Low-cost inference GPU with wide cloud availability.
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.
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 T4 has 16 GB of VRAM. In practice, that's enough memory for roughly 8B parameters at 4-bit or 2B 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) | T4s needed | Cost /hr | Cost /mo |
|---|---|---|---|---|
Qwen3.8-27B
27B
|
19 GB
|
2
|
$1.28
|
$922
|
Gemma 4 31B
30.7B
|
24 GB
|
2
|
$1.28
|
$922
|
GPT-OSS-120B
117B · 5.1B active
|
68 GB
|
8
|
$5.12
|
$3,686
|
GLM-5.3-Flash
320B · 18B active
|
178 GB
|
13
|
–
|
–
|
MiniMax-M3
428B · 23B active
|
241 GB
|
17
|
–
|
–
|
DeepSeek V4.1 Flash
552B · 16B active
|
306 GB
|
22
|
–
|
–
|
Kimi K3
2.8T · 104B active
|
1,545 GB
|
108
|
–
|
–
|
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. How we estimate costs.
Nvidia T4 · Per GPU
| Compute · dense | |
|---|---|
| FP16 / BF16 | 65 TFLOPS |
| INT8 | 130 TOPS |
| FP32 | 8.1 TFLOPS |
| Precision support | FP16FP32INT4INT8 |
| Memory | |
|---|---|
| Capacity | 16 GB GDDR6 |
| Bandwidth | 300 GB/s |
| Bus width | 256-bit |
| ECC | Yes |
| Silicon | |
|---|---|
| Architecture | Turing |
| Process | TSMC 12nm FFN |
| Transistors | 13.6 billion |
| Shader cores | 2,560 CUDA cores |
| Matrix cores | 320 Tensor cores |
| Compute units | 40 SMs |
| Fabric and host | |
|---|---|
| Host interface | PCIe 3.0 x16 |
| Power | |
|---|---|
| Board power | 70 W |
| Cooling | Passive |
Source: official Nvidia T4 datasheet.
At 720 hours per month, one T4 costs an estimated $461 at the median on-demand price. Cheapest verified in stock: $50 per month on-demand, $36 spot.
GetDeploying currently tracks T4 configs from 11 providers. Vast.ai and Zenlayer have verified in-stock on-demand configs. See the full price comparison above for every provider and config.
As of September 16, 2026, the median on-demand price has fallen about 6% over the past 90 days to $0.64 per GPU per hour, though it is about 5% above where it was a year ago.
One T4 runs models up to roughly 8B parameters at 4-bit quantization or 2B 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.
16GB GDDR6 in a 70W low-power form factor. Very low cost per hour across cloud providers. INT8 Tensor Cores for efficient inference on smaller models.
Turing architecture lacks BF16 and FP8. 16GB VRAM limits model sizes. Slow for training. For production inference, the L4 offers significantly better throughput.
Nvidia L40S
48 GB GDDR6 · 2.9x the bandwidth of the T4
Nvidia L4
24 GB GDDR6 · released Q1 2023
Nvidia A10
24 GB GDDR6 · 2x the bandwidth of the T4
Last updated
