Top Ada Lovelace consumer GPU for local AI research and rendering.
Weekly median price per GPU per hour
By provider shows one card per company, placed where its best offer ranked. By configuration lists every offer.
What you can rent comes first: in stock, then waitlist, then not reported, then out of stock. Priced offers rank ahead of quote-only, and on-demand ahead of other billing types.
Within a group, five factors set the order:
We match the provider, its country, GPU model, form factor, billing type, availability and instance name. Matching is partial and case-insensitive. Everyday words work too, so "interruptible" finds spot and "sold out" finds out of stock.
USA
6 configs
1x-8x
USA
1 config
1x
USA
107 configs
1x-16x
USA
20 configs
1x-8x
USA
1 config
1x
USA
28 configs
1x-8x
USA
2 configs
1x-2x
France
7 configs
1x-8x
UAE
6 configs
1x-16x
Singapore
2 configs
1x-2x
Singapore
4 configs
1x-8x
USA
1 config
1x
No providers match .
Heads up: A provider's own page may quote a different figure, for example on tax, region or a promotion. A monthly-only plan shows a derived hourly rate. Verify before provisioning. More on how we price.
One RTX 4090 has 24 GB of VRAM. In practice, that's enough memory for roughly 28B parameters at 4-bit or 6B at 16-bit, assuming a 32K context. Below are some open-weight LLMs, with the estimated memory and GPUs each one needs.
| Model | Memory (INT4 / FP4) | RTX 4090s needed | Cost /hr | Cost /mo |
|---|---|---|---|---|
Qwen3.8-27B
27B
|
18 GB
|
1
|
$0.45
|
$324
|
Gemma 4 31B
30.7B
|
22 GB
|
1
|
$0.45
|
$324
|
GPT-OSS-120B
117B · 5.1B active
|
67 GB
|
4
|
$1.80
|
$1,296
|
GLM-5.3-Flash
320B · 18B active
|
178 GB
|
9
|
–
|
–
|
MiniMax-M3
428B · 23B active
|
239 GB
|
12
|
–
|
–
|
DeepSeek V4.1 Flash
552B · 16B active
|
306 GB
|
15
|
–
|
–
|
Kimi K3
2.8T · 104B active
|
1,544 GB
|
72
|
–
|
–
|
Estimates based on the median on-demand rate. Memory is weights plus FP8 KV cache at 32K context per request (FP16/BF16 in the 16-bit column). GPU counts assume 90% of advertised VRAM is usable. No guarantee of runtime support, usable performance, or that a matching quantized build exists. How we estimate costs.
Nvidia RTX 4090 · Per GPU
| Compute · dense | |
|---|---|
| FP8 | 330.3 TFLOPS 660.6 with sparsity |
| FP16 / BF16 | 165.2 TFLOPS 330.4 with sparsity |
| INT8 | 660.6 TOPS 1,321.2 with sparsity |
| FP32 | 82.6 TFLOPS |
| Precision support | FP8FP16BF16TF32FP32INT4INT8 |
| Memory | |
|---|---|
| Capacity | 24 GB GDDR6X |
| Bandwidth | 1,010 GB/s |
| Bus width | 384-bit |
| ECC | No |
| Silicon | |
|---|---|
| Architecture | Ada Lovelace |
| Process | TSMC 4N |
| Transistors | 76.3 billion |
| Shader cores | 16,384 CUDA cores |
| Matrix cores | 512 Tensor cores |
| Compute units | 128 SMs |
| Fabric and host | |
|---|---|
| Host interface | PCIe 4.0 |
| Power | |
|---|---|
| Board power | 450 W |
| Cooling | Active |
Source: official Nvidia RTX 4090 datasheet.
As of September 14, 2026, the median on-demand price is $0.45 per GPU per hour across 18 providers with a priced on-demand config.
| Billing type | Configs | Median /GPU/hr | Cheapest |
|---|---|---|---|
On-demand | 98 | $0.45 | $0.23 (in stock, Theta EdgeCloud) |
Reserved | 68 | $0.47 | $0.27 (12 mo, in stock, Novita) |
Spot | 47 | $0.17 | $0.12 (in stock, Vast.ai) |
The cheapest verified in-stock on-demand RTX 4090 tracked by GetDeploying is $0.23 per GPU per hour from Theta EdgeCloud. Cheapest verified in stock on other billing types: reserved $0.27 (12 mo) from Novita and spot $0.12 from Vast.ai.
At 720 hours per month, one RTX 4090 costs an estimated $324 at the median on-demand price. Cheapest verified in stock: $166 per month on-demand, $194 reserved (12 mo), $86 spot.
GetDeploying currently tracks RTX 4090 configs from 20 providers. The cheapest verified in-stock on-demand configs come from Theta EdgeCloud, Vast.ai, Lium and Novita. See the full price comparison above for every provider and config.
As of September 14, 2026, the median on-demand price has fallen about 11% over the past 90 days to $0.45 per GPU per hour, and about 14% over the past 12 months.
One RTX 4090 runs models up to roughly 28B parameters at 4-bit quantization or 6B at 16-bit, assuming a 32K context. Larger models run across multiple GPUs: the model table above shows the estimated memory and GPU count for popular open-weight LLMs.
24GB GDDR6X with Ada Lovelace FP8 Tensor Cores. Highest-performing consumer GPU of the Ada generation. Can run 13B parameter models at FP16 or 30B+ quantized. Popular for AI research and Stable Diffusion.
Consumer GeForce GPU. No ECC memory. 24GB VRAM is limiting for larger LLMs. Subject to Nvidia's consumer EULA, which may restrict data center use. For production inference, consider the L40S.
Last updated
