median of $1.47 per GPU-hour across 30 providers and the RTX PRO 6000 Blackwell $2.04 across 24, so the L40S is 28% cheaper per hour. Among the 17 providers that price both, the L40S is cheaper at 17 and the RTX PRO 6000 Blackwell at 0. 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 L40S that is 2 cards at $2.94/hr; on the RTX PRO 6000 Blackwell one card at $2.05/hr.
What is the real difference between them?
48 GB GDDR6 at 864 GB/s (Ada Lovelace) against 96 GB GDDR7 at 1,792 GB/s (Blackwell). 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.