Ideogram 4 quantization test: int8convrot is 30-40% faster but clearly worse than fp8
Exciting-Action6636 · reddit · 2026-11-08
A Reddit user benchmarked Ideogram 4's two quantized variants:
- Speed: int8convrot runs about 30-40% faster than fp8.
- Quality: fp8 is clearly better — int8convrot softens and blurs textures.
- Why: Ideogram 4's bf16 weights were never released, only fp8. The community int8convrot builds were made by dequantizing fp8 to fp16 and then re-quantizing to int8, so the double quantization costs quality.
The author's conclusion: the usual "int8convrot beats fp8" rule of thumb doesn't hold for Ideogram 4 — it keeps the speed advantage but loses the quality edge. Pick fp8 for quality, int8convrot for speed.
More from Infra
- Hugging Face launches live training streams, starting with RL-tuned LoRA adapters — art_zucker · 2026-11-08
- Lobsters thread: how AI traffic is degrading the web for human developers — vboykis · 2026-11-08
- Oracle's quarterly capex up 2,540% to $28.5B in five years, fastest growth among hyperscalers — Beth_Kindig · 2026-11-08
- Decentralized Compute Network Lium Hits $2.32M GPU Spend in October, Up 74% MoM — markjeffrey · 2026-11-08
- Lex Sokolin: Google wants a piece of AI datacenters in space — minchoi · 2026-11-08
- Megaspeed exec recruited to spy on Nvidia CEO, US intelligence memo reportedly claims — TechCrunch AI · 2026-11-08