For the Turing architecture, the NVIDIA Tesla T4 was the flagship GPU for inference, lightweight training, and general-purpose cloud workloads.
But in 2026, with newer and significantly more powerful GPUs widely available, is it still worth using?
Takeaways
<ul><li>The <strong>NVIDIA T4 price</strong> varies dramatically depending on the provider.</li><li>Marketplaces like Vast.ai offer the lowest pricing. But this is subject to availability; there are currently only 4 listings on the marketplace.</li><li>Hyperscalers (AWS, Azure, GCP) charge significantly more for the same aging hardware.</li><li>Most major providers phased it out in favor of newer alternatives.</li><li>At current pricing, the T4 competes directly with <strong>newer GPUs that offer significantly better performance per dollar</strong>.</li></ul>
| Provider | Instance Name | Price (USD/hr) |
|---|---|---|
| Vast.ai | $0.15 | |
| Lyceum | $0.39 | |
| Google Cloud | n1-standard-4 + 1x T4 GPU | $0.40 |
| AWS | g4dn.xlarge | $0.53 |
| Cerebrium | $0.59 | |
| Azure | Standard_NC4as_T4_v3 | $0.59 |
| Replicate | $0.81 | |
| * Last update: May 19, 2026 | ||
T4 vs Newer GPUs
| GPU | VRAM | Price/hr |
|---|---|---|
| T4 | 16GB | $0.15-$0.81 |
| RTX A6000 | 48GB | $0.35 |
| A100 | 80GB | $0.78 |
Modern GPUs outperform the T4 at similar or slightly higher prices.
AWS G4 Instances
AWS offers NVIDIA T4 GPUs in its EC2 G4 Instances.
G4 instance characteristics:
<ul><li>NVIDIA T4 GPUs</li><li>4-96 vCPUs</li><li>16-384GB RAM</li><li>125-1800GB NVMe SSD storage</li></ul>
As usual for AWS instances, egress fees are $0.09/GB after the first free 100GB.
| Provider | SKU | T4 GPUs | VCPUs | RAM | Hourly price | Price per-GPU |
|---|---|---|---|---|---|---|
| AWS | g4dn.xlarge | 1 | 4 | 16 | $0.53-$0.63 | |
| AWS | g4dn.2xlarge | 1 | 8 | 32 | $0.75-$0.90 | |
| AWS | g4dn.4xlarge | 1 | 16 | 64 | $1.20-$1.45 | |
| AWS | g4dn.8xlarge | 1 | 32 | 128 | $2.18-$2.61 | |
| AWS | g4dn.16xlarge | 1 | 64 | 256 | $4.35-$4.69 | |
| AWS | g4dn.12xlarge | 4 | 48 | 196 | $3.91-$5.22 | $0.98-$1.31 |
| AWS | g4dn.metal | 8 | 96 | 384 | $7.82-$9.39 | $0.98-$1.17 |
Azure NC T4 Instances
Through the NCas_T4_v3-series, Azure offers compute-optimized virtual machines tailored for cost-effective AI deployment.
Powered by NVIDIA Tesla T4 Tensor Core GPUs and AMD EPYC 7V12 (Rome) processors, these instances provide an ideal platform for running real-time machine learning inference, standard CUDA-accelerated workloads, and interactive graphics applications.
| Provider | SKU | T4 GPUs | VCPUs | RAM | Hourly price | Price per-GPU |
|---|---|---|---|---|---|---|
| Azure | Standard_NC4as_T4_v3 | 1 | 4 | 28 | $0.59 | |
| Azure | Standard_NC8as_T4_v3 | 1 | 8 | 56 | $0.85 | |
| Azure | Standard_NC16as_T4_v3 | 1 | 16 | 110 | $1.36 | |
| Azure | Standard_NC64as_T4_v3 | 4 | 64 | 440 | $4.91 | $1.23 |
Google Cloud N1 instances
Through general-purpose and custom N1 machine types, GCP offers flexible compute attachment with NVIDIA T4 GPUs:
<ul><li>N1 Standard, High-Memory, and High-CPU (n1-standard, n1-highmem, n1-highcpu) attached with NVIDIA T4 16GB.</li><li>Custom N1 configurations (custom-) that allow you to choose your own vCPU and RAM layout alongside the T4 16GB.</li></ul>
Being a highly customizable instance, the table below is just illustrative of the smallest and largest machine types.
| Provider | SKU | T4 GPUs | VCPUs | RAM | Hourly price | Price per-GPU |
|---|---|---|---|---|---|---|
| GCP | n1-standard-4 + 1xT4 | 1 | 4 | 15 | $0.57 | |
| GCP | n1-standard-96 + 4xT4 | 4 | 96 | 360 | $6.85 | $1.71 |
| GCP | n1-highcpu-4 + 1xT4 | 1 | 4 | 3.6 | $0.51 | |
| GCP | n1-highcpu-96 + 4xT4 | 4 | 96 | 86.4 | $5.31 | $1.33 |
| GCP | n1-highmem-4 + 1xT4 | 1 | 4 | 26 | $0.62 | |
| GCP | n1-highmem-96 + 4xT4 | 4 | 96 | 624 | $8.14 | $2.04 |
Methodology (why you can trust these numbers)
<ul><li>Pricing reflects <strong>public on-demand rates</strong> as of May 2026.</li><li>Only <strong>comparable instance types</strong> (single GPU, standard configurations) were included.</li><li>Marketplace pricing (e.g., Vast.ai) reflects <strong>currently available listings</strong>, which may fluctuate.</li><li>Hyperscaler pricing (AWS, Azure, Google Cloud) reflects <strong>standard on-demand pricing</strong>, excluding reserved or spot discounts.</li></ul>
Why this matters for developers
This chart compares how much it costs to run an NVIDIA T4 for 10 hours across major cloud providers.
Thunder Compute offers NVIDIA RTX A6000 GPUs for $0.35/hr which have 3 times the VRAM as NVIDIA T4.
Using an outdated GPU like the T4 at similar prices to modern alternatives can significantly increase cost per experiment and slow down iteration cycles.
FAQ
What is the NVIDIA T4 still used for?
In 2026, the NVIDIA T4 is used for inference workloads, video processing, and light machine learning tasks.
How much does the NVIDIA T4 cost?
The NVIDIA T4 price ranges from $0.15/hr to $0.81/hr depending on the provider.
How much memory does the NVIDIA T4 have?
The NVIDIA T4 has 16GB of GDDR6 VRAM.
Can the NVIDIA Tesla T4 handle modern deep learning?
In 2026, it can handle small models and inference, but it's not ideal for modern deep learning workloads due to limited VRAM and performance.
Is the NVIDIA T4 outdated?
Yes, while still functional, it is being phased out and replaced by more powerful and cost-efficient GPUs.