Nvidia just announced its next ultra-high-end Titan-series GPU called TITAN RTX, which it variously refers to as ‘The Titan of Turing’ and ‘T-Rex’. Nvidia claimed TITAN RTX, as the world’s most powerful desktop GPU, that provides massive performance for AI research, data science, and creative applications.
Power of GPU is everything when it’s come to training a Machine Learning Model or especially when it is the Deep Learning Model. When you train a Deep Learning Model, Forward Pass and Backward Pass are two main operations include in it, and if your GPU is not enough powerful then your time consumption for this process will be more.
Nivida’s TITAN RTX is offering a massive power that can really boost up the processing speed of a computer. The TITAN RTX, dubbed T-Rex is driven by NVIDIA Turing architecture. Its new card delivers 130 teraflops of deep learning performance and 11 GigaRays of ray-tracing performance.
At the Conference on Neural Information Processing Systems in Montreal, Canada today, Jensen Huang, founder and CEO of Nvidia, repeated his claim that “Turing is Nvidia’s biggest advance in a decade – fusing shaders, ray tracing, and deep learning to reinvent the GPU”.
Huang added, more specifically, that “The introduction of T-Rex puts Turing within reach of millions of the most demanding PC users — developers, scientists, and content creators.”
Using Nvidia’s own metrics, ray tracing performance is up to 11 Gigarays per second compared to 10 on the GeForce RTX 2080 Ti, while deep learning performance throughput is rated at 130 Teraflops compared to 57 Teraflops. The Titan RTX has the same 1350MHz base clock speed as the GeForce RTX 2080 Ti, but a higher boost clock speed of 1770MHz instead of 1635MHz.
Nividia’s TITAN RTX is perfectly handcrafted for ease in applications like AI researchers, data scientists, and deep learning developers. It provides Turing Tensor Cores will deliver breakthrough FP32, FP16, INT8, and INT4 performance, allowing faster training and inference of neural networks.
The TITAN RTX is joined by 72 Turning RT cores to assist with real-time ray tracing. It is comprised of 4,608 programmable CUDA cores (compared to 4352 in a GeForce RTX 2080 Ti) for 3D graphics. The TITAN RTX also comes packed with 24GB of GDDR6 RAM, which is an impressive amount but runs on a 384-bit bus for an effective throughput increase to 672GBps, up from 616GBps.
If we compare TITAN RTX to GeForce RTX 2080 Ti, then it is a somewhat beefier version of GeForce RTX 2080 Ti. It uses the same TU102 GPU but without any of its hardware disabled. The TU102 GPU has approximately 18.6 billion transistors and is fabricated using a 12nm process. Power consumption is rated at 280W. The GPU base/boost clock is 1350/1770MHz.
TITAN RTX is most suitable GPU for RAPIDS, one of the most popular data science workflows and Data scientists will look forward to the accelerates there data analytics project with this new GPU.
It will also be able to handle all the games that its GeForce RTX siblings can, with full support for Nvidia’s much-touted ray tracing and Deep Learning Supersampling (DLSS) algorithm.
The powerful new graphics card is also being touted as a tool for studios that work on games and VR experiences built around ray tracing. In fact, NVIDIA claims that the TITAN RTX will provide enough power and memory bandwidth for real-time 8K video editing.
A listing for the GPU is already live on NVIDIA’s website, but Nvidia said that the Titan RTX will become available in the US and Europe later this month and will be priced at whopping $2,499.
The price in India will be Rs. 2,24,000 including taxes, which is lower than the Rs. 2,81,800 list price of the outgoing Titan V model, which was launched almost exactly a year ago.
The Titan RTX will be an alternative to Nvidia’s pro-grade Quadro RTX GPUs, bridging the workstation and gaming markets.
More in AI :