GPU model · price + stock · hourly snapshot
Nvidia RTX 5090 cloud rental prices
5 providers list the Nvidia RTX 5090, 2 have it in stock. Prices run from $0.59 to $0.70 per GPU hour — the same card, 1.2x apart.
LIVE Prices and stock checked 1 minute ago
Providers
Every provider renting the Nvidia RTX 5090
| Provider | From | Stock | Grab |
|---|---|---|---|
| Vast.ai | $0.59 | In stock | Grab |
| Runpod | $0.69 | In stock | Grab |
| Lium | $0.40 | Out of stock | Grab |
| Excess Supply | $0.65 | Out of stock | Grab |
| Phyntec | $0.70 | Out of stock | Grab |
In stock first, then cheapest. Stock is what the provider reported at our last check. One GPU for a month (730h) at $0.59: $433. Lium publishes live counts: 258 GPUs at $0.40, 100% rented.
The Nvidia RTX 5090 sells out fast. Get one email when it is back in stock.
Price history
Price trend
$0.400 /GPU/hr · Oct 22
One point per day: the lowest listed $/GPU/hr across every provider we track, captured daily. 7 offers in the latest snapshot.
Configurations
All 7 configurations
| Configuration | VRAM | Per GPU/hr | Stock |
|---|---|---|---|
| 4x RTX 5090Vast.ai | 127 GB | $0.59 | In stock |
| 1x RTX 5090Runpod | 32 GB | $0.69 | In stock |
| 1x NVIDIA GeForce RTX 5090Lium | — | $0.40 | Out of stock |
| 8x RTX5090Excess Supply | 256 GB | $0.65 | Out of stock |
| 4x RTX5090Excess Supply | 128 GB | $0.65 | Out of stock |
| 8x RTX5090Excess Supply | 256 GB | $0.70 | Out of stock |
| 8x RTX5090Phyntec | 256 GB | $0.70 | Out of stock |
Model fit
What fits in 32 GB
The question behind most rentals: will the model load. Sized against one Nvidia RTX 5090.
| Model | 16-bit | 8-bit | 4-bit |
|---|---|---|---|
| DeepSeek-V3 DeepSeek · 671B params · 37B active | 1,610 GB 51× · 22 GB free | 805 GB 26× · 27 GB free | 403 GB 13× · 13 GB free |
| Llama 3.1 405B Meta · 405B params | 972 GB 31× · 20 GB free | 486 GB 16× · 26 GB free | 243 GB 8× · 13 GB free |
| Qwen3 235B-A22B Alibaba · 235B params · 22B active | 564 GB 18× · 12 GB free | 282 GB 9× · 6 GB free | 141 GB 5× · 19 GB free |
| Mixtral 8x22B Mistral · 141B params · 39B active | 338 GB 11× · 14 GB free | 169 GB 6× · 23 GB free | 85 GB 3× · 11 GB free |
| Mistral Large 2 Mistral · 123B params | 295 GB 10× · 25 GB free | 148 GB 5× · 12 GB free | 74 GB 3× · 22 GB free |
| gpt-oss-120b OpenAI · 117B params · 5.1B active | 281 GB 9× · 7 GB free | 140 GB 5× · 20 GB free | 70 GB 3× · 26 GB free |
| Qwen2.5 72B Alibaba · 72.7B params | 174 GB 6× · 18 GB free | 87 GB 3× · 9 GB free | 44 GB 2× · 20 GB free |
| Llama 3.3 70B Meta · 70.6B params | 169 GB 6× · 23 GB free | 85 GB 3× · 11 GB free | 42 GB 2× · 22 GB free |
| Qwen3 32B Alibaba · 32.8B params | 79 GB 3× · 17 GB free | 39 GB 2× · 25 GB free | 20 GB 1× · 12 GB free |
| Gemma 2 27B Google · 27.2B params | 65 GB 3× · 31 GB free | 33 GB 2× · 31 GB free | 16 GB 1× · 16 GB free |
| Mistral Small 3 Mistral · 24B params | 58 GB 2× · 6 GB free | 29 GB 1× · 3 GB free | 14 GB 1× · 18 GB free |
| Llama 3.1 8B Meta · 8B params | 19 GB 1× · 13 GB free | 10 GB 1× · 22 GB free | 5 GB 1× · 27 GB free |
Required memory is weights plus 20% for KV cache and runtime at short context; long contexts and large batches need more. Cells tinted green fit on a single Nvidia RTX 5090.
Specifications
Nvidia RTX 5090 specifications
| Architecture | Blackwell |
|---|---|
| Memory per GPU | 32 GB GDDR7 |
| Memory bandwidth | 1,792 GB/s |
| Release date | Q1 2025 |
| FP4 compute rate | 1,676 TFLOPS (dense), 3,352 with sparsity |
| FP8 compute rate | 419 TFLOPS (dense), 838 with sparsity |
| FP16 / BF16 compute rate | 209.5 TFLOPS (dense), 419 with sparsity |
| INT8 compute rate | 838 TOPS (dense), 1,676 with sparsity |
| Process node | TSMC 4N |
| Board power | 575 W |
Alternatives