Ray Ozzie

This is simply incredible. Sambanova just launched a cloud inference platform allowing developers access to this blazing fast AI platform that has the best inference speeds on Llama models in the market. What is unique here is that all models are run with 16 bit precision as well as being extremely cost efficient. Start developing #FastAI ➡ http://cloud.sambanova.ai

48

Bubble Alert: If folks don't realize, one of the best ways for a company to lose value and revenue is delaying new products that spur sales. So what happens when product technical delays plus competition equals drops in revenue? Now due to US government regulation Chinese companies can no longer buy Nvidia Cloud GPUs directly unless they are the clocked down variants. They can still back channel though other Asian distributors. Meanwhile Microsoft, Amazon, Google, Meta/Facebook, Samsung, Huawei, Qualcomm, AMD, and ARM have all been hard at work building chipset competitors to Nvidia Cloud GPUs. So will public shareholders punish Nvidia when revenue goes down due to both the delays to the Blackwell B200 and the active competition where everyone wants to shift to their own chipsets to get away from the massive markup of the over-priced Nvidia Cloud GPU. Remember, Nvidia does Consumer GPUs that range from $100 to $1000, then enterprise workstation GPUs that run $1,000 to $3,000. Then Cloud GPUs that run from $10,000 to $40,000. From $100 to $40,000, the GPUs are not 400x times as expensive to make. Cloud GPUs have massive profit, giving incentive for every other tech giant who can, to build their own. So what happens to the AI Bubble when Nvidia revenue and profits dip and investors don't see magic numbers go up forever? No bubble lasts forever. ----- The Verge: "Nvidia reportedly delays its next AI chip due to a design flaw/ The company’s next big AI chip may not ship in large numbers until next year, reports The Information." Nvidia has reportedly told Microsoft and at least one other cloud provider that its “Blackwell” B200 AI chips will take at least three months longer to produce than was planned, according to The Information. The delay is the result of a design flaw discovered “unusually late in the production process,” according to two unnamed sources. B200 chips are the follow-up to the supremely popular and hard-to-get H100 chips that power vast swaths of the artificial intelligence cloud landscape (and helped make Nvidia one of the most valuable companies in the world). Nvidia expects production of the chip “to ramp in 2H,” according to a statement that Nvidia spokesperson John Rizzo shared with The Verge. “Beyond that, we don’t comment on rumors.” Nvidia is now reportedly working through a fresh set of test runs with chip producer TSMC, and won’t ship large numbers of Blackwell chips until the first quarter. The Information writes that Microsoft, Google, and Meta, have ordered “tens of billions of dollars” worth of the chips. The report comes just months after Nvidia said that “Blackwell-based products will be available from partners” starting in 2024. The new chips are supposed to kick off a new yearly cadence of AI chips from the company as several other tech firms, such as AMD, work to spin up their own AI chip competitors. The Verge: https://lnkd.in/gDMjxusQ #ai #cloudai #cloudgpu #gpucloud #cloud #nvidia

5

🚀 BREAKTHROUGH | Elon Musk’s VERY Aggressive Push in the LLM Race🚀 NVIDIA CEO Jensen Huang just highlighted Musk's incredible feat: building an AI data center in 19 days, using 100,000 chips to power his upcoming Grok LLM – aiming to outpace GPT. Elon isn’t just playing catch-up; he’s charging ahead in the AI space. His ambition is clear: fuel this powerful infrastructure to supercharge all his application companies - everything from Tesla to SpaceX to Twitter (X). This aggressive move positions Musk to redefine AI and make his mark in the LLM race, with Grok set to shake up the landscape soon. https://lnkd.in/eDKgnG5T #AI #xAI #LLM #Grok #ElonMusk #TechRevolution #AIInfrastructure

8

6 Comments

Give a talk about PerfXLM, a high performance large language model inference engine, at RISC-V Summit Europe 2024

1

Is AI 90% marketing and only 10% reality as Linus Torvalds claims? Rather than being a product to solve a specific problem… …is it a product searching for a problem to solve? #ai #storytelling #digitalmarketing #basicsalesstrategy

