- Synthetic
- Posts
- AI chip wars accelerate as Nvidia, AMD, and Intel battle it out
AI chip wars accelerate as Nvidia, AMD, and Intel battle it out
Plus, Anthropic chief expects most work to end by 2029
Steve Brown
June 06, 2024
In partnership with
This Week in AI
Talk to anybody working in the AI safety field trying to help researchers build responsible AI. One of their major fears will be a ‘race condition’ where rivals cut safety corners and cause significant harm in the rush to build powerful AGI first. OpenAI insiders are blowing the whistle on a culture of recklessness and secrecy that they believe may lead to its AI systems becoming dangerous. They also complain about non-disparagement agreements employees are asked to sign when they leave or forfeit high-value vested stock options. In an open letter, a group of former employees call for whistleblower protections and for OpenAI and other research labs to embrace external scrutiny and adopt transparent governance. Interestingly, AI luminaries Yoshua Bengio, Geoff Hinton, and Stuart Russell endorse the letter. Google declined to comment, though Synthetic’s experience, Google DeepMind takes a far more responsible approach to AI safety and invites external scrutiny of its research and products through a safety council.
Nvidia has become the second-most valuable company in the world, leaping ahead of Apple with a $3 trillion valuation, despite the U.S. Federal Trade Commission and Department of Justice opening antitrust investigations into OpenAI, Microsoft, and Nvidia.
At Computex earlier in the week, CEO Jensen Huang revealed plans for another leap in AI performance in 2026 with the “Rubin” platform, signaling an annual innovation cadence from now on. He also emphasized Nvidia’s interest in making AI more physical in the form of advanced robotics. Nvidia’s announcements include:
Rubin - Named after astronomer Vera Rubin, this next-generation AI platform will launch in 2026. It will use new HBM4 memory and feature a super-high-speed NVLink 6 interconnect running at 1.8 terabits per second. Rubin Ultra will follow in 2027.
Project G-Assist - Originally an April Fool’s Day hoax in 2017, this has now become a real Nvidia product. It’s a chatbot that guides you through video games with game-specific, context-aware tips and tricks. It also tunes and optimizes the operation of your PC to maximize your gaming experience.
NIMs - Building on the information revealed at GTC earlier in the year, Huang shared more details on Nvidia Inference Microservices (NIMs) that package together everything developers need to run a model in optimized containers that can be easily deployed in the cloud, data centers, or workstations. Over 40 NIMs are already available, covering areas including text, image, and video generation, digital biology, speech, and the generation of life-like human avatars. Nvidia has even packaged Meta’s LLaMA-3 model inside a NIM and made it available for free on Hugging Face.
Synthetic’s Perspective: NIMs scream proprietary lock-in, but they will make life much easier for any developer who plans to always live inside an Nvidia world and pay Nvidia prices.
Not everyone was thrilled with what they heard from Nvidia. One commentator laments the incredible power consumption of Nvidia chips and the human cost of the company’s focus on robots and synthetic humans in a piece titled I watched Nvidia’s Computex keynote and it made my blood run cold.
In 2021, in a pre-ChatGPT world, OpenAI pulled the plug on its robotics research team, citing a lack of relevance to its core mission. Recent events have convinced OpenAI to reverse that decision and bring its robotics research efforts back to life. In just the last 18 months, a broad range of humanoid robots have emerged. Nvidia has invested heavily in robotics, and OpenAI’s collaboration with leading robot company Figure turned many heads.
Video: Mr Bean explains AI image generation
Ever wonder how an AI can generate an image of nearly anything from a simple text description? Most use a technique known as diffusion, which starts with a random image (essentially white noise) and then strips away the noise to ‘reveal’ an image beneath. The AI is trained on images that have had noise progressively added over time and learns to run that process in reverse. This video, produced by the same people who brought you Ryan Gosling explains LLMs a few weeks back, shows (a deepfake of) Mr Bean explaining the process in detail. Enjoy!
