Ornith 35B-A3B MoE hits 78 tok/s with 100% accuracy on NVIDIA RTX Spark in first tests
draginol · x · 2026-10-19
Blogger draginol tested the Ornith 35B-parameter model with 3B active params (MoE) on an NVIDIA RTX Spark N1X (Blackwell, 64GB RAM): prefill 1,833 tok/s, decode 78.1 tok/s, 674ms first token, 64K context. It scored 100% on 80 simple tasks; tougher benchmarks coming next.
Related event: Ornith 35B/3B MoE Runs Fast and Accurate Locally(2 posts)→
More from Infra
- GPU programmer says the field needs lighter, more predictable inference engines — brandon_xyzw · 2026-10-19
- llama-doctor: CLI that diagnoses your llama.cpp server from /metrics with severity-graded findings — msalsas · 2026-10-19
- Underwater AI data center planned in Alaska's Cook Inlet, home to 330 endangered belugas — anselm · 2026-10-19
- DeepInfra selling DeepSeek v4.1 at 49% off sparks doubts it can turn a profit — gharik · 2026-10-19
- 330MB EmbeddingGemma2 indexes 6,000+ screenshots: text beats visual descriptors in cosine similarity — jupiterbjy · 2026-10-19
- OpenFaaS bought 4x DGX Sparks for air-gapped coding and red-teaming — natesiggard · 2026-10-19