4

Last week Andreessen Horowitz announced plans to build an AI infratructure cluster. Here’s why they did it. A16Z is accumulating 20,000 NVIDIA HV100 chips. These chips are going for about $30,000 a piece, meaning that the total investment is around $600M. Now why would they invest that much capital into this type of project? Because GPU’s are becoming a valuable commodity in the AI startup ecosystem. And by amassing and operating a cluster of servers themselves, A16Z can attract startups that build LLM’s and need access to the server in exchange for equity. It’s quite the gamble. This means that over the next 5-7 years A16Z needs to generate around $1.5B in value to make a positive return after accounting for the costs of the initial investment, operating and maintaining the servers and the time value of money. I think that over time this AI infrastructure play should pay off for Andreessen Horowitz as we see momuntal increases in productivity from AI models and more use cases for the technology being developed every day but it is a little risky. What do you think? https://lnkd.in/gkNTpmvM #investing #vc #startups #founders #ai #opinion

11

2 Comments

I just witnessed a VIP demo of Blitzy. Mind Blowing 🤯 If you are building enterprise software and want to get 1000X better, you should reach out to Brian/Pardeshi Link Ventures

57

1 Comment

With all the excitement about massive NVIDIA DGX and EGX systems like the upcoming BG200 NVL72 being used to train huge LLMs and MoEs, I thought it would be nice to ground the discussion back to good old HPC. It's been around for ever. You know, supercomputing. Accelerating computing is not new, and yes much of the hockey stick stuff folks are getting excited about has to do with Moore's Law (not inspite of it), advancements in HBM, advanced packaging for closer-to-wafer-scale accelerators (thanks to Cerebras Systems), and accounting for FP8 and FP4 datatype support for Hopper and Blackwell respective from FP16 for Ampere and prior which I call the "hockey stick" factor. At a system level, there are the benefits of water-cooling which seems to be all the rage these days. Yes, bigger chips mean denser systems, also thermal problems and huge energy draws. I outline this stuff or at least hint at it in my neXt Curve GTC 2024 recap report. Any whoo, for fun I took at peak at TOP500 database to get the latest snapshot in the supercomputing race. Interesting to see that Lenovo tops the list of vendors with the most systems under its belt with 163, and Hewlett Packard Enterprise delivered the two exaflop supercomputers on Earth including the Frontier supercomputer which runs AMD's 3rd Gen Epyx CPUs, and Instinct MI250x GPUs, and the second largest based on Intel Corporation CPUs and accelerators. The third is the Eagle supercomputer by Microsoft based on Nvidia's H100 and Intel CPUs. I think they have five of these beasts that they offer through Azure AI. Just wanted to share because I think we need to be clear on how Moore's Law (transistor density) is one thing and these accelerated computing TFLOP curves are something different in my view. Let me know if you disagree. I just see too many people out there talking about this stuff in an "interesting" way arriving at "interesting" conclusions. I hope this helps. Have a great weekend! Forbes Forbes Technology Council Questex Matt Hamblen neXt Curve #hpc #acceleratedcomputing #supercomputing #generativeai

6

5 Comments

Nvidia CEO Jensen Huang tells Stanford kids about "ample doses of pain & suffering." Interesting take. Here's my 1 and a half cents: 1/ High Expectations = Low Grit: Entitled = brittle. Makes sense. Success requires resilience, the ability to bounce back from punches. Can't build that on a bed of marshmallows. 2/ Easy Wins = Shallow Gains: Struggle builds character. Getting everything handed to you creates...well, people who expect things to be handed to them. Not exactly a recipe for startup success. - But is suffering the only path to the top? 1/ Alternative Routes: Plenty of founders are just, you know, talented and hardworking. They learn from mentors, avoid dumb mistakes, and grind like crazy. No need for a character-building crisis (although, life throws those your way eventually). 2/ Mental Health Matters: Let's not glorify suffering. Building a business is a marathon, not a torture test. Finding joy (and maybe even therapy) can fuel a more sustainable journey. The Bottom Line: Hardship can be a teacher, but it's not the only one. Focus on resilience, value your effort, and prioritize your mental health. There are multiple paths to greatness, some with less collateral damage. - What are your best startup "suffering" stories (or how you avoided it)? #startuplife #resilience #founders #mentalhealth