AI Tech and Innovation
As part of his ongoing push to restore Intel to a market leadership position, CEO Pat Gelsinger announced new Xeon 6 chips and “Lunar Lake,” a highly integrated chip to build AI PCs. He also revealed pricing for the Intel Gaudi 3 chip that competes against Nvidia’s H100. It will cost 2/3 of the price. Intel revealed Xeon 5 only six months ago, demonstrating their rapid driving pace. Lunar Lake will feature integrated memory (similar to Apple’s M-series processors), all-day battery life, and be the first manufactured on Intel’s leading-edge 18A process. If you haven’t upgraded your laptop in a while, Lunar Lake could be the trigger you’ve been waiting for.
In a shot across the bows of Intel and AMD, Nvidia is rumored to have developed a chip for PC makers to build an AI PC (or Copilot+ PC if you prefer the Microsoft branding.) Built around an ARM core blended with their latest Blackwell GPU architecture, the product could offer formidable competition to traditional PC chip makers. Microsoft’s latest Surface Laptop and pro tablets are built with Qualcomm's ARM-based AI PC chip.
Due to launch later this month, Microsoft’s Recall feature for Copilot+ PCs is designed to make searching and retrieving information on your local PC easy. To perform its function, Recall takes a screenshot every five seconds. While all information is stored locally and never used to train Microsoft’s AIs, security specialists have noted that information is unencrypted and stored locally in plain text, making it a hacker’s dream. One ethical hacker did just that to prove the point.
Researchers at the University of Chinese Academy of Sciences have outlined a proposed concept to improve AI decision-making. Foundation agents are generally capable across physical and virtual space and designed to handle open-ended decision tasks using an LLM’s understanding of the world.
AI Insights
“The economically and politically relevant comparison on most tasks is not whether the language model is better than the best human; it is whether they are better than the human who would otherwise do that task.”
A senior worker for a frontier AI company considers the implications of a world without work, which they anticipate might come within the next few years. Assuming that economic and material needs are met with some universal basic income, will we be happy without work? If AI brings forced retirement to most workers before the end of the decade, how will you feel? What will you do to fill your days? How will we create meaning in our lives? Are we ready for such dramatic change?
In an effort to reduce costly returns, Amazon will use computer vision to find defects in shipments before they are dispatched from North American fulfillment centers. Known as Project P.I., the AI will scan millions of packages each day, looking for defective products and checking products are the correct color and size. Amazon hopes to significantly reduce unwanted returns, wasted packaging, and carbon emissions. Early trials have been highly successful, and Amazon plans to widen deployment in 2024.
Synthetic’s take on the urgent need to reimagine education is simple: if we were to build our education system from scratch today to meet the needs of both today and tomorrow, it would look very different than it does now. Securing a quality education comes with crippling costs that burden students for decades; the pedagogy is little changed from that used in the 18th and 19th centuries, and students are not being prepared to thrive in the workplaces of the future. Artificial intelligence can help. And if traditional educators fail to take action and fundamentally transform themselves, they will be disrupted by well-funded and highly motivated players.
Toolkit for the Future
An entirely new way to present ideas
Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.
Need to expand internationally but unsure where to start? Oyster makes it easy to find, hire, and retain local talent. They handle the details (country-specific labor laws, international tax laws, compliance, and global payroll) so you can focus on finding the right talent and ramping their impact.
Build intelligent personal shopping experiences by adding the Manifest AI chatbot to your site. Help shoppers find what they need, faster. Double add-to-cart and conversion rates and get 25% higher AOV. Easy Shopify and help desk integration. Free 14-day trials.
Hire top talent with industry-specific expertise to build world-class engineering teams and solve complex business problems. AI-assisted sourcing, vetting, and hiring delivers the right talent, at the right time, globally!
Maintain healthy calendar habits, improve your productivity, optimize cross-team meetings, boost collaboration, and improve work-life balance. Reclaim schedules 1:1s and defends your calendar so you can focus. Syncs multiple calendars and integrates with Google Calendar, Zoom, Slack, and HubSpot.
Take customer support to the next level with powerful AI features including chatbots, workflow automation, emotion insights, knowledge bases, data-drive insights, real-time reporting, and empathy coaching to help your support teams increase productivity and customer satisfaction.
Incorporate your company, access one-click growth tools, stay compliant, and manage everything your business needs — all online, from anywhere. Launch your U.S. business in minutes with no paperwork or legal headaches.