The Dilemma of AI Accelerator Chip Manufacturers

 The Dilemma of AI Accelerator Chip Manufacturers

  1. When scaling out - the high-end market: To achieve optimal performance, it becomes an incredibly challenging task to design not only the accelerator chip and board but also incorporate HBM, high-bandwidth low-latency networking, a compiler that optimizes workloads for multiple chips within multiple boards, tiles, and layers on top of large-scale equipment.

  2. When not scaling out - the low-end market: Even if we simplify the workload optimization problem to L=M=1 and tolerate some inconvenience with the compiler, we still need to produce chips with high quality and performance at a lower cost than the RTX 4090, reaching around 4N in terms of production volume. Can NVIDIA lower the manufacturing cost of the RTX 4090 to compete with this market? It seems highly unlikely.

In the short term, abandoning option 1 and pursuing option 2, focusing on a narrow market segment with low-cost high-performance solutions, seems to be the only viable strategy. Option 1 is not feasible for short-term goals.

General-purpose chips become a bet with high probability of moderate success or a low probability of striking gold. Although there is a possibility of transitioning gradually from a narrow market to the general-purpose chip market after achieving moderate success, it remains uncertain whether these chips can be made more affordable and powerful than NVIDIA's offerings.

The market for LLM based on transformers in the narrow market segment is expected to witness fierce competition among startups. While higher pricing may be possible in the short term, over time it may become difficult to justify charging premium prices unless the chips outperform competitors significantly.

Amidst all this, there is the emergence of llama.cpp, a hardware like MacBooks that can generate LLM tokens without accelerators. It even supports speech recognition and offers various other functionalities. In summary, it is a challenging market where everyone knows the difficulties, yet the general sentiment seems to be dismissive.

Comments

Popular posts from this blog

Supercomputer Multiverse Evolutionary Theory

LLaMA wants a body to live an actual life.

You know that saying, "Poverty is inherited"?