26

14 Comments

Nvidia CEO Jensen Huang: Every $1 cloud service providers spend with Nvidia translates to $5 worth of GPU rentals. Jensen has come up with a great business strategy: "The more you buy, the more you earn" 😂 #nvidia #artificialintelligence #llms #chatgpt #generativeai #gpu #servers #datacenter #cloud #ai #technology #innovation #business #Semiconductorindustry #electronics #semiconductor #supplychain

41

2 Comments

Recently at SIGGRaph, the host asked an interesting question about Jensen Huang about Nvidia going up stack into software while Hyperscalers going down stack into chips. https://lnkd.in/ecetv8rK 11:27 Jensen gave an technically correct answer about Nvidia always being a software company but he actually dodged the question. You see, those same software insights are already being tackled by Hyperscalers (and AMD) as they not only build the chips but - as a prerequisite - also introduce an open source layer conceptually similar to CUDA so that that layer can organize the context so that the accelerated computing can happen at the chip level. So while what Jensen outwardly said about Nvidia being a software company insofar as CUDA is concerned is true, all things being equal, if he only stays at the CUDA layer and goes no further up stack, Nvidia will eventually stagnate. And he knows that. What we are really seeing is Nvidia going above the CUDA layer into NIMs - i.e specialized LLMs for e.g. Robotics - which have clients from established Robotics companies like Boston Dynamics to who's who in Robotics startups like Figure.ai. We are even seeing Nvidia going higher to the eventual AI app layer - witness OpenAI's DALL-E similar looking app and the Avatar app just to name a few. It's unclear whether Nvidia is using it only as proof of concept to demonstrate the power of its GPUs or as a gateway delving eventually into the App layer. I suspect its more the latter - potentially at least - as Jensen is keeping his options open. The main reason for holding back is that his main customers are the Hyperscalers but he is also well aware they eventually want to eat his lunch. Inside a fortune cookie: Avoid Strength, Attack Weakness: Striking Where the Enemy is Most Vulnerable... By holding out advantages to him, he can cause the enemy to approach of his own accord - Sun Tzu 😉

4

3 Comments

AI that can use computers like humans? Anthropic just made it a reality. Anthropic's latest announcement is reshaping the AI landscape: 1. Upgraded Claude 3.5 Sonnet with industry-leading software engineering skills 2. New Claude 3.5 Haiku model matching top-tier performance at lower cost and speed 3. Groundbreaking "computer use" capability allowing AI to navigate interfaces like humans 4. Significant improvements in coding, with Claude 3.5 Sonnet outperforming all publicly available models on SWE-bench Verified 5. Early adoption by tech giants for automating complex, multi-step processes This leap forward in AI capabilities opens up exciting possibilities for automation, software development, and open-ended research tasks. However, it also raises important questions about responsible deployment and potential risks. What are your thoughts on the implications of AI systems that can use computers like humans? How might this change the future of work and technology? #ArtificialIntelligence #TechInnovation #FutureOfWork #AIEthics #SoftwareEngineering Share your insights in the comments!

5

2 Comments

Seeing stellar performance with HEAVY.AI HeavyDB on the Star Schema Benchmark (1000 GB), running on the NVIDIA GH200 Grace Hopper superchip and an 8XA100 GPU instance on Vultr Cloud. It takes only ~1.5 seconds to run all 13 queries, 50X+ faster than other data warehouses for the price.

