中文English
Blackwell架构芯片ichaiyang 2024-05-20 9:45 39
1, 8 years of 1000x growth, Nvidia brings the most successful product in history 2, Autonomous Driving Computing & Domain control platform -NVIDIA3, When does the 50 series graph...

Blackwell architecture chip (What is the meaning of the chip architecture)

8 years of 1000x growth, Nvidia brings the most successful product in history

Huang Renxun said. In the past eight years, the demand for AI computing power has increased 1,000 times. Under the Blackwell architecture, chips can be connected to build a large AI supercomputing cluster to support greater computing needs. "It's Nvidia's most successful product." Huang further introduced that Blackwell has 208 billion transistors, more than double the 80 billion transistors of the previous generation chip "Hopper", and can support AI models with up to 10 trillion parameters.

The entire training speed in the SFT phase, in the same mode of adaptationAfter the type structure, under the same training set, the entire training process is also shown in the figure, which is more than 3 times the best open source capability in the current industry; RHF training speed in reinforcement learning is about twice that of DeepSpeed; In terms of throughput force, after adapting the same model structure, TFlOPS are also faster than DeepSpeed under the same training set.

At the GTC China Conference in December 2019, Nvidia announced the latest generation of autonomous driving computing SoC Orin. This chip is composed of 1Consisting of 7 billion transistors, it integrates Nvidia's next-generation GPU architecture with Arm Hercules CPU cores and a new deep learning and computer vision accelerator capable of running up to 200 trillion calculations per second. Compared to the previous generation Xavier, the performance has improved by 7 times.

The Hopper architecture's H100 NVL GPU is a stunning debut, with 40Tb/s of amazing I/O bandwidth and 12x performance leap, providing unprecedented support for enterprise applications. 56 core server GPU,A symbol of Nvidia's endless potential in the data center space. Which one you choose depends on your target area and performance needs. Whether it's gaming, design, research or artificial intelligence, Nvidia has the perfect GPU for you to explore.

● History: NVIDIA introduced its first graphics card, the NVIDIA STG-2000X, in 1995, but sales were lukewarm due to incompatibility with system formats. The second product, RIVA128, launched in 1997, focused on compatibility, especially with Microsoft systems, and was on par with Voodoo in performanceHowever, it has stronger compatibility, so it is a big success.

NVIDIA Chief Scientist: Breaking the physical limits of chip technology On December 14, according to media reports, Nvidia chief scientist Bill Dally said before the annual China GPU Technology Conference that the GPU allows the performance of artificial intelligence (AI) to be doubled every year. Nvidia researchers are defining ways to build faster AI chips in systems with higher bandwidth that are easier to program.

Autonomous drivingComputing & Domain Control Platform -NVIDIA

1, R-Car-V3M: Smart Camera of choice Designed for economical smart cameras, R-Car-V3M supports ASIL and deep learning, suitable for driver monitoring, circumnavigation and liDAR, NCAP features, low power consumption and high efficiency. The latest Drive Orin and Drive Atlan show NVIDIA's future direction in autonomous driving computing power, positioning themselves as powerful AI processors and next-generation platforms in 2025, respectively. And for the existing hardwareLike Drive PXDrive CX, Drive Hyperion, and Drive Pegasus, NVIDIA provides detailed specifications and feature descriptions.

3, S32K - Universal MCU/ : S32K as a universal microcontroller, suitable for a variety of automotive electronic systems, providing flexible computing power. S32R - Radar Processing / : The S32R focuses radar signal processing to provide accurate perception for Advanced Driver assistance systems (ADAS). S32Z/E - Real-time Processor / : Designed for real-time controlIt will use TSMC's 3nm process and is expected to be produced in September 2024.

2, series graphics card is expected to be released in 2024. The 50 Series graphics cards are expected to be released in the fourth quarter of 2024, the 5090 in October 2024, the 5080 in November, and the 70 and 60 in 2025. Nvidia-style graphics cards, on the other hand, will be available in 2023. However, it will take about five years for a 50-series graphics notebook to hit the market, meaning it won't be available until around 2027. Finally, the 50 series graphics card will be released in 2024.

3, yearsThere will be a 50-series graphics card. In the latest NVIDIA graphics card boss interview, it is also revealed that 50 series of graphics cards will be released by the end of 2024. According to the current Revelations, it should be Intel and Samsung for foundry. If you are planning to buy a video card, you may wish to wait until 50 series video card out to buy, although the price will be more expensive, but its performance is absolutely guaranteed.

4, series of graphics cards may be out in 2023, but the specific release time is still subject to official news. Since last year has launched a part of the 40 series of graphics cards, so sayThis year, the remaining 40 series graphics cards will be released, as well as some 50 graphics cards. The series graphics notebook should take about five years to market, that is, it will not be available until around 2027. The current 30 series has just been released.

Will the 50 series graphics card size be larger

Process improvements: The RTX50 series uses a more advanced 3nm process, compared to the 5nm process for the 40 Series. This means that 50 Series graphics cards have a smaller transistor size and a higher clock rateAnd lower power consumption. CUDA Core Additions: The RTX50 series graphics cards are equipped with more CUDA cores, the core unit that performs graphics rendering in Nvidia graphics cards.

Generally speaking, 50 series graphics cards will be more expensive than 40 series graphics cards. This is because 50 series graphics cards are more powerful than 40 series graphics in terms of performance, with more video memory and higher core frequencies, and therefore more expensive. Of course, different brands and models of graphics card prices will vary. Overall, if you need higher graphics performance, then the 50 series graphics card may be a better choice, but it needs to payA higher price.

Recently came the latest news of the rtx50 series graphics card, it is said that the structure of the 50 series graphics card will change greatly, and the performance will also improve a large section, I hope we can look forward to it. How about the 50 series graphics card: Very good. Recently, foreign news about the rtx50 series graphics card, it is said that the structure of the 50 series graphics card will be greatly changed, which is the largest performance improvement in the history of NVIDIA.

50 series of graphics cards will be released this year. In the latest NVIDIA system graphics card boss interview Revelations, there are also explosionsIt is expected that 50 series graphics cards will be released by the end of 2024. According to the current Revelations, it should be Intel and Samsung for foundry. If you are planning to buy a video card, you may wish to wait until 50 series video card out to buy, although the price will be more expensive, but its performance is absolutely guaranteed.

No. According to relevant public information, the performance and power consumption of the 50 series mobile graphics card are limited, and there is a certain performance gap between the desktop graphics card, and it is impossible to narrow this gap. The 50 series mobile graphics card uses the NVIDIAGeForceGTX105Laptop graphics cards with 0, 1050Ti, 1060, 1070, 1080 chips.