Dual-RTX-3090 user weighs $7,000 CMP 170HX upgrade for next open models
eribob · reddit · 2026-09-27
A local-LLM hobbyist running Qwen 3.8 27B on dual RTX 3090s (fp8, 129k context) debates upgrading for upcoming Qwen 3.8 Flash, DeepSeek V4 Flash and GLM 5.3 Flash. Cheapest path: two CMP 170HX 64GB cards at $6-7,000 in Europe for 176GB total VRAM (3-4bit quants); 4x V100 32GB is half the price but slower, power-hungrier and fussier. Free option: wait for Qwen 4 27B.
More from Infra
- Pathway's 150M-parameter BDH reasons in latent space, beyond Transformers — bigdata · 2026-09-27
- Stale Data Was an Analytics Problem. Agents Make It an Operations Problem — bigdata · 2026-09-27
- Managing 5 LLM providers: which gateway handles billing, spend caps, and fallbacks? — InsuranceDependent71 · 2026-09-27
- Warp shows 3 ways software factories improve themselves: skills, memory, model routing — AI Engineer · 2026-09-27
- MEM v3 Open-Sourced: Dynamic Batch Scaling Against CUDA OOM, Survives +10GB VRAM Shocks — uBazzyZ- · 2026-09-27
- Three myths of hosted LLM inference: sticker prices, interchangeable endpoints, and self-hosting — TangeloOk9486 · 2026-09-27