Tesla’s groundbreaking Supercomputer is going to be unveiled this week, with sources like @Sawyer Merritt buzzing about its imminent launch. The new Supercomputer is not just for AI; its sheer power makes it suitable for high-performance computing tasks. Elon Musk’s new beast boasts a peak performance surpassing that of Leonardo, a top-tier global Supercomputer, with 10,000 Nvidia GPUs. Potentially outpacing competitors, this technological marvel is set to revolutionize Tesla’s full-driving (FSD) training. A key figure at Tesla, Tim Zaman highlighted their vast training datasets, hinting at their edge in real-world video training.
However, Nvidia’s struggle to meet the H100 GPU demand has pushed Tesla. Dojo, Tesla’s $1 billion investment, a Supercomputer with custom-designed chips aiming to redefine FSD training. Alongside Dojo, Tesla’s Nvidia H100 GPU cluster is also coming online, a strategic move to dominate automotive computer power. Recent revelations of Elon Musk indicate Tesla’s ambitious $4 billion investment in AI & FSD training over the next 2 years.
A known Tesla enthusiast, Sawyer Merritt has shared deeper insights into the system’s intricacies on X.Com. Tesla’s reliance on GPUs, especially Nvidia’s H100, underscores their commitment to harnessing AI for autonomous driving. The hefty tag of $30,000 of the H100 each hasn’t deterred Tesla, who’s investing heavily in a 10,000 GPU training cluster. Yet the industry’s insatiable demand for the H100 has outpaced Nvidia’s supply capabilities. The inception of Dojo might have been circumvented had Nvidia met the demand, as hinted by Musk. As emphasized by Merritt, the challenge isn’t just computational power, it’s about leading & innovating in the automotive realm.
The vision of Tesla extends beyond supercomputers; they’re crafting unique data centers, as job listings in Austin suggest. These data centers are pivotal for Tesla, especially when handling vast video data essential for their self-driving software. However, it’s not all smooth sailing; Elon Musk’s recent comments hint at delays for FSD beta access in their Hardware 4 (HW4) vehicles. This six-month delay has HW4-equipped Models, X & Y owners in a quandary, as they might miss out on FSD updates. The strategy of Musk seems clear: perfect FSD on HW3 before rolling it out to the newer hardware.
The delay has sparked concerns, especially among those who have paid for FSD beta access on their HW4 vehicles. The drive of Musk on FSD v12, though impressive, showed the system’s imperfections, like not halting completely at a stop sign. Yet, the FSD v12’s ability to mimic human drivers, learning from videos & not relying on coded instructions, is revolutionary. The commitment of Tesla to data quality is evident, with Elon Musk emphasizing the importance of top-tier data for driving improvements. While showcasing advancements, the FSD v12 demo also highlighted areas needing more refinement before public release.
With Musk projecting a jump from 4 million to a staggering 10 million vehicles in a few years, Tesla’s fleet is set to expand. The world watches, as Tesla continues its relentless pursuit of AI-driven automotive excellence, anticipating the next leap in automotive driving.