NVIDIA RTX 5080 (Blackwell)
Overview
NVIDIA RTX 5080 is the second-tier flagship of the RTX 50 series, launched January 30, 2025, positioned in the mid-to-high range. With 16GB GDDR7 and 10,752 CUDA cores, it delivers approximately 50% of the RTX 5090's performance.
Core Specifications
| Item | Spec |
|---|
| Architecture | Blackwell (GB203) |
| Process | TSMC 4N |
| CUDA Cores | 10,752 |
| Tensor Cores | 336 (5th Gen) |
| RT Cores | 84 (4th Gen) |
| Boost Clock | 2.62 GHz |
| Memory | 16 GB GDDR7 |
| Memory Bandwidth | 960 GB/s (30 Gbps × 256-bit) |
| FP16 Tensor | 196 TFLOPS (sparse) |
| FP8 Tensor | 392 TFLOPS (sparse) |
| FP4 Tensor | 1,568 TOPS (sparse) |
| TDP | 360 W |
| Launch Price | $999 |
| Launch Date | 2025-01-30 |
Comparison with RTX 4080 SUPER
| Metric | RTX 5080 | RTX 4080 SUPER | Improvement |
|---|
| CUDA Cores | 10,752 | 10,240 | 1.05× |
| Memory | 16GB GDDR7 | 16GB GDDR6X | Same capacity / faster |
| Memory Bandwidth | 960 GB/s | 736 GB/s | 1.30× |
| FP4 Tensor | 1,568 TOPS | N/A | New |
| TDP | 360W | 320W | 1.13× |
| Price | $999 | $999 | Same |
| Model | Quantization | Performance |
|---|
| Llama 3 8B | FP16 | ~50 tok/s |
| Llama 3 13B | INT4 | ~38 tok/s |
| Mistral 7B | FP16 | ~62 tok/s |
| Mixtral 8x7B | INT4 | ~25 tok/s |
16GB memory can run 13B quantized models or 7B at full precision.
| Item | Detail |
|---|
| Vendor | NVIDIA |
| Launch Price | $999 |
| Target Market | Mainstream consumer AI, 1080p/1440p gaming, Stable Diffusion |