GMKtec EVO-X2 (128GB unified)
The $/GB winner: runs 70B-class and quantized big-MoE (V4 Flash tier) that no consumer GPU fits
L40S cloud pricing, August 2026
Live L40S quotes (48 GB VRAM) across the providers we track, cheapest first. Marketplace prices are real asks from the Vast.ai order book; secure/on-demand tiers carry datacenter SLAs (what the tiers mean).
Quotes refreshed 2026-08-23
Price floor: $0.467/hr (marketplace) at
Vast.ai.
Rent a L40S →
| Provider | Tier | $/hr per GPU | Median $/hr | Live offers | |
|---|---|---|---|---|---|
| Vast.ai | marketplace | $0.467 | $0.534 | 15 | Rent |
| Verda | spot | $0.685 | - | - | |
| RunPod | community | $0.790 | - | - | Rent |
| RunPod | secure | $0.990 | - | - | Rent |
| Verda | on-demand | $1.37 | - | - |
What fits on a single L40S
Models from our tracked open-weight set that fit in 48 GB, with KV-cache headroom included. Bigger models need multi-GPU nodes (math in the calculator).
| Model | Params | Fits at | VRAM needed | Cheapest API $/1M out |
|---|---|---|---|---|
| GLM-4.7 Flash | 31B | FP8 | 43 GB | compare → |
| Qwen3.6 27B | 27.8B | FP8 | 39 GB | compare → |
| Qwen3.5 27B | 27.8B | FP8 | 39 GB | compare → |
| Qwen3 32B | 32.8B | FP8 | 46 GB | compare → |
| Qwen3 30B A3B | 30.5B | FP8 | 42 GB | compare → |
| Mistral Small 3.1 | 24B | FP8 | 33 GB | |
| Nemotron 3 Nano | 30B | FP8 | 42 GB | compare → |
| IBM Granite 4.1 30B | 30B | FP8 | 42 GB | |
| IBM Granite 4.1 8B | 8.8B | FP8 | 13 GB | compare → |
| GPT-OSS 20B | 21B | FP8 | 29 GB | compare → |
| Llama 3.1 8B | 8B | FP8 | 11 GB | compare → |
| Llama 3.2 3B | 3.2B | FP8 | 5 GB | compare → |
| Gemma 3 27B | 27B | FP8 | 38 GB | compare → |
| Gemma 3 12B | 12B | FP8 | 17 GB | compare → |
| Phi-4 | 14.7B | FP8 | 21 GB | compare → |
| DeepSeek R1 Distill Qwen 32B | 32.8B | FP8 | 46 GB | compare → |
| Qwen3.6 35B A3B | 36B | INT4 | 25 GB | compare → |
| Qwen3.5 35B A3B | 36B | INT4 | 25 GB | compare → |
| Mixtral 8x7B | 46.7B | INT4 | 33 GB | compare → |
Prefer to own the hardware?
Renting wins for bursty loads; owning wins at sustained 24/7 use. Run your hours through the calculator. If you're buying, these are the machines this audience actually buys:
NVIDIA RTX 5090 (32GB)
Bandwidth-per-dollar king for the A3B-MoE and 27-32B dense workhorses
NVIDIA RTX PRO 6000 Blackwell (96GB)
The startup standard: only sub-$10K single card that runs 70B+ comfortably
As an Amazon Associate we earn from qualifying purchases · hardware links may be referral links · picks are editorial, not paid · disclosure
FAQ
How much does it cost to rent a L40S?
As of 2026-08-23, L40S rentals start at $0.467/hr (marketplace tier at Vast.ai). Datacenter capacity with SLAs typically costs more than community or marketplace hardware.
What can you run on a L40S?
With 48 GB of VRAM, a single L40S fits models up to roughly 34B parameters at FP8 or 68B at INT4 quantization, including KV-cache headroom.
Cite or embed today's floor
Live badge for a README. It updates itself with every daily refresh:
[](https://llmhosting.ai/gpus/l40s)
Citation: L40S rental floor $0.467/hr (llmhosting.ai, 2026-08-23). Free to reuse with a link back.