Dave Lister, Principal Database Consultant
Yesterday NVIDIA announced its next generation of video cards, the GeForce RTX 30 series. Based upon the gaming and graphics variant of NVIDIA’s Ampere architecture and built on an optimised version of Samsung’s 8nm process, NVIDIA is promoting the new cards as delivering some of their greatest gains ever in graphics performance.
In addition, the latest generation of GeForce will also be coming with some new features to further set the cards apart from and ahead of NVIDIA’s older Turing-based RTX 20 series.
Starting September 17th, NVIDIA and their partners (such as ASUS) will release the GeForce RTX 3090, GeForce RTX 3080, and GeForce RTX 3070 (see table below for specs).
The RTX 3090 is akin to the RTX 20 Series TITAN
NVIDIA have indicated that the performance gains for the RTX 30 series may be around 30% compared to their equivalents in the RTX 20 series. Worthy of note, the RTX 30 series supports PCI Express 4.0 (as well as 3.0), although current graphics cards are not able to saturate the full potential of PCI Express 3.0. Only AMD‘s third-generation Ryzen and Thread ripper processors can take advantage of PCI Express 4.0 potential as Intel has yet to deliver processors that can support that level of connectivity.
NVIDIA GPUs are the leading computational engines powering the AI revolution, providing tremendous speedups for AI training and inference workloads. In addition, NVIDIA GPUs accelerate many types of high performance computing and data analytics applications and systems.
Data science is one of the fastest growing fields of computer science and impacts every industry while data science problems involve data on a massive scale and require large-scale processing capabilities.
NVIDIA are focused on advancing GPU technologies that can be used by scientists to wrangle, prep, train and deploy models quickly and accurately.
Specifically, the new GA102 Ampere GPU that the RTX 3080 is based on delivers 30 shader TFLOPs and does two shader calculations per clock, versus 1 per clock in the previous generation. Ampere’s second gen RT cores also deliver roughly 2x the ray triangle intersection throughput of Turing, with 58 RT TFLOPS (versus 34 RT-TFLOPS for Turing) for double the ray tracing performance, and 238 Tensor TFLOPS (versus 89 in Turing) for more than double its machine learning throughput. Ampere’s 3rd gen Tensor cores will power NVIDIA’s DLSS 2.0 (Deep Learning Super Sampling) image processing engine for better graphics fidelity.
The peak available throughput is more likely to be 2.7X in some cases (Shader and Tensor ops), or a little less than 2x (1.7x) in the case of ray tracing. The new GA102 Ampere GPU in itself, with all of this additional throughput and capability is massive, comprising some 28 billion transistors for a GA102 chip, versus 18.5 billion in Turing’s TU102.
Another area that NVIDIA has caught some flak in past releases, is their pricing model. However, this time around, NVidia’s price to performance strategy seems on point and sharper, the fact that a GeForce RTX 2080 Ti still retails for USD$1200 - $1400, whereas the new GeForce RTX 3080 at $699 is a highly attractive offer.
For more information regarding the new NVIDIA A100 GPU based on the new NVIDIA Ampere GPU architecture, refer to:
https://developer.nvidia.com/blog/nvidia-ampere-architecture-in-depth/