A summary of the benchmarked GPUs

NVIDIA’s newest products are included, including the Ampere GPU generation. The functionality of multi-GPU configurations, such as a quad RTX 3090 arrangement, is also assessed.

This section covers some options for local GPUs that are currently some of the best suited for deep learning training and development owing to their compute and memory performance and connectivity with current deep learning frameworks.

GPU NameDescription
GTX 1080TINVIDIA’s traditional GPU for Deep Learning was introduced in 2017 and was geared for computing tasks, featuring 11 GB DDR5 memory and 3584 CUDA cores. It has been out of production for some time and was just added as a reference point.
RTX 2080TIThe RTX 2080 TI was introduced in the fourth quarter of 2018. It has 5342 CUDA cores structured as 544 NVIDIA Turing mixed-precision Tensor Cores with 107 Tensor TFLOPS of AI capability and 11 GB of ultra-fast GDDR6 memory. This GPU was discontinued in September 2020 and is no longer available.
Titan RTXThe Titan RTX is powered by the most powerful TuringTM architecture. With 576 tensor cores and 24 GB of ultra-fast GDDR6 memory, the Titan RTX provides 130 Tensor TFLOPs of acceleration.
Quadro RTX 6000The Quadro RTX 6000 is the server variant of the famous Titan RTX, with enhanced multi-GPU blower ventilation, expanded virtualization functionality, and ECC memory. It uses the same TuringTM core as the Titan RTX, which has 576 tensor cores and delivers 130 Tensor TFLOPs of productivity as well as 24 GB of ultra-fast GDDR6 ECC memory.
Quadro RTX 8000The Quadro RTX 8000 is the RTX 6000’s bigger sibling. With the same GPU processing unit but double the GPU memory (48 GB GDDR6 ECC). In fact, it is presently the GPU with the highest accessible GPU memory, making it ideal for the most memory-intensive activities.
RTX 3080One of the first GPU models to use the NVIDIA AmpereTM architecture, with improved RT and Tensor Cores and new live multiprocessors. The RTX 3080 has 10 GB of ultrafast GDDR6X memory and 8704 CUDA cores.
RTX 3080 TiThe RTX 3080’s bigger brother, featuring 12 GB of ultra-fast GDDR6X memory and 10240 CUDA cores.
RTX 3090The GeForce RTX 3090 belongs to NVIDIA’s AmpereTM GPU generation’s TITAN class. It is powered by 10496 CUDA cores, 328 Tensor Cores of the third generation, and innovative streaming multiprocessors. It, like the Titan RTX, has 24 GB of GDDR6X memory.
NVIDIA RTX A6000The NVIDIA RTX A6000 is the Quadro RTX 6000’s Ampere-based update. It has the same GPU processor (GA-102) as the RTX 3090, however it supports all CPU cores. As a result, there are 10752 CUDA cores and 336 third-generation Tensor Cores. Furthermore, it has twice as much GPU memory as an RTX 3090: 48GB GDDR6 ECC.