Nvidia’s DRIVE AGX Orin platform with 12 Hercules cores has 17 billion transistors
Nvidia has just announced its latest platform for self-driving cars called DRIVE AGX Orin. The said platform owes its power to an SoC (system on a chip) processor called Orin. This chip has the latest generation architecture of Nvidia graphics processors and its processing cores are of the Hercules ARM architecture. are A large number of functional features based on artificial intelligence are also included in this processor is included. In fact, the target market of Orin’s DRIVE AGX platform is self-driving cars and robots, and this computing system should have much better performance and higher efficiency than its previous generation, DRIVE Pegasus.
This platform, which is made of the new generation Orin processor, will replace the Drive PX Pegasus platform. The Drive PX Pegasus system consists of several Xavier processors and is based on the Turing architecture and performs up to 320 tera operations (TOPs) at 500 watts. Each of Orin’s processor cores, with an impressive number of 17 billion transistors, is supposed to perform 200 terabytes of operations in one second and be 7 times more efficient than Xavier’s SoCs. For comparison, it is necessary to mention that the largest consumer-grade graphics cards currently being manufactured, Volta GV100 and Turing TU102, have only 21.1 and 18.6 billion transistors, respectively.
- Nvidia showed its advances at the GTC conference in the field of artificial intelligence
- Nvidia: with the introduction of seven-nanometer graphics processors , to everyone’s surprise
Nvidia did not specifically mention the type of new GPU in the Orin chip, but the chip maker’s Ampere series graphics cards from next year will be launched on the market. The lithography of the mentioned graphics processor is below 12 nm and the architectural upgrade in this product is bigger than what we saw in the evolution of Volta to Turing. Also, the Xavier processor was the first member of the Nvidia chip family that was introduced under the Volta graphics architecture. So it is not surprising that Orin, which is the first chip with Ampere architecture, replaces this product.
Sam Abolsamid, senior research issues analyst at Navigant Research Institute, also says: /p>
Nvidia’s long-term commitment to the transportation industry combined with innovative tools and platforms coupled with continuous development has resulted in a vast ecosystem. Almost every company working on self-driving cars has used Nvidia [chips] for their computing system. Orin seems to be an impressive step forward that could help usher in a big new chapter in the story of this growing technology. , several different configurations of the Orion-named chip are listed on the Drive platform. Orion seems to have several optimized variants in workload and thermal design power, including the basic version with one camera and the ability of 36 tera operations per second at 15 watts of power, a version with 4 cameras and the ability of 100 tera operations per second at 40 watts of power. There is a version with two chips and a capacity of 400 terabytes of operations per second at a power of 130 watts, and a flagship version with two Orion chips and two separate graphics cards and a capacity of 2000 terabytes of operations per second at a power of 750 watts. It seems that the thermal design power in the version that has two hundred TOPs is around 60 to 70 watts.
A next-generation GPU that appears to have the same technology as the Orion processor was also showcased at the GTC 2019 event. The aforementioned GPU performs up to 800 terabytes of operations (TOPs) at 310 watts. The mentioned figures are the result of the fact that two chips from the Orion processor have performed 400 TOPs at a thermal design power of 130 watts and are placed above the DRIVE platform with a power of 620 watts. The thermal design power of 310 watts per graphics card means that these products perform 800 terabytes of operations per second when executing the Int8 (eight-bit) instruction set. Currently, the Quadro RTX 8000 graphics card, which is the flagship of the TU102 series, performs a total of 206.1 TOPs of operations at a thermal design power of 295 watts. This alone represents a 3.5x performance jump over the Turing graphics architecture and defines exactly what to expect from Nvidia’s next generation graphics cards. The annual GPU Technology Conference (GTC) is one of the biggest events in the field of graphics cards, processors and artificial intelligence.
Nvidia stated that Orin’s DRIVE AGX platform Level 2 to 5 autonomous vehicles are compatible and extensible. The aforementioned computing system is designed to meet security system standards such as ISO 26262 ASIL-D while running various applications that are based on artificial intelligence or deep neural networks (DNN). Just like its predecessor Xavier, the Orin processor is supposed to be used in cars that will be produced in 2022, that is, two years after the unveiling of this chip. As a result, the construction of this platform should start earlier than the mentioned time and probably from 2021.
Comparison of different generations of Nvidia’s Drive PX platform
Soc Name product name Nvidia Drive PX Nvidia Drive PX Nvidia Drive Xavier Nvidia Drive Pegasus Nvidia Drive AGX Orin Tegra X1 Parker Xavier Xavier Orin Lithography 20 nm 16 nm FinFET 12 nm FinFET 12 nm FinFET 12 nm FinFET Unspecified Number of transistors on SoC 2 billion unknown 7 billion on each Xavier chip 7 billion on each Xavier chip 17 billion on each Orin chip Graphic architecture Maxwell with 256 cores Pascal with 256 cores Volta with 512 cores Volta with 512 cores Ampere? Processor 16-core ARM 12-core ARM 8-core ARM 16-core ARM 12-core ARM processor architecture
8 Cortex A57 cores
8 Cortex A53 cores
4 Denver core
8 Cortex A57 cores
8 64-bit Carmel ARM cores have 8 MB of level 2 cache memory and 4 MB of level 3 cache
8 64-bit cores Carmel ARM has 8 megabytes of level 2 cache and 4 megabytes of level 3 cache /td>
Tera operations per second, based on Deep Learning (DLTOPs) unknown 20 DLTOPs
align:center”>320 TOPs 200 TOPs Number of chips 2 Tegra X1 chips
2 Tegra X1 chips 2 Pascal MXM graphics processors 1 Xavier chip -align:center”>
2 Volta chips
2 Turing chips
1 Ampere chip System Memory LPDDR4
8 GB LPDDR4
16 GB 256-bit LPDDR4 memory
LPDDR4 and GDDR6
unknown Recorder Graphics card Unspecified
4 GB GDDR5
200 GB/s Thermal Design Power (TDP) 20 watts
Xavier processor was also unveiled in 2016, but its use in vehicles since the end of 2017 and It started in early 2018. With all the above mentioned, Volta’s Drive PX Pegasus platform graphics processor was launched in the form of Titan V graphics card one year after the platform was unveiled. The mentioned time frames can provide users with a correct understanding of when the next generation of Nvidia graphics cards under the Ampere architecture will be unveiled. As we get closer to CES 2020, we will hear more details from Nvidia about the company’s new architecture and product roadmap.