Running Gemma-4 26B at 124 tokens/SEC on a CPU, no GPU
Get picks like this daily. Today's launch numbers, the top launches, AI and tech news, and a daily opportunity pick, straight to your inbox.
Want to reach these readers? Sponsor an issue.
I wanted to know how fast a 26B mixture-of-experts model could run on a desktop CPU with no GPU. Got ~40 tok/s single-stream (lossless) and ~124 batched. The surprising part was the byte budget: for this model you compress the output head (32% of per-token bytes), not the experts (16%). The writeup has the bandwidth roofline and the dead-ends; the repo has the reproducible recipe. Happy to answer questions.Repo: https://github.com/arun-prasath2005/gemma4-cpu-moe
HN · launched Jun 30, 2026 · 10 upvotes (top 41%) · 0 views
Niche: llm inference engines · 133 launches, 4 in the last 7 days ▼ -33% · Explore this niche →
Category and tags
- Horizontal
- Model & infra
- Developer
- AI-native
- Hobby / open-source project
Visit website ↗ Source listing ↗
More like this: all 133 launches in llm inference engines →
Could you build this?
No Achieving 124 tokens/sec for a 26B MoE model on a CPU without a GPU requires high-performance systems engineering, low-level CPU vector assembly (AVX-512/AMX), and novel quantization/compression schemes.
What it would actually take: Requires custom C++/assembly or high-performance C inference kernels tailored to specific CPU cache architectures and SIMD instruction sets (AVX-512, AMX, or ARM NEON). The core difficulty lies in writing custom quantization kernels specifically optimizing output-head matrix multiplication alongside memory-bandwidth-bound MoE routing without hardware accelerators, far beyond general LLM API wrappers.
Discussion
1 comment analyzed.
Concerns raised: Output head byte budget is surprisingly high
Feature requests: More aggressive compression of head while keeping experts mostly untouched
Competitors
Other products that read as similar to this one : 207 launches clear the similarity bar, closest 8 shown.
Attention rank: #100 of 208 (itself plus its competitors, highest first, normalized so YC and Product Hunt are compared fairly).
Launched 244 days after the earliest competitor.
- Open-source engine running Gemma 4 26B in 2 GB RAM on any M-series Mac · hn · 2026-07-29 · 919 upvotes · similarity 0.61
- I made a Gemma 4 Mac app that names screenshots with local AI · hn · 2026-05-31 · 7 upvotes · similarity 0.52
- RunNburn · hn · 2026-07-30 · 11 upvotes · similarity 0.52
- GPU Rated · ph · 2026-10-26 · 1 upvotes · similarity 0.50
- A new engine to run Kimi K3 on a laptop · hn · 2026-07-29 · 7 upvotes · similarity 0.48
- Cactus Hybrid: We taught Gemma 4 to know when it's wrong · hn · 2026-07-22 · 191 upvotes · similarity 0.48
- ExANS · hn · 2026-08-05 · 15 upvotes · similarity 0.47
- Google Gemma 4 12B · ph · 2026-06-04 · 309 upvotes · similarity 0.46
The niche, labels, vibe code rating and competitor comparison on this page are analysis by Nicheloom. If you use them, please credit nicheloom.com. How you may use our data.