Budget legacy GPU when VRAM matters more than speed.
Weekly median price per GPU per hour · Get the data
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.
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 P40 has 24 GB of VRAM. In practice, that's enough memory for roughly 21B 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) | P40s needed | Cost /hr | Cost /mo |
|---|---|---|---|---|
Qwen3.8-27B
27B
|
19 GB
|
1
|
–
|
–
|
Gemma 4 31B
30.7B
|
24 GB
|
2
|
–
|
–
|
GPT-OSS-120B
117B · 5.1B active
|
68 GB
|
4
|
–
|
–
|
GLM-5.3-Flash
320B · 18B active
|
178 GB
|
9
|
–
|
–
|
MiniMax-M3
428B · 23B active
|
241 GB
|
12
|
–
|
–
|
DeepSeek V4.1 Flash
552B · 16B active
|
306 GB
|
15
|
–
|
–
|
Kimi K3
2.8T · 104B active
|
1,545 GB
|
72
|
–
|
–
|
Estimates based on the median on-demand rate. Memory is weights plus FP16/BF16 KV cache at 32K context per request. GPU counts assume 90% of advertised VRAM is usable. No guarantee of runtime support, usable performance, or that a matching quantized build exists. Pricing methodology.
Nvidia P40 · Per GPU
| Compute · dense | |
|---|---|
| INT8 | 47 TOPS |
| FP32 | 12 TFLOPS |
| Precision support | FP16FP32INT8 |
| Memory | |
|---|---|
| Capacity | 24 GB GDDR5 |
| Bandwidth | 346 GB/s |
| Bus width | 384-bit |
| ECC | Yes |
| Silicon | |
|---|---|
| Architecture | Pascal |
| Process | TSMC 16nm FinFET |
| Transistors | 11.8 billion |
| Shader cores | 3,840 CUDA cores |
| Compute units | 30 SMs |
| Fabric and host | |
|---|---|
| Host interface | PCIe 3.0 x16 |
| Power | |
|---|---|
| Board power | 250 W |
| Cooling | Passive |
Source: official Nvidia P40 datasheet.
The cheapest verified in-stock estimate is $58 per month on-demand, $50 spot.
GetDeploying currently tracks P40 configs from 1 provider. Vast.ai has verified in-stock on-demand configs. See the full price comparison above for every provider and config.
One P40 runs models up to roughly 21B 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 GDDR5 at very low cost. Legacy option for workloads that need more VRAM than the P4. Some availability from budget cloud providers.
Pascal architecture with no mixed precision support. GDDR5 bandwidth is very low. Only viable for legacy batch inference where VRAM matters more than speed.
Nvidia L4
24 GB GDDR6 · 13% less bandwidth than the P40
Nvidia T4
16 GB GDDR6 · 13% less bandwidth than the P40
Nvidia A10
24 GB GDDR6 · 73% more bandwidth than the P40
Last updated
