Blockade Labs just upgraded its Skybox AI platform with Model 3, enabling creators to generate and download seamless 360-degree environments in 8K resolution. The new model packs 77% more pixels to target Apple Vision Pro headsets, alongside a new remix tool for existing panoramic images.
NVIDIA just rolled out full-stack optimizations for Meta's brand-new Llama 3 across its data center GPUs, edge devices, and PCs. In early benchmarks, a single H200 GPU churned out roughly 3,000 tokens per second on the 70B model, enough to handle 300 users simultaneously. The 8B model is also landing on RTX PCs and Jetson Orin modules.
Intel just unveiled its Gaudi 3 AI accelerator at Vision 2024, claiming it outperforms Nvidia's dominant H100 GPU. Built on TSMC's 5nm process with 128GB of HBM2e memory, Intel says Gaudi 3 runs models like LLaMA up to 1.7x faster and is 2.3x more power-efficient. Air-cooled models roll out in Q3, followed by liquid-cooled units in Q4.
Nvidia just swept all seven categories in the latest MLPerf inference benchmarks, showcasing its new H200 chip and TensorRT-LLM software optimizations. Inference now drives roughly 40% of Nvidia's data center revenue. Intel submitted results for Gaudi2 and 5th Gen Xeon, but AMD once again chose not to submit its MI300.
NVIDIA just unveiled its next-generation Blackwell computing platform at GTC, promising a massive leap forward for trillion-parameter AI models. The new architecture delivers up to 5x better FP4 inference performance than Hopper, alongside the new GB200 Grace Blackwell Superchip and NIM software microservices. Why it matters: Accelerated computing is officially replacing general-purpose silicon, and tech giants are already lining up to build the next wave of generative AI.
Nvidia just launched Nvidia Quantum Cloud, a new microservice enabling researchers to simulate quantum processing units directly on GPU hardware. Built on its CUDA-Q platform, it helps teams test complex algorithms in chemistry, biology, and finance before fault-tolerant quantum hardware matures. The company also unveiled cuPQC to accelerate post-quantum encryption.
Nvidia just introduced Project GR00T, a general-purpose foundation model platform tailored for humanoid robots, alongside a dedicated Jetson Thor computer powered by the Blackwell architecture. Almost every major robotics player—from Boston Dynamics to Figure AI—is already on board, positioning the chip giant as the core brain behind embodied AI.
Nvidia just rolled out its Earth-2 climate digital twin cloud platform at GTC 2024. Powered by generative AI models like CorrDiff and Omniverse APIs, it simulates high-resolution weather at a two-kilometer scale up to 1,000 times faster and 3,000 times more energy-efficiently than traditional numerical models. Early adopters include Taiwan's weather bureau and The Weather Company.
Perplexity is finalizing a new funding round that will push the AI search startup to a $1 billion valuation, according to The Wall Street Journal. That nearly doubles its $520 million valuation from just two months ago, alongside topping $10 million in annual revenue. Investors are clearly eager to back a challenger to Google's search empire.
Microsoft just detailed DirectSR, a new Windows API built with Nvidia, AMD, and Intel to streamline AI super resolution across PC games. Instead of writing custom integrations for DLSS, FSR, and XeSS separately, developers can now support all three through a single code path. It rolls out soon via an Agility SDK preview.
Humanoid robot maker Figure AI just raised $675 million at a $2.6 billion valuation from Big Tech heavyweights, including Jeff Bezos, Nvidia, Microsoft, and OpenAI. Along with the cash, Figure signed a deal with OpenAI to develop next-generation multimodal models for its robots. Why it matters: the tech giants are officially racing to put frontier AI into physical bodies.
Google just released Gemma, a new lightweight family of open models built on the same research as Gemini. Available in 2B and 7B parameter sizes, it runs directly on laptops or desktops, beats rival open models at its scale, and comes with permissive commercial licensing. It gives local and open-source AI developers a massive new alternative to Llama.
Groq just went viral with an AI chip architecture that makes standard cloud GPUs look sluggish. Powered by custom Language Processing Units (LPUs), independent benchmarks clocked Groq hitting 247 tokens per second—over 13 times faster than Microsoft’s cloud baseline. Why it matters: this raw speed bypasses memory bottlenecks, finally making lag-free, real-time voice conversations with AI viable.
NVIDIA just released Chat with RTX, a free tech demo that lets you run customized chatbots completely offline on Windows. By connecting local files and YouTube links to open-source models like Mistral and Llama 2 via TensorRT-LLM, you can query notes and documents without sending data to the cloud. You will need a GeForce RTX 30 Series GPU or higher with at least 8GB of VRAM.
NVIDIA CEO Jensen Huang just told world leaders at Dubai's World Governments Summit that every nation needs to own its own "sovereign AI" to protect local data, culture, and history. Huang urged leaders to build out their own compute infrastructure, arguing that natural language has already turned everyone into a programmer.
OpenAI CEO Sam Altman is reportedly in talks to raise an eye-popping $5 trillion to $7 trillion to overhaul the global semiconductor industry and expand AI infrastructure. Altman has long warned that chip shortages are throttling AI growth, and he's now pitching sovereign wealth funds like the UAE to help challenge Nvidia's dominance. Why it matters: If realized, it is an unprecedented moonshot that would fundamentally reshape global supply chains.
Mark Zuckerberg just confirmed that Meta is actively training Llama 3, aiming for state-of-the-art coding and reasoning capabilities to pursue AGI. To power the push, Meta is stockpiling over 340,000 Nvidia H100 GPUs by the end of 2024, building a computing fortress that rivals almost any competitor in the industry.
AMD just launched its Instinct MI300X accelerator and MI300A APU to challenge Nvidia’s dominance in large language model training and inference. The company claims the MI300X beats Nvidia’s H100 by 1.4x in Llama 2 inference, with Microsoft and Meta already lining up to deploy the hardware in cloud and data centers.
Nvidia CEO Jensen Huang just predicted that artificial general intelligence could arrive within the next five years. Speaking at the DealBook Summit, Huang defined AGI as AI that can match normal human intelligence on standard tests, revealing that Nvidia's H-100 chips already require AI to design, though models still struggle with multi-step reasoning.
Nvidia just launched its generative AI drug discovery platform, BioNeMo, on AWS via Amazon SageMaker and ParallelCluster. Researchers can scale digital biology models like ESM-2 across up to 512 H100 GPUs, cutting training runs from a month down to just a few days. Why it matters: Biotech teams can now fine-tune foundation models directly on their proprietary lab data using standard AWS cloud infrastructure.
