NVIDIA GTX 1070 Ti - 8GB usable
8GB Pascal that still runs 7B models on llama.cpp. No tensor cores and 256 GB/s, so it is a starting point, not a destination.
Specifications
| Brand | NVIDIA |
|---|---|
| Model | GTX 1070 Ti |
| Usable VRAM | 8GB |
| Architecture | Pascal |
| CUDA / Stream Processors | 2,432 |
| Memory Bandwidth | 256 GB/s |
| TDP | 180W |
| FP32 TFLOPS | 8.2 |
Current Offers
Used from $100New from $120
Prices last updated:
GPUDojo is reader-supported. When you buy through links on our site, we may earn an affiliate commission.
Price History
- eBay$100mid-range
- Amazon$120current
- Newegg$199near low
For AI / LLM Use
Limited VRAM restricts you to 7B quantized models. Slower generation, usable but not snappy. Older architecture may have limited software support (check CUDA compatibility).
What Models Can It Run?
- 7B Q6_K, 14B Q3_K (tight)
- 7B Q4_K_M only
Estimated Performance
Generation: ~19 tokens/sec
Prefill: ~146 tokens/sec
Recommended Quantisations
- Q4_K_M for 7B models
- Q3_K for larger experiments
Pros & Cons
Pros
- Consumer card: easy to install, display output
Cons
- Only 8GB usable VRAM: limited to small models
- Low memory bandwidth: slower token generation
- Older Pascal architecture: verify current CUDA support
Community Verdict
No community reviews yet for the GTX 1070 Ti. Know a good review? Let us know.