> For the complete documentation index, see [llms.txt](https://docs.clore.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.clore.ai/guides/getting-started/pricing.md).

# GPU Pricing & Availability

What is actually listed on the Clore.ai marketplace right now, and what it costs per GPU hour

{% hint style="info" %}
**Snapshot: 24 August 2026.** Numbers below come from the live marketplace listing: **2,202 servers / 6,791 GPUs**, of which **1,198 servers were free to rent** at the time of the snapshot. Prices are **per GPU per hour**, on-demand, in USD. The marketplace itself is the source of truth — check [clore.ai/marketplace](https://clore.ai/marketplace) before you budget.
{% endhint %}

Clore.ai bills per minute. A price shown here as `$0.20/hr` means you pay roughly `$0.0033` per minute of uptime and nothing when the order is closed. The marketplace UI can display either **per hour** or **per day** — this page, and every guide in this collection, quotes **per GPU per hour**.

## What it costs

Ranges are the 10th to 90th percentile of servers that were free to rent in the snapshot, so the low end is what a patient renter pays and the high end is a premium listing.

| GPU                    | VRAM     | $/GPU/hr       | Median    | Free servers |
| ---------------------- | -------- | -------------- | --------- | ------------ |
| RTX 3060               | 12GB     | $0.03–0.07     | $0.04     | 32           |
| RTX 3060 Ti            | 8GB      | $0.03–0.10     | $0.05     | 15           |
| RTX 3070               | 8GB      | $0.03–0.09     | $0.05     | 54           |
| RTX 3080               | 10GB     | $0.05–0.19     | $0.08     | 68           |
| RTX 3080 Ti            | 12GB     | $0.06–0.19     | $0.10     | 48           |
| **RTX 3090**           | **24GB** | **$0.07–0.21** | **$0.09** | **90**       |
| RTX 4060 Ti            | 8GB      | $0.04–0.12     | $0.06     | 16           |
| RTX 4070               | 12GB     | $0.04–0.20     | $0.07     | 24           |
| RTX 4070 Ti            | 12GB     | $0.07–0.25     | $0.12     | 24           |
| RTX 4070 Ti SUPER      | 16GB     | $0.08–0.19     | $0.12     | 28           |
| RTX 4080 SUPER         | 16GB     | $0.09–0.17     | $0.12     | 22           |
| **RTX 4090**           | **24GB** | **$0.14–0.42** | **$0.20** | **98**       |
| RTX 5060 Ti            | 16GB     | $0.04–0.10     | $0.06     | 29           |
| RTX 5070               | 12GB     | $0.07–0.29     | $0.11     | 40           |
| RTX 5070 Ti            | 16GB     | $0.10–0.25     | $0.13     | 103          |
| RTX 5080               | 16GB     | $0.12–0.62     | $0.17     | 56           |
| **RTX 5090**           | **32GB** | **$0.25–0.77** | **$0.29** | **121**      |
| RTX PRO 6000 Blackwell | 96GB     | $0.92–1.38     | $1.21     | 4            |
| Tesla V100 SXM2        | 32GB     | $0.05–0.06     | $0.05     | scarce       |
| RTX A6000              | 48GB     | \~$0.36        | $0.36     | 2 GPUs total |
| H100 NVL               | 94GB     | \~$1.04        | $1.04     | 4 GPUs total |
| L40S                   | 48GB     | \~$0.50        | $0.50     | 1 GPU total  |

The fleet is overwhelmingly consumer silicon. **RTX 50-series is already 19.5% of all listed GPUs** (1,321 cards), the 30-series is the volume tier, and the 40-series sits in between.

## What is not on the marketplace

{% hint style="warning" %}
**A100, H200 and B200 are not listed on the marketplace today.** Not "rare" — zero. There are 4 H100 NVL GPUs across 2 servers, 2 A6000s and 1 L40S. If a guide sizes a workload at "8× H100", that configuration cannot be rented on-demand here.
{% endhint %}

Datacenter capacity is sold separately as [bare metal](https://clore.ai/bare-metal) — dedicated H100/H200/B200 nodes on request, contract rather than per-minute. For everything else, the practical ceiling is a multi-GPU consumer rig.

**The largest rigs listed today:**

| Rig                            | Total VRAM | Rig $/hr |
| ------------------------------ | ---------- | -------- |
| 4× RTX PRO 6000 Blackwell 96GB | 380GB      | \~$5.00  |
| 11× RTX 5090 32GB              | 341GB      | \~$5.42  |
| 10× RTX 5090 32GB              | 310GB      | \~$5.00  |
| 8× Tesla V100 SXM2 32GB        | 256GB      | \~$0.40  |
| 8× RTX 4090 24GB               | 192GB      | \~$1.60  |

Availability by size, from the snapshot:

| You need                          | Servers listed | Free at snapshot |
| --------------------------------- | -------------- | ---------------- |
| ≥ 24GB (one 3090/4090 class card) | 1,299          | 677              |
| ≥ 48GB                            | 653            | 322              |
| ≥ 80GB                            | 328            | 153              |
| ≥ 96GB                            | 182            | 81               |
| ≥ 192GB                           | 54             | 14               |
| ≥ 242GB                           | 47             | 9                |
| ≥ 352GB                           | 2              | 2                |

