bigyabai

Damn, somebody's insecure.

It took CXMT 10 years to make a profitable DDR5 line, with a similar advantage of billions of dollars, contracts, political leverage, clean room fabs, and allegedly stolen IP.

It would probably be easier for Micron to double their memory production, than it would be for Intel to beat Micron's annual shipments.

I don't see the microprocessor comparison here. Dedicated inference ASICs already have a great chip design parallel - ASICs. Most PCs and smartphones avoid ASICs entirely to control things via the SOC if possible. The most popular modern microprocessors are all generally not ASIC designs.

Crypto miners already charged down this route for mining, and now there are thousands of ASIC mining rigs that are worth nothing, when GPU rigs are resold at a profit. ASICs are a dead-end, and I think that's why Nvidia couldn't give a rats ass what Taalas builds.

All I know for certain is that the future will have new models, and baked-in ASICs won't get them.

About ChatGPT, sure. Not Llama 3.1 8B.

I don't think it's obvious. The 8B model that the Taalas cards run aren't any larger than what a modern smartphone can inference with GPU compute shaders. Users don't need SRAM or dedicated hardware to get interactive tok/s at that scale.

To push beyond the envelope of what a mobile GPU can do, you'd have to build gigantic coprocessors for 30-100B models, which would probably cost more than the phone/laptop itself to integrate on-die. Nevermind the fact that SRAM is several times more expensive than DRAM gig-for-gig.

For the record, "the future" you are describing is running a tiny LLM from 2 years ago.

Those on-device weights are horribly outdated, underperform for their size, and cannot be updated to use a different model. I don't think Nvidia is afraid of Taalas.

> You could have manufactured enough ARC video cards to meet demand.

I don't know if anyone told you, but GDDR is included in the current RAM shortage.

It depends on what the tasks are. Text-native work like programming or interacting with a terminal should be a lot more immediate for LLMs than something abstract like Starcraft, DOTA, Minecraft or Mindustry. LLMs certainly know the rules for these games, but processing game state and formulating a plan in the form of a Python script or MCP call is a rough debugging loop.

Minecraft is really the perfect example, because humans get all sorts of unfair advantages. Even a young player has enough spatial reasoning to recognize and remember landmarks, visualize large projects or explore without getting lost. These are simple instincts that LLMs will struggle to match even with huge context sizes and native multimodal input.

Yes, but that doesn't contradict my statement or support your original claim. Tesla would be challenged to polarize European customers any further than they have already.

> what about my statement "Tesla is gaining share in Europe despite Chinese EVs" do you disagree with?

It's just missing the context that Tesla hemorrhaged European market share for 8 consecutive quarters.

From their current position, it would be genuinely difficult to find ways to piss off more of their customers.

Robotics has it's own supply chain issues. Apple's forced divestment away from China creates a supply gap that will be difficult or impossible to bridge with US suppliers. It could easily develop into another "Tesla vs. BYD" situation that depends on federal protectionism to work (and outright fails as a business outside of the US). When you look at it from that perspective, home robotics is just another Vision Pro product on the roadmap. It will have hardware limits, the price tag will be astronomically unaffordable, and competing systems will probably undercut Apple in seriously threatening ways. By the time that Apple can iterate robotics into something that customers want, the market will have festered into something beyond their control.

At risk of sounding like a glazer, Nvidia has a super clean path to home robotics no matter who wins the consumer segment. They have experience selling their hardware to system integrators like Nintendo and Tesla, and have shipped low-power GPGPU platforms on an SOC since the mid-2010s. The value add of CUDA is enormous, versus the struggle Apple will have bringing developers into the CoreML ecosystem. Nvidia has the momentum with Tegra and Jetson, the hardware and software lead will take years to catch up with.

I guess in a fair essay, this is the point where I should commend Apple for their output. They make the fastest AArch64 CPUs on the market, have an excellent "ecosystem" sell to the billions of iPhone owners in the world, and can ship any service from advertisement to F1 and sports streaming. Individually these are very impressive accomplishments that Tim Cook deserves commendation for. Taken as a whole, I don't see any coherent strategy that ties it all together. The "iPhone ecosystem" concept can't save expensive masterpiece products like the Vision Pro, even if it's a great headset. Users despise Apple's advertisement and service upsells, and single-core CPU performance isn't constraining home robotics. All of Apple's strongest tactics are pretty benign in the light of who they'll have to oppose.

This was a good discussion, and it's fun to game-theory future products with other people. Admittedly, asking Apple to compete with Nvidia's tech stack is like asking the coughing baby to defuse a hydrogen bomb. I recognize that the full scope of my demands for competition are unrealistic, but I think there's some meat on the bone that Apple is leaving for no good reason. The Mac Pro is one of those lowest-hanging fruit that didn't have to die, or at the very least could have been redesigned to champion the advantages of Apple Silicon. The Mac Studio is not a very confident future-facing product, and I feel a bit of worry for Apple's future designs seeing it as their flagship Mac. Apple has all the resources to push out the compute frontier for consumers and datacenters alike - here's hoping they join the fray, for all of our sakes.