Groq LPU

What hackers pick instead

• Google TPUGoogle's custom AI chips, rented through Google Cloud, are recommended for cheap, energy-efficient inference and large-scale clusters. Common caveat: hard to program and behind Nvidia at top performance.73recommend
• Taalas HC1Model-in-silicon chip praised for very fast, low-power inference; the main caveat is that the model cannot be updated without replacing the chip.59recommend
• Taalas hardwired LLM chipPeople recommend it for very fast LLM inference from a model etched into silicon. Critics note the chip is locked to one model and updates are limited.45recommend
• Cerebras wafer-scale chipWafer-scale chip that people say runs LLM inference far faster than GPUs. Critics agree it is costly and struggles with large models.29recommend
• Cerebras CS-4People recommend it for very fast AI inference, above 1,000 tokens per second on large models. Critics say it is scarce, hard to access, and not affordable for individuals.12recommend

Every quote

The 12 most detailed of 12 comments. Each links to the original.

1. Groq's technical decisions are unsound in a normal world. But being HBM-free may have some merit in a world where HBM supply is constrained.

   Mixed · Dec 2025 · in “Nvidia buying AI chip startup Groq for about $20B in cash”

2. Google's TPU and Groq's LPU are the only real, commercially viable ways to provide all the compute required for the inference people want.

   Recommends · Jun 2026 · in “Apple reveals new AI architecture built around Google Gemini models”

3. The most rapid path that Intel has to selling competitive GPUs, would be to licence designs from Groq

   Recommends · Feb 2026 · in “Intel will start making GPUs”

4. A cluster of LPUv2s would probably be faster and cheaper than an Infiniband cluster of Epycs.

   Recommends · Dec 2025 · in “macOS 26.2 enables fast AI clusters with RDMA over Thunderbolt”

5. DRAM-based GPUs/TPUs are quite inefficient for inference compared to SRAM-based Groq/Cerebras

   Recommends · Feb 2026 · in “How Taalas "prints" LLM onto a chip?”

6. No HBM needed on this card, and 32x faster inference than today's best GPUs at inference!

   Mixed · May 2026 · in “Higher usage limits for Claude and a compute deal with SpaceX”

7. Groq's chips only have 230MB of SRAM per chip vs 80GB on an H100

   Mixed · Dec 2025 · in “Nvidia's $20B Antitrust Loophole (Not an Acquisition)”

8. Their LPUs are way more efficient for inference.

   Recommends · Jan 2026 · in “Furiosa: 3.5x efficiency over H100s”

9. they scale the number of chips to fit a single model when they have no dram

   Criticizes · Dec 2025 · in “Nvidia buying AI chip startup Groq for about $20B in cash”

10. Gives NVIDIA a whole lineup of inference-focused hardware that iirc can retrofit into existing air cooled data centres

    Recommends · Jan 2026 · in “Furiosa: 3.5x efficiency over H100s”

11. has near impossible requirements for deployment

    Mixed · Dec 2025 · in “Nvidia buying AI chip startup Groq for about $20B in cash”

12. now worthless ASICs are rotting

    Criticizes · Jul 2026 · in “Kimi K3 is now live”