NVIDIA T4 - 16GB usable
70W, single-slot, 16GB, and it runs off slot power in almost any server. Only 320 GB/s, so it favours always-on inference over speed.
Specifications
| Brand | NVIDIA |
|---|---|
| Model | T4 |
| Usable VRAM | 16GB |
| Architecture | Turing |
| CUDA / Stream Processors | 2,560 |
| Memory Bandwidth | 320 GB/s |
| TDP | 70W |
| FP32 TFLOPS | 8.1 |
Current Offers
Used from €690
Prices last updated:
GPUDojo is reader-supported. When you buy through links on our site, we may earn an affiliate commission.
Price History
- eBay€690at high
For AI / LLM Use
Good for 14B models. 30B requires aggressive quantization. Slower generation, usable but not snappy. Datacenter card with no display output, may need aftermarket cooling.
What Models Can It Run?
- 14B Q6_K, 30B Q3_K (tight)
- 14B Q4_K_M, 7B full precision
- 7B Q6_K, 14B Q3_K (tight)
- 7B Q4_K_M only
Estimated Performance
Generation: ~24 tokens/sec
Prefill: ~145 tokens/sec
Recommended Quantisations
- Q4_K_M for 14B models
- Q6_K for 7B-8B models
- Q8 for 7B if VRAM allows
Pros & Cons
Pros
- Only 70W TDP: power efficient
- Turing architecture: broad CUDA software support
Cons
- 16GB usable VRAM: may need quantization for 30B+ models
- Moderate memory bandwidth: not the fastest for inference
- No display output: headless only
- May need aftermarket cooling solution
Community Verdict
No community reviews yet for the T4. Know a good review? Let us know.