Model (Full Name) | Type | Launch Time | Architecture | CUDA Cores | Tensor Cores | GPU VRAM | vGPU Support | NVLink Support | FP32 Performance | Power Consumption |
---|---|---|---|---|---|---|---|---|---|---|
NVIDIA GeForce GT 710 | Consumer | 2014 | Kepler | 192 | No | 2GB DDR3 | No | No | 0.336 TFLOPS | 19W |
NVIDIA GeForce GT 730 | Consumer | 2014 | Kepler | 384 | No | 2GB DDR3 | No | No | 0.692 TFLOPS | 38W |
NVIDIA Quadro K620 | Workstation | 2014 | Maxwell | 384 | No | 2GB DDDR5 | No | No | 0.863 TFLOPS | 45W |
NVIDIA Quadro P600 | Workstation | 2017 | Pascal | 384 | No | 2GB GDDR5 | No | No | 1.2 TFLOPS | 40W |
NVIDIA Quadro P620 | Workstation | 2017 | Pascal | 512 | No | 2GB GDDR5 | No | No | 1.2 TFLOPS | 40W |
NVIDIA Quadro P1000 | Workstation | 2017 | Pascal | 640 | No | 4GB GDDR5 | No | No | 1.894 TFLOPS | 47W |
NVIDIA RTX T1000 | Workstation | 2021 | Turing | 896 | 28 | 4GB GDDR6 | No | No | 2.5 TFLOPS | 50W |
NVIDIA GeForce GTX 1650 | Consumer | 2019 | Turing | 896 | No | 4GB GDDR6 | No | No | 3.0 TFLOPS | 75W |
NVIDIA GeForce GTX 1660 | Consumer | 2019 | Turing | 1408 | No | 6GB GDDR6 | No | No | 5.0 TFLOPS | 120W |
NVIDIA GeForce RTX 2060 | Consumer | 2019 | Turing | 1920 | 240 | 6GB GDDR6 | No | No | 6.5 TFLOPS | 160W |
NVIDIA GeForce RTX 3060 Ti | Consumer | 2020 | Ampere | 4864 | 152 | 8GB GDDR6 | No | No | 16.2 TFLOPS | 200W |
NVIDIA RTX A4000 | Workstation | 2021 | Ampere | 6144 | 192 | 16GB GDDR6 | Yes | No | 19.2 TFLOPS | 140W |
NVIDIA RTX A5000 | Workstation | 2021 | Ampere | 8192 | 256 | 24GB GDDR6 | Yes | Yes | 27.8 TFLOPS | 230W |
NVIDIA RTX A6000 | Workstation | 2021 | Ampere | 10752 | 336 | 48GB GDDR6 | Yes | Yes | 38.71 TFLOPS | 300W |
NVIDIA GeForce RTX 4060 | Consumer | 2023 | Ada Lovelace | 3072 | 96 | 8GB GDDR6 | No | No | 15.11 TFLOPS | 115W |
NVIDIA GeForce RTX 4090 | Consumer | 2022 | Ada Lovelace | 16384 | 512 | 24GB GDDR6X | No | Yes | 82.6 TFLOPS | 450W |
NVIDIA GeForce RTX 5060 | Consumer | 2025 | Ada Lovelace | TBD | TBD | TBD | No | No | 23.22 TFLOPS | 180W |
NVIDIA GeForce RTX 5090 | Consumer | 2025 | Ada Lovelace | TBD | TBD | TBD | No | Yes | 109.7 TFLOPS | 575W |
NVIDIA Tesla K80 | Data Center | 2014 | Kepler | 4992 | No | 24GB GDDR5 | No | Yes | 8.73 TFLOPS | 300W |
NVIDIA Tesla V100 | Data Center | 2017 | Volta | 5120 | 640 | 16GB HBM2 | Yes | Yes | 14 TFLOPS | 250W |
NVIDIA Tesla P100 | Data Center | 2016 | Pascal | 3584 | No | 16GB HBM2 | Yes | Yes | 9.5 TFLOPS | 250W |
NVIDIA RTX A40 | Workstation | 2020 | Ampere | 10752 | 336 | 48GB GDDR6 | Yes | Yes | 37.48 TFLOPS | 300W |
NVIDIA A100(40G) | Data Center | 2020 | Ampere | 6912 | 432 | 40GB HBM2e | Yes | Yes | 19.5 TFLOPS | 400W |
NVIDIA A100(80G) | Data Center | 2020 | Ampere | 6912 | 432 | 80GB HBM2e | Yes | Yes | 19.5 TFLOPS | 400W |
NVIDIA H100 | Data Center | 2022 | Hopper | 16896 | 528 | 80GB HBM3 | Yes | Yes | 183 TFLOPS | 700W |
Type: Refers to its main use and target users. GeForce is for gaming and everyday graphics, Quadro/RTX A is for professional work like 3D design and rendering, and Tesla/A/H series is for AI, high-performance computing, and data centers.
Architectuer: Represents the underlying chip design generation of the GPU, impacting performance, power efficiency, AI capabilities, and feature support. Each architecture brings advancements tailored for specific compute needs.
CUDA Cores: Determines parallel processing power, essential for AI training and complex computations.
Tensor Cores: Boosts AI computations and deep learning efficiency, available in NVIDIA’s latest GPUs.
VRAM (Video Memory): Affects handling of large datasets; critical for deep learning and video rendering.
Memory Bandwidth: Impacts data transfer speed between GPU and memory, important for high-performance computing.
FP32 Performance(TFLOPS): Measures floating-point computing power, crucial for machine learning and simulations.
Power Consumption (Wattage): Important for energy efficiency and infrastructure planning.