AI Zona’s Post

A new Blackwell GPU, GB200 has been introduced which offers upto 20 petaflops of FP4 horsepower. It is the first TEE-I/O capable GPU that helps in protecting sensitive data from unauthorised access in AI models. Its transformer engine uses Blackwell Tensor Core along with TensorRT-LLM for accelerating the inference and training for LLMs and Mixture-of-Experts (MOE) models. The Decompression Engine used helps in accelerating the full pipeline of database queries for the highest performance in data analytics and data science. #nvidia #gpu #gpucomputing #transformermodels #innovation #innovationinengineering

LLM breakthrough: $0.1 per 1 million tokens. The beauty of Neural Architecture Search is that no matter what open source model comes next, NAS will consistently produce 2-3X more efficient architectures. On top of any breakthrough or bag of tricks, NAS is here to stay. Efficiency is the most important metric. Go Deci!

Whatever power it may have, a solution must be able to solve the problems. Is there any of the artificial Intelligence (AI) products you know can answer the following questions, with the power of such technology? "Who, in England of UK, has new US patents granted on the nearest Tuesday, when the USPTO releases the newly granted US patents on a weekly basis?" "Who, in the "江蘇‘’ province of China, has new US patents granted on the nearest Tuesday, when the USPTO releases the newly granted US patents on a weekly basis?" With our intellectual property (IP), we can. Do you or any of your contacts need our expertise and our IP? We're selling, NOT just talking. Thanks. Here is our pitch deck: "Experiment results showed that with our intellectual property (IP), a copyrighted multilingual metadata, we are doing what AI, like ChatGPT, can't do in data analytics. Without metadata, NO data can be found/retrieved, even by the most advanced technologies, like AI, supercomputers, etc. https://lnkd.in/g-aJFnXR Our IP can also make your information service UNIQUE in the world." P.S. We wrote this post ourselves, without using any AI tool.

This is will revolutionise AI eco system. Enterprises will have to adopt Chips to enable AI capability to the fullest..

Very interesting! (🧙♂️*🤖)*NVIDIA=🚀😀!! El regalo perfecto para el Día del Padre !! y el Día de la Madre (en mayo) !!  Introducing Blackwell. It’s not a chip; it’s a platform. —Jensen Huang, CEO, NVIDIA Blackwell platform for trillion-parameter scale generative AI.  Main features: - AI Superchip, 208B transistors - 2nd Gen Transformer Engine: FP4/FP6 Tensor Core - 5th Generation NVLink: Scales to 576 GPUs - RAS Engine: 100% In-System Self-Test - Secure AI: Full Performance Encryption - Decompression Engine: 800GB/sec The most powerful GPU in the market. Exciting announcement, live from GTC. -- Os dejo un link a una noticia muy interesante relacionada: https://lnkd.in/dyp7Hrn6

Everyone talks about data. Data is the new oil. However, no one talks about how to compute that data. What Nvidia focused on, its computation unlike other competitors, is really commendable. Not only in hardware, but they're also expanding vertically by creating their own open-source frameworks and toolkits in fields such as ASR, TTS, NLP, Vision, and generative AI, which one can use to train their customized models using their GPUs. Additionally, they're collaborating with industries to increase productivity. No surprise that its shares have given more than 100% return in just 6 months, over 250% return in a year, and more than 1900% return in 5 years. Interestingly the return of Bitcoin in one year is 130%, and in 5 years, it's more than 1700%. It's no surprise that as crypto transactions increase, more mining is done, and more GPUs are required, ultimately increasing their value. If you see, the return of Nvidia and Bitcoin almost go hand in hand. On a lighter note, if you fear crypto investment but are fascinated by its returns, if you're unsure which company will lead in AI in the end, you can at least confidently bet on Nvidia, as all these things require a GPU at the very least.