65

1 Comment

There's a large HPC audience on here - and everyone talks about CPU vs GPU, and the large bank of legacy codes that don't fit to GPU acceleration (or you can't take the time to adapt them). Elad Raz and NextSilicon just lifted the lid on the third way - and it's a Maverick solution. Bring your standard C++/Fortran/MPI workflows, and end up with better-than-GPU acceleration and efficiency without having to port. While everyone's talking about ML, HPC still makes the world go round. Can't wait to see this in person at Supercomputing. https://lnkd.in/esuUyu2y

106

7 Comments

Did you know that this company that has surpassed Apple was once near bankruptcy? The story of NVIDIA's ascent is one for the ages: 🔎 In 1993, Nvidia was founded by dreamers Jensen Huang, Curtis Priem, and Chris Malachowsky. Their quixotic quest? Crafting processors to supercharge video games. But their 1995 chip NV1 was a brutal reality check - a flop that nearly bankrupted them. 🔎 Salvation arrived in 1997 with NV3. Outperforming rivals, it landed Nvidia deals with PC titans Dell Technologies and Gateway. They'd been granted a seat at the 3D graphics table. 🔎 1999's GeForce 256 GPU was Nvidia's big bang. Bringing massive parallel processing to PCs, it revolutionized gaming visuals overnight. The world took notice of these graphics gurus.  🔎 Nvidia charted its own path, unconventionally championing gaming as the next cultural wave before it crested. Jensen Huang's foresight cemented their iconic brand. 🔎 Restlessly innovating, they expanded GPUs to mobile, automotive, and kickstarted the AI/deep learning era. Nvidia emerged as a multidimensional computing force. 🔎 Today's climax? Nvidia's stock skyrocketed past $AAPL to become the 2nd largest $SPX component at $1164 - an all-time high with a 1:10 split looming. The former underdog has outgrown the giants. 🔎 From near-bankruptcy to industry dominance, NVIDIA's rise is a GOAT tale of perseverance, vision, and constant reinvention. They dreamed of changing graphics computing forever, and delivered that in spades.

11

🧠🔬 We talk a lot about non-invasive neural interfaces, but let's take a trip down memory lane and look at the OG invasive BCIs that paved the way for what we're doing today! 👀 Ever wondered how a tiny chip in your brain could let you control a robotic arm? Or how some folks are playing video games... without using their hands? 🚀 Our team just dropped an in-depth blog post diving into the wild world of invasive Brain-Computer Interfaces. We're talking Neuralink, Blackrock Neurotech, Synchron - the whole gang! 💡 Highlights include: - Mind-blowing stories of BCI users regaining independence - The latest breakthroughs that sound straight out of a sci-fi movie - A peek into the future of human-machine interaction Whether you're a tech geek, a healthcare pro, or just curious about the future of neurotechnology, this post is a must-read. 🔗 https://bit.ly/46KDstY 🎧🧠 Want to stay in the loop on all things neurotech? Hit subscribe on our newsletter for exclusive updates and insights. Trust me, you don't want to miss what's coming next! What do you think? Is the future of tech... in our heads? Drop your thoughts below! 👇 #Wisear #NeuralInterfaces #Wearables #BCI #BrainImplants #Neuralink #Synchron #Blackrock

22

A new foreigner has arrived in town… #Molmo, an #opensource family of state-of-the-art #multimodal #AI models which outpeform top proprietary rivals including OpenAI’s GPT-4o, Anthropic’s Claude 3.5 Sonnet, and Google’s Gemini 1.5 on several third-party benchmarks. Molmo consists of four main models of different parameter sizes and capabilities: 1. Molmo-72B (72 billion parameters, or settings — the flagship model, based on based on Alibaba Cloud’s Qwen2-72B open source model) 2. Molmo-7B-D (“demo model” based on Alibaba’s Qwen2-7B model) 3. Molmo-7B-O (based on Ai2’s OLMo-7B model) 4. MolmoE-1B (based on OLMoE-1B-7B mixture-of-experts LLM, and which Ai2 says “nearly matches the performance of GPT-4V on both academic benchmarks and user preference.”) molmo.org looks really promising. It's going to be fun!🤩 #innovation #tech

11

1 Comment