NVIDIA GeForce RTX 3090 24GB Specifications for AI Enthusiasts
The NVIDIA GeForce RTX 3090 is NVIDIA's flagship consumer graphics card launched in September 2020, featuring the Ampere architecture based on the GA102 GPU (GA102-300-A1 variant). Built on Samsung's 8nm process with 28.3 billion transistors and a die size of 628 mm², it offers 10,496 CUDA cores across 82 streaming multiprocessors, 328 3rd-generation tensor cores, and 82 2nd-generation RT cores. The card features an unprecedented 24GB of GDDR6X memory running at 19.5 Gbps on a 384-bit bus, delivering 936.2 GB/s of memory bandwidth. With a TDP of 350W, the RTX 3090 was designed for 8K HDR gaming and professional creative workflows. The 3rd-generation Tensor Cores support fine-grained structured sparsity (2:4 pattern), enabling up to 571 INT8 TOPS with sparsity and 143 FP16 TFLOPS for dense tensor operations. The RTX 3090 uses a PCI-Express 4.0 x16 interface and is one of the few consumer GPUs to support 2-way NVLink. At launch, it was positioned as the spiritual successor to NVIDIA's Titan series, offering professional-grade memory capacity in a GeForce product.
- Release Date: September 24, 2020
- GPU Architecture: Ampere
- Hardware-Accelerated GEMM Operations:FP16 FP32 BF16 FP8 INT8 INT4 TF32 FP64 INT1
- CUDA Compute Capability : 8.6
Specifications for NVIDIA GeForce RTX 3090
| Raw Performance | |
|---|---|
| FP32 TFLOPs: 35.58 | |
| FP16 TFLOPs: 142.74 | |
| Tensor Core Count: 328 | |
| Memory Capacity (GB): 24 | |
| Memory Bandwidth (GB/s): 936.2 | |
| Int8 TOPs: 285 |
Real-time NVIDIA GeForce RTX 3090 GPU Prices
Compare Price/Performance to other GPUs
References
- https://www.nvidia.com/content/PDF/nvidia-ampere-ga-102-gpu-architecture-whitepaper-v2.1.pdf
- https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3090-3090ti/
- https://www.techpowerup.com/gpu-specs/geforce-rtx-3090.c3622
- https://www.tomshardware.com/features/nvidia-ampere-architecture-deep-dive
- https://forums.developer.nvidia.com/t/rtx-3090-peak-performance/198049