## What fits

Rough mapping from model weight size to what you should rent. Quantized GGUF sizes are the ones that matter in practice.

| Weights on disk | Rent this                           | Example                                                          |
| --------------- | ----------------------------------- | ---------------------------------------------------------------- |
| ≤ 16GB          | 1× RTX 3090 / 4090                  | Qwen3.8-27B Q4, Mistral Small 3.1, SDXL, most TTS                |
| 16–30GB         | 1× RTX 5090                         | Qwen3.8-27B FP8, 32B Q4/Q6, FLUX at high res                     |
| 30–60GB         | 2× RTX 4090 / 5090, or RTX PRO 6000 | Mistral Small 4 Q3–Q4, 70B Q4                                    |
| 60–192GB        | 4–8× RTX 5090                       | DeepSeek V4-Flash FP8, MiniMax M3 Q2                             |
| 192–380GB       | 8–11× RTX 5090, 4× RTX PRO 6000     | Nemotron 3 Ultra Q2, GLM-5.2 Q2                                  |
| > 380GB         | Not on the marketplace              | Kimi K3 at any quant → [bare metal](https://clore.ai/bare-metal) |

## Spot vs on-demand

Clore.ai runs two order types: **on-demand** at the host's fixed price, and **spot**, where you bid and can be outbid by a higher offer.

{% hint style="info" %}
**Spot is not automatically cheaper.** In the snapshot, **63% of servers listed spot at exactly the on-demand price**. Of the 34% that were cheaper, the median discount was **13.4%**. A handful were priced above on-demand. Check the actual pair on the server card rather than assuming a discount.
{% endhint %}

Spot still makes sense for interruptible work — batch rendering, dataset preprocessing, long fine-tunes with checkpointing — because you can also bid *above* the listed spot price to hold a machine other bidders want.

## Partial GPU rental

Roughly **60% of listed servers (1,331 of 2,202) allow partial rental**: you take 1 or 2 GPUs out of an 8-GPU rig and pay only for those, instead of renting the whole box. For single-GPU workloads this is usually the cheapest path to a good card, because large rigs often price below small ones per GPU. See [Partial GPU Rental](/guides/getting-started/partial-gpu-rental.md).

## Real cost examples

Computed from the median prices above, at per-minute billing.

| Job                                  | GPU                 | Time     | Cost        |
| ------------------------------------ | ------------------- | -------- | ----------- |
| 1,000 SDXL images                    | RTX 3090 @ $0.09/hr | \~2 h    | **\~$0.18** |
| 1M tokens through a 27B Q4 model     | RTX 4090 @ $0.20/hr | \~20 min | **\~$0.07** |
| 10 × 6-second video clips            | RTX 4090 @ $0.20/hr | \~20 min | **\~$0.07** |
| LoRA fine-tune, 7B, 1 epoch          | RTX 4090 @ $0.20/hr | \~3 h    | **\~$0.60** |
| Whisper transcription, 10 h of audio | RTX 3060 @ $0.04/hr | \~40 min | **\~$0.03** |
| TTS API running 24/7 for a month     | RTX 3060 @ $0.04/hr | 730 h    | **\~$29**   |
| Always-on 27B chat endpoint, a month | RTX 5090 @ $0.29/hr | 730 h    | **\~$212**  |

## Checking prices yourself

The marketplace listing is public. Prices in the feed are **per server per day** in USD — divide by 24 for hourly, and by the GPU count for per-GPU:

```bash
curl -s https://clore.ai/webapi/marketplace/servers \
  | python3 -c '
import json,sys,re
for s in json.load(sys.stdin)["all_servers"]:
    gpu = s["specs"]["gpu"]
    n = int(re.match(r"^(\d+)x", gpu).group(1)) if re.match(r"^(\d+)x", gpu) else 1
    usd = (s["price"]["usd"] or {}).get("on_demand_usd") or 0
    if usd and "4090" in gpu and not s["rented"]:
        print(f"{gpu:32} ${usd/n/24:.3f}/GPU/hr")
'
```

## Next Steps

* [GPU Comparison](/guides/getting-started/gpu-comparison.md) — which card for which workload
* [Partial GPU Rental](/guides/getting-started/partial-gpu-rental.md) — rent 1 GPU out of a rig
* [Cost Calculator](/guides/getting-started/cost-calculator.md) — budget a specific job
* [CUDA & PyTorch Compatibility](/guides/getting-started/cuda-pytorch-compatibility.md) — pick an image that runs on the card you rented

### Links

* [Marketplace](https://clore.ai/marketplace) · [Bare metal](https://clore.ai/bare-metal)
* [Rent an RTX 4090](https://clore.ai/rent-4090.html) · [Rent an RTX 5090](https://clore.ai/rent-5090.html)
* [GPU Cloud Pricing Comparison — Clore.ai vs Major Providers](https://blog.clore.ai/gpu-cloud-pricing-comparison/)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.clore.ai/guides/getting-started/pricing.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.