Desktop GPU delivers 130TFLOPS for deep learning research
At the Conference on Neural Information Processing Systems in Montreal, Nvidia introduced the Titan RTX desktop GPU which delivers 130TFLOPS of deep learning performance and 11GRays of ray tracing performance. It is, says Nvidia, the wold’s most powerful desktop GPU and has already been dubbed the T-rex.
Running on new Nvidia Turing architecture which fuses shaders, ray tracing and deep learning, the Titan RTX combines AI, real-time ray-traced graphics, next-gen virtual reality and high performance computing, says the company. It features new RT Cores to accelerate ray tracing, plus new multi-precision Tensor cores for AI training and inferencing.
The 576 multi-precision Turing Tensor cores provide up to 130TFLOPS of deep learning performance while the 72 Turing RT cores deliver up to 11 GigaRays per second of real-time ray-tracing performance.
A 24Gbyte GDDR6 memory with 672Gbytes per second of bandwidth is twice the memory of previous-generation Titan GPUs. The 100Gbytes per second Nvidia NVLink can pair two Titan RTX GPUs to scale memory and compute.
The performance and memory bandwidth facilitate real-time 8K video editing and the VirtualLink port provides the performance and connectivity required by next-generation virtual reality (VR) headsets.
The Titan RTX targets AI researchers and deep learning developers. It transforms the PC into a supercomputer for these groups, with multi-precision Turing Tensor cores for what Nvidia describes as “breakthrough performance from FP32, FP16, INT8 and INT4”, allowing faster training and inference of neural networks. The increased memory capacity along with NVLink allow researchers to experiment with larger neural networks and data sets.
The Titan RTX can also be used to accelerate data analytics with Nvidia Rapids, the open source libraries which integrate with popular data science workflows to speed up machine learning.
Titan RTX will be available later this month in the US and Europe.