Tag: Blackwell Architecture

July 24, 2025

Nvidia Showcases RTX 5050, AI, and Gaming with Indian OEMs

Key Takeaways

1. Nvidia has launched the RTX 50 Blackwell series, starting with the GeForce RTX 5050 for laptops and desktops, priced at $249.
2. The RTX 5050 features 2,560 CUDA cores, 80 tensor cores, 20 ray tracing cores, and 8 GB of GDDR6 VRAM, while the laptop version has 8 GB of GDDR7 VRAM.
3. Nvidia’s new GPUs utilize advanced technologies like DLSS 4, multi-frame generation, RTX Neural Shaders, and Reflex 2.
4. The company showcased new hardware from various manufacturers and presented impressive AI and ML tools powered by the RTX Blackwell GPUs.
5. Nvidia hosted a “Future of AI” event in New Delhi, announcing the availability of the RTX 50 series in the region.

Nvidia has finished rolling out its RTX 50 Blackwell series with the launch of the GeForce RTX 5050 for laptops and desktops. The company recently hosted a “Future of AI” event in New Delhi, India, to announce the availability of these and other RTX 50 series graphics cards in the region.

Specifications and Pricing

To summarize, the RTX 5050 comes with a price tag of $249 and serves as the next generation after the RTX 3050. It is built on the GB207 GPU, featuring 2,560 CUDA cores, 80 tensor cores, and 20 ray tracing (RT) cores. The GPU is equipped with 8 GB of GDDR6 VRAM on a 128-bit interface and has a total graphics power (TGP) of 130 W. The laptop variant shares many specs, however, it utilizes 8 GB of GDDR7 VRAM and has a TGP limited to 50 W.

Technical Insights

Nvidia’s Jeff Yen, who is the Director of Technical Marketing for the APAC region, along with John Gillooly, Senior Technical Marketing Manager for APAC South, discussed the technical features of the Blackwell architecture. They highlighted how all RTX 50 GPUs, from the entry-level RTX 5050 to the top-tier RTX 5090 monster, leverage technologies like DLSS 4, multi-frame generation, RTX Neural Shaders, and Reflex 2.

New Hardware and Demos

In addition to a variety of pre-built desktops from manufacturers like Ant PC, The MVP, EliteHubs, and Vishal Peripherals, new laptop designs that just hit the Indian market were also showcased.

We have already tested some of these configurations and are looking forward to reviewing others and more in the near future.

Nvidia also presented some impressive demonstrations, especially focused on AI and ML workflows. The company showcased tools like ChatRTX, AnythingLLM, and AI Workbench, all powered by Tensor cores found in the RTX Blackwell GPUs.

Tags: AI Technologies, Blackwell Architecture, Nvidia RTX 5050
January 7, 2025

NVIDIA Targets Tesla with Thor Chip and 4D Driving Simulator

NVIDIA revealed a brand new automotive chip called Thor during its CES 2025 keynote presentation. This chip is built on the Blackwell platform and boasts 20 times more processing power compared to the current generation of automotive chips. It is designed to aid in virtual simulations for autonomous vehicles.

New Approach to Autonomous Driving

Rather than relying on logging millions of miles like Tesla does with its Full Self-Driving (FSD) feature, NVIDIA is investing in a synthetic world simulator named Cosmos. This innovative tool will complement the three chips that are essential for AI in autonomous vehicles.

The Cosmos platform comprises cutting-edge generative world foundation models, tokenizers, guardrails, and a speedy video processing pipeline. These components aim to enhance the creation of physical AI systems, including autonomous vehicles and robots, according to NVIDIA.

Upgraded AI Processing Power

In addition to the new simulator, NVIDIA has upgraded its AI processing chips with the Thor automotive chip, which is the successor to the Orin line. The Thor chip features the Blackwell architecture, similar to the recently launched GeForce RTX 5070 gaming card, which NVIDIA claims can match the performance of an RTX 4090 at a price of $549.

Thor provides 1,000 TFLOPS of powerful AI computing and offers roughly 20 times the AI capability of its predecessor. Its efficiency is crucial as it helps reduce the overall costs of System on Chip (SoC) in vehicles that integrate both autonomous driving and infotainment systems.

Flexible Architecture for Car Manufacturers

Automakers and developers of autonomous vehicle solutions can allocate the 1,000 TFLOPS of computing power between self-driving tasks and infotainment systems, making the architecture highly adaptable. According to ARM, “When used in level 3 or higher autonomous driving scenarios, a single Nvidia DRIVE AGX Thor can replicate the functionality of multiple advanced devices currently found in vehicles, leading to considerable advantages in performance for car manufacturers.”

NVIDIA also highlighted multiple safety and security certifications for its new Thor-based Drive AGX platform, including recognition from the ANSI National Accreditation Board and TÜV Rheinland.

Partnerships and Innovative Simulation Features

NVIDIA has already partnered with various automakers and autonomous driving solution providers such as Mercedes and Nuro. They are utilizing NVIDIA’s self-driving chips and Cosmos simulation software, which includes features like OmniMap that combines mapping and satellite imagery to create 3D environments that are drivable. The Neural Reconstruction Engine generates high-fidelity 4D simulation environments from AV sensor data. Additionally, Edify 3DS can search for existing scenes or create 3D objects based on text or image prompts, transforming scenes in NVIDIA Omniverse that are grounded in 4D. Finally, Cosmos is capable of generating countless variations of driving scenarios using simple text prompts.

However, it is yet to be determined if the Cosmos simulator will enable traditional automakers and electric vehicle startups to bridge the gap with Tesla’s FSD solution, which relies on extensive real-world driving data to enhance its AI algorithms.

Source: Link

Tags: Blackwell Architecture
August 7, 2024

Nvidia Developing AI GPU with 144GB HBM3E Memory

Nvidia is gearing up for the release of its next-generation B100 and B200 GPUs, which are built on the Blackwell architecture, as reported by TrendForce. These advanced GPUs are slated to launch in the latter half of this year and will cater primarily to Cloud Service Providers (CSPs) for their cloud computing needs. Additionally, Nvidia plans to introduce a simplified version, the B200A, aimed at OEM enterprise customers with edge AI requirements.

Tight Packaging Capacity

TSMC’s CoWoS-L packaging capacity, utilized by the B200 series, remains limited. The B200A will transition to the more straightforward CoWoS-S packaging technology to better meet the needs of Cloud Service Providers.

B200A Technical Specifications

Although the full technical specifications of the B200A are not yet detailed, it is confirmed that the HBM3E memory capacity will be reduced from 192GB to 144GB. Additionally, the number of memory chip layers will be cut from eight to four, although the capacity of each individual chip will increase from 24GB to 36GB.

Power Consumption and Cooling

The B200A will consume less power than the B200 GPUs and will not require liquid cooling, making it easier to install with an air cooling system. This new GPU model is expected to be available to OEM manufacturers by the second quarter of next year.

Supply chain surveys indicate that Nvidia’s primary high-end GPU shipments in 2024 will be based on the Hopper platform, with H100 and H200 targeting the North American market and H20 for the Chinese market. Since the B200A will be released around the second quarter of 2025, it is anticipated not to overlap with the H200, which is scheduled to arrive on or after the third quarter.

(Source: TrendForce)

Tags: Blackwell Architecture