Tesla participated in Hot Chips 34 and shared a ton of mind-boggling information about their Dojo supercomputer and chip architecture.
The details shared about Dojo in Hot Chips 34 by Tesla’s Emil Talpes, who worked at AMD for about 17 years for Opteron processors, are only about the hardware and capabilities of the tiles and Dojo as a whole. The performance will be discussed at Tesla’s AI Day II on September 30th.
The goal with Dojo is, according to Musk, “to be really good at video training. We have probably the fourth or approaching the third most powerful computing center in the world for AI training. Our first goal with Dojo is to make it competitive and be more effective and neural net training than a whole bunch of GPUs.”
Since Tesla needs a lot of computing power to process the video data from the vehicles in its fleet, it has built a proprietary system-on-wafer solution. According to ServeTheHome, “Each D1 die is integrated onto a tile with 25 dies at 15kW. Beyond the 25 D1 dies, there are also 40 smaller I/O dies.”
All of the power and cooling is integrated directly on the Training Tile, which is capable of 10 TB/s on-tile bisection bandwidth and 36 TB/s off-tile aggregate bandwidth. This architecture allows for the tiles to be scaled with 9TB/s links between them. They can also be plugged in and do not require their own server.
“The defining goal of our application is scalability,” Talpes said at the end of the presentation. “We have de-emphasized several mechanisms that you find in typical CPUs, like coherency, virtual memory, and global lookup directories just because these mechanisms do not scale very well when we scale up to a very large system. Instead, we have relied on a very fast and very distributed SRAM storage throughout the mesh. And this is backed by an order of magnitude higher speed of interconnect than what you find in a typical distributed system.”
The inside look into what Tesla’s building behind the scenes continues to prove how and why Tesla is at the forefront in artificial intelligence and neural net training. It gives the Tesla community an added sense of comfort knowing that Tesla will always have scalability and innovative technology at the forefront of everything the automotive company does.
Watch Anastasi In Tech’s Recap of Tesla’s Hot Chip 34 Presentation
Subscribe
Subscribe to our newsletter to stay up to date on the latest Tesla news, upcoming features and software updates.
Tesla’s ambitious 4680 cell program has been pivotal for its vehicle roadmap - and in particular, for the Cybertruck. Bonnue Eggleston, Tesla’s Senior Director for the 4680 cell project at Tesla, recently sat down with Sandy Munro on Munro Live, offering valuable insights into cell development, manufacturing hurdles, and Tesla’s future trajectory. You can watch the video in its entirely below.
The 4680 cell, like many batteries, is characterized by its dimensions: 46mm in diameter and 80mm long. Tesla is currently producing the 2nd generation of the 4680 - internally known as the Cybercell - which is shipped with every variant of the Cybertruck. This Gen 2 variant is a considerable step up from Gen 1 - whose limited production was cancelled following the slow charging issues with the 4680 Model Y.
Prototypes are Easy; Production is Hard
Bringing the 4680 from a concept cell to mass production hasn’t been easy, but according to Tesla, it has now become Tesla’s cheapest cell per kWh. Eggleston emphasized in the video that scaling up was an immense challenge - and required an extreme attention to detail.
With a team possessing a broad skill set, it took considerable effort to bring the 4680 to life, starting from the raw electrode material and progressing through the crucial formation process.
Breaking Barriers
To overcome these hurdles, Eggleston’s team leaned into innovation and focused on new processes that had not been utilized in the battery world yet. The groundbreaking new dry electrode process is the key here, which eliminates the use of toxic solvents and large ovens required in traditional production methods. This reduces internal factory footprint, while also being cleaner and safer, building a better cell from the ground up.
Complementing this, Tesla has also been developing a custom electrolyte formulation in-house, tailored specifically for their anode, cathode, and separator materials, all aimed at expanding their deep vertical integration.
This vertical integration has been key to the 4680 program, and Tesla has further extended it, with in-house production of components like cell cans serving to optimize the process and reduce waste. Eggleston also pointed out the unique terminal design on the 4680, which allows for easier and more reliable welding, contributing to the high production output that Tesla is aiming for.
Sustainability
On the sustainability front, Tesla has been hard at work recovering and recycling materials right from the manufacturing line to minimize waste. Eggleston highlighted this as part of Tesla's effort to promote sustainability, which ties in with the environmental benefits gained from avoiding solvents in the dry electrode process.
Structural Battery Packs
While the 4680 is intrinsically linked to the Cybertruck, we expect Tesla to expand this to its future vehicles eventually - whether through use of the specific cell format, or the technologies learned through its development. Eggleston noted that the efficiency of the Cybertruck is partly due to his team’s cooperation and work with the vehicle team. The structural battery pack minimizes weight and provides additional support and protection to the cabin and occupants.
