American semiconductor giant, NVIDIA has announced that it is working on brand new next-generation GPU architecture. Dubbed as the Turing GPU architecture, the much-awaited technology was unveiled by the tech giant’s CEO Jensen Huang at the company’s SIGGRAPH 2018, as a part of a keynote presentation.
The Turing architecture:
The next generation architecture has incorporated a slew of new features and it is also expected that it will be available for mainstream production later this year. Although most of the information unveiled about the architecture revolved around the Professional Visualization (ProViz) there wasn’t much information released by the company regarding products made using this architecture.
Despite this, it is most likely that the company will be launching products using this architecture in the near future. The company also went on to note that the presentation doesn’t reveal all the features of the Turing architecture and there may be some changes or additions to the final launch of the technology.
The architecture brings a slew of features and capabilities to NVIDIA’s products. The headline feature is definitely the hybrid rendering capability that mixes the traditional rasterization with the ray tracing to highlight the strengths of these dual technologies. In fact, it must be remembered that NVIDIA made an announcement earlier this year on its RTX technology and this new unveiling is nothing but a continuation of the development of the former.
In fact, NVIDIA is all set to add more ray tracing related hardware coupled with the Turing architecture. This is all set to provide a quicker and more efficient ray tracing acceleration.
Nvidia to host a Geforce Gaming Celebration event on August 20-21
RT Core and Tensor:
NVIDIA has implemented a new feature called the RY Core. Although not much information has been shared by NVIDIA regarding this technology, it is assumed that this is likely to be dedicated ray tracing processors. It is expected that these processors will help accelerate both of the ray-triangle intersection checks apart from the BVH (bounding volume hierarchy) manipulation, as they are popular means to store items for ray tracing.
The company also claims that products with the architecture built in can cast an impressive 10 Giga rays per second. This is a whopping 25 times increase from the current unaccelerated Pascal.
Apart from this, the architecture is also carrying over the Tensor Cores from Volta with a significant enhancement being made to these cores. These tensor cores are critical to NVIDIA’s technologies as they help speed up the ray tracing and reduce the required number of rays in an image with the help of Artificial Intelligence. While this may not be a key feature that the company may have presented in the SIGGRAPH keynote presentation, the company complete neural network and Artificial Intelligence is built on tensor cores. NVIDIA also uses AI denoising to help cleaning up an image so that the number of rays required is drastically reduced.
The architecture will also help speed up the workload of various processes in addition to increasing the rendering. Turing tensor cores do support INT 8 as well as INT 4 precisions over and above the FP 16 precision mode that Volta brings. The only element that feels as a let-down is on the hybrid rendering part, as it isn’t as powerful as it was expected to be, especially considering the amount of upgrade the technology is receiving.
When compared to the Pascal, the performance is only expected to improve by six times. The company will also be launching an SDK, called as NVIDIA NGX that will help to integrate image processing along with neural networking, also using methods such as Deep Learning Anti-Aliasing (DLAA).
The post NVIDIA Unveils Turing GPU Architecture; Mainstream Production To Start Next Year appeared first on MySmartPrice.
from MySmartPrice https://ift.tt/2wmWhmf
No comments:
Post a Comment