RTX 4090-Ready Models
These popular AI models are optimized to run on the RTX 4090. This is not a complete list—browse our full model library to discover more options.
DeepSeek OCRRTX 4090 GPU Details
| Category | Detail |
|---|---|
| GPU Name | AD102 |
| GPU Codename | AD102 |
| Architecture | Ada Lovelace |
| GPCs | 11 |
| TPCs | 64 |
| SMs | 128 |
| CUDA Cores / SM | 128 |
| CUDA Cores / GPU | 16384 |
| Tensor Cores / SM | 4 (4th Gen) |
| Tensor Cores / GPU | 512 (4th Gen) |
| OFA TOPS | 305 |
| RT Cores | 128 (3rd Gen) |
| GPU Boost Clock (MHz) | 2520 |
| Peak FP32 TFLOPS (non-Tensor) | 82.6 |
| Peak FP16 TFLOPS (non-Tensor) | 82.6 |
| Peak BF16 TFLOPS (non-Tensor) | 82.6 |
| Peak INT32 TOPS (non-Tensor) | 41.3 |
| RT TFLOPS | 191 |
| Peak FP8 Tensor TFLOPS with FP16 Accumulate | 660.6/1321.22 |
| Peak FP8 Tensor TFLOPS with FP32 Accumulate | 660.6/1321.22 |
| Peak FP16 Tensor TFLOPS with FP16 Accumulate | 330.3/660.62 |
| Peak FP16 Tensor TFLOPS with FP32 Accumulate | 165.2/330.42 |
| Peak BF16 Tensor TFLOPS with FP32 Accumulate | 165.2/330.42 |
| Peak TF32 Tensor TFLOPS | 82.6/165.22 |
| Peak INT8 Tensor TOPS | 660.6/1321.22 |
| Peak INT4 Tensor TOPS | 1321.2/2642.42 |
| Frame Buffer Memory Size and Type | 24 GB GDDR6X |
| Memory Interface | 384-bit |
| Memory Clock (Data Rate) | 21 Gbps |
| Memory Bandwidth | 1008 GB/sec |
| ROPs | 176 |
| Pixel Fill-rate (Gigapixels/sec) | 443.5 |
| Texture Units | 512 |
| Texel Fill-rate (Gigatexels/sec) | 1290.2 |
| L1 Data Cache/Shared Memory | 16384 KB |
| L2 Cache Size | 73728 KB |
| Register File Size | 32768 KB |
| Video Engines | 2 x NVENC (8th Gen), 1 x NVDEC (5th Gen) |
| TGP (Total Graphics Power) | 450 W |
| Transistor Count | 76.3 Billion |
| Die Size | 608.5 mm2 |
| Manufacturing Process | TSMC 4N NVIDIA Custom Process |
| PCI Express Interface | Gen 4 |
