gpu catalog: t4
The classic low-power inference card: 70 watts and 16 GB, still everywhere and still fine for embeddings and classic ML serving.
| GPU | VRAM | Regions | Availability | Price / GPU | |
|---|---|---|---|---|---|
| T4 | 16G | — | out of stock | — | |
T4 is out of stock right now — these are live: The classic low-power inference card: 70 watts and 16 GB, still everywhere and still fine for embeddings and classic ML serving. VRAM16 GB GDDR6 Memory bandwidth0.3 TB/s FP16 tensor65 TFLOPS ArchitectureTuring InterconnectPCIe 3.0 Board power70 W CUDA cores2,560 Released2018 | |||||
| RTX PRO 6000 | 96G | 4 regions | high availability | $2.03/hr | launch |
| RTX A6000 | 48G | 2 regions | high availability | $0.54/hr | launch |
| RTX 6000 ADA | 48G | 2 regions | high availability | $0.78/hr | launch |
| L40 | 48G | 1 region | high availability | $0.94/hr | launch |
| RTX 5090community | 32G | 5 regions | high availability | $0.52/hr | launch |
| RTX 4090community | 24G | 7 regions | high availability | $0.47/hr | launch |
| RTX 3090community | 24G | 6 regions | high availability | $0.19/hr | launch |
| A100 | 80G | 4 regions | available | $1.48/hr | launch |
| H200 SXMcommunity | 141G | 5 regions | available | $4.31/hr | launch |
| H200 SXM | 141G | 5 regions | available | $6.35/hr | launch |
| RTX 4090 | 24G | 4 regions | available | $1.07/hr | launch |
| H100 SXM | 80G | 3 regions | available | $5.27/hr | launch |
| RTX 5080community | 16G | 7 regions | available | $0.28/hr | launch |
| RTX 3080community | 10G | 4 regions | available | $0.13/hr | launch |
| A16 | 16G | 6 regions | available | $0.56/hr | launch |
| L40Scommunity | 48G | 4 regions | available | $0.72/hr | launch |
| H200 NVLcommunity | 141G | 5 regions | available | $4.33/hr | launch |
| RTX A4000community | 16G | 4 regions | available | $0.10/hr | launch |
| RTX 4080community | 16G | 4 regions | running low | $0.29/hr | launch |
| A40 | 48G | 3 regions | running low | $0.71/hr | launch |
| L40S | 48G | 2 regions | running low | $0.96/hr | launch |
| H100 PCIE | 80G | 1 region | running low | $2.98/hr | launch |
| B200community | 192G | 1 region | running low | $9.64/hr | launch |
| L4community | 24G | 2 regions | running low | $0.53/hr | launch |
| A100 40GBcommunity | 40G | 3 regions | running low | $0.45/hr | launch |
| A40community | 48G | 1 region | running low | $0.72/hr | launch |
| RTX PRO 4500 | 32G | 1 region | running low | $0.86/hr | launch |
| RTX 5090 | 32G | 1 region | running low | $1.67/hr | launch |
| H100 SXMcommunity | 80G | 2 regions | running low | $4.50/hr | launch |
| B200 | 192G | 1 region | running low | $9.59/hr | launch |
| A100community | 80G | 1 region | running low | $1.33/hr | launch |
| H200 NVL | 141G | 1 region | running low | $3.95/hr | launch |
| V100community | 16G | 2 regions | running low | $0.14/hr | launch |
| RTX A6000community | 48G | 1 region | running low | $0.58/hr | launch |
| H100 NVLcommunity | 94G | 1 region | running low | $4.39/hr | launch |
| RTX A5000community | 24G | 1 region | running low | $0.28/hr | launch |
lowest live on-demand rate per GPU, refreshed continuously · per-second billing
Our floor price for the T4, sampled continuously across every provider in the fleet and rolled up to one point per day. A dot on the zero line is a day we checked and found nothing launchable; a shaded break is a day we did not sample. Those are different facts, so this chart never draws one as the other.
We have not collected enough daily samples for the T4 yet. The trend appears here once history accumulates — we do not draw a line through data we never had.
faq
T4 capacity is temporarily out of stock on GPU.ai. When available, it's billed per second at the live rate shown here. Check this page again. Stock and prices refresh continuously.
The NVIDIA Tesla T4 has 16 GB of GDDR6 memory with up to 320 GB/s of bandwidth. GPU.ai reports the minimum VRAM actually delivered across our fleet, so the figure you see at launch is the figure you get.
GPU.ai aggregates a dozen-plus GPU clouds and routes each launch to the cheapest one with stock, so the same silicon is typically far cheaper than hyperscaler list rates. The comparison table on this page shows the current public on-demand list prices side by side.
Billing is per second, starting when your instance is SSH-ready and stopping the moment you terminate. There are no minimum commitments, no reservation fees, and no egress surprises: you pay the listed hourly rate pro-rated to the second.