🔥What a remarkable achievement for NVIDIA - 𝗧𝗵𝗲 𝗻𝗲𝘄 𝗕𝗹𝗮𝗰𝗸𝘄𝗲𝗹𝗹 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺 𝗮𝗻𝗱 𝗚𝗕𝟮𝟬𝟬: • 1000X AI compute in 8 years. • First machine exceeding 1 exaFLOP in one rack. • LLM inference improved by up to 30 times comparing with the Hopper (previous generation). • Training a 1.8 trillion parameter model would have previously taken 8,000 Hopper GPUs and 15 megawatts of power, while today 2,000 Blackwell GPUs can do it while consuming just 4 megawatts for the same two months period. 📌 𝙼𝚘𝚜𝚝 𝚖𝚎𝚖𝚘𝚛𝚊𝚋𝚕𝚎 𝚖𝚘𝚖𝚎𝚗𝚝: Jensen: "This is what makes it possible. That's the back of DGX (supercomputer in a rack). 130tb/sec goes to the back of that chassis. 𝗧𝗵𝗮𝘁 𝗶𝘀 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝘁𝗵𝗲 𝗮𝗴𝗴𝗿𝗲𝗴𝗮𝘁𝗲𝗱 𝗯𝗮𝗻𝗱𝘄𝗶𝗱𝘁𝗵 𝗼𝗳 𝘁𝗵𝗲 𝗶𝗻𝘁𝗲𝗿𝗻𝗲𝘁." Audience: ... Audience: *𝘍𝘪𝘯𝘢𝘭𝘭𝘺 𝘳𝘦𝘢𝘭𝘪𝘻𝘦𝘴* Starts clapping. ❓Considering this breakthrough, how do you foresee the future direction of #generativeai?

WOW - My team at Deci AI keeps breaking records! I'm thrilled to introduce Deci’s Generative AI Platform featuring a new series of proprietary LLMs, an inference engine, and an AI inference cluster management solution. 🔍 Here a key things to know about Deci-Nano: 🏆 With an average Open LLM  Leaderboard score of 70.76  & a 7.41 MT-Bench score, Deci-Nano outperforms Mistral-7b-instruct-v0.2 Gemma-7b-it ⚡Ultra Fast inference speed -  Deci is off-the charts with E2E latency (256 tokens) coming at 4.56 seconds 🤯  measured on NVIDIA A100 GPUs. 💰 Exceptional price: $0.1 per 1 Million tokens. Deploy our models via API, or own your infrastructure to keep your data in-house. You can try it now with a free trial: 💻 Try Deci-Nano in our playground or get your API trial token> https://lnkd.in/dgaV5bhJ 📚 Read the blog > https://hubs.ly/Q02pspZM0 📘 Check out the QuickStart notebook > bit.ly/Deci-Nanonotebook #genai #llm #inference #deployment

NVIDIA GB200 Grace Blackwell Superchip: Tailored for the generative AI era. - Core Components: Combines two NVIDIA B200 Tensor Core GPUs with an NVIDIA Grace CPU. - Interconnect Technology: Utilizes a 900GB/s NVLink for high-speed CPU-GPU communication. - Manufacturing Process: Employs a TSMC 4NP custom process. - Transistor Count: Boasts 208 billion transistors, indicating advanced computational capability. - Chip Design: Features a dual-die design connected by a 10TB/s chip-to-chip interconnect. - NVLink Switch: Facilitates the connection of four NVLinks at 1.8TB/s for scalable AI processing. - Strategic Focus: Aims at setting new benchmarks in scalable, efficient AI computing infrastructure.

From NVIDIA #gtc2024 three highlights for me so far - Blackwell The new GPU, Blackwell, is 4X faster than the previous generation of GPUs and 25X more energy efficient. https://lnkd.in/dRkNyvTk - GR00T GR00T is a general-purpose foundation model that promises to transform humanoid robot learning in simulation and the real world. https://lnkd.in/dgVyEHmp - NIMS NIM microservices simplify the AI model deployment process by packaging algorithmic, system, and runtime optimizations and adding industry-standard APIs. This enables developers to integrate NIM into their existing applications and infrastructure without extensive customization or specialized expertise. https://lnkd.in/dM4bM_aE Videa Credit: Steve Nouri _____________________________________ Follow me for regular updates and insights on Generative AI and my journey as a CTO in AI product development.