4680 in the Future
Eggleston expressed a considerable amount of confidence in Tesla’s 4680 program and the progress - citing significant improvements in throughput, yields, and product quality since he took leadership.
He acknowledged the ambitious targets that Tesla and Elon have set - and mentioned that the use of metrics like headcount per gigawatt helps drive production efficiently. This metric essentially measures labor efficiency – producing more battery capacity (gigawatt-hours) with fewer people indicates a more streamlined and cost-effective manufacturing process.
While Eggleston hinted at future developments, and we have previously heard of Tesla working on even more cells for the future, the battery technology race has been progressing rapidly around the world. While Tesla has been pushing 4680 production and deploying 325kW-capable Superchargers (and soon 500kW), they continue to face challenges from the competition.
Brands like China’s Zeekr are demonstrating new LFP batteries capable of charging from 10-80% in under 10 minutes, while achieving sustained speeds of 400kW+. Currently, the Cybertruck can only sustain the 325kW cap speed for a few minutes at best, resulting in a sub-par charging curve compared to upcoming competitors.
Tesla will have to focus on developing and producing new cells that maintain that cost-competitive advantage the 4680 has built, while also achieving faster charging speeds across its entire lineup. For now, these new faster charging speeds are restricted to the Cybertruck, but with refreshes for the Model S and Model X on the horizon, we expect that Tesla’s updated flagship vehicles will make the best use of this tech until it is ready for the rest of the lineup.
According to a report from Reuters, obtained from three sources with inside knowledge, Tesla is delaying the launch of its more affordable model - the cheaper variant of the Model Y. Tesla has promised the launch of a more affordable model in the first half of 2025, but we have yet to see anything come to light.
Tesla has been especially secretive with new product launches - the fact that nobody leaked the announcement of the Robovan or the final design of the Cybercab for We, Robot shows Tesla’s commitment to keeping things under wraps.
Tesla’s Affordable Vehicle Plan
The Reuters report sheds some light on what Tesla's immediate affordable vehicle plans might actually entail with a changed approach: a stripped-down version of its best-selling Model Y SUV, internally code named E41. This approach seems to align with earlier speculation that Tesla might opt for more affordable versions of existing platforms rather than launching an entirely new vehicle like the often-rumored "Model Q" or "Project Redwood."
According to Reuters' sources, the US production launch for this cheaper Model Y has slipped. Instead of the first half of 2025, the target for the start of US production is now reportedly somewhere between Q3 2025 and early 2026. The specific reason for this delay wasn't made clear in the report.
‘E41’ Speculations
Despite the delay, Tesla appears to have set a production target for the vehicle in the US, aiming for 250,000 vehicles in 2026. This suggests that Tesla is banking on this being a high-volume vehicle, and expects significant demand at the price point. The report also reiterates previous accounts that this affordable Model Y variant is planned for eventual production in China and Europe as well.
Sources familiar with the Chinese plans indicated a 2026 launch there, with the vehicle potentially being smaller and costing 20% less to produce than the current refreshed Model Y. The timing for a European rollout remains unclear, but would likely follow a Chinese launch.
Affordable Model 3
Interestingly, the Reuters report also mentions, based on the three sources, that Tesla is planning a similar strategy for the Model 3, intending to launch a bare-bones version. This will likely be an even more cut-down version of the Mexican Model 3. This further supports the idea that Tesla's path to affordability, for now, involves cost-optimizing its existing popular models.
These more affordable vehicles will be crucial for Tesla. The company reported its first annual delivery decline recently, and faces analyst expectations of another drop this year. Factors cited include increased Chinese competition and potential brand reputation impacts. A cheaper Model Y and Model 3 could help attract new customers.
Strategy Pivot
This strategy also exists in the context of Tesla's potential pivot away from a dedicated, all-new $25,000 platform. Elon had previously indicated that the project was deprioritized in favor of focusing resources on developing the Robotaxi network and Cybercab.
The E41 and cheaper Model 3 appear to be the revised approach to hitting lower price points using existing manufacturing lines and platforms, albeit likely not reaching the originally hoped-for $25,000 base price - perhaps after the Federal EV Rebate.
Tariff Impact
Finally, the report touched upon the challenging geopolitical and economic landscape, particularly potential US tariffs. Automakers are wary of rising costs from tariffs on imported vehicles and parts. Tesla has reportedly increased its North American sourcing for many components over the past couple of years, a move that could lessen the tariff impact on a US-produced E41.
This contrasts with recent reports suggesting Tesla suspended plans to ship some components from China for the Cybercab and Semi programs, specifically due to tariff concerns. This highlights the complex supply chain calculations Tesla is making as it tries to balance production cost, tariffs, and vertical supply chain integrations.
With Tesla's Q1 2025 earnings call scheduled for Tuesday, everyone, including us, will be keenly listening for the next steps on these crucial affordable models.