Building a DDR4+HBM2 local inference box: capacity over speed
Ancapta · x · 2026-09-03
A Reddit user asks which local LLM hardware upgrades actually feel meaningful, arguing that going from 32+16GB to 64+16GB is less worthwhile than 32+32GB. He plans to build a DDR4 + HBM2 inference machine to complement his main 32GB DDR5 + 16GB GDDR7 system, trading speed for greater total capacity to run a wider variety of models, and asks whether 32+32, 128+32 or 64+64 is the most logical capacity target.
Related event: Reddit Debates RAM vs VRAM Upgrades for Local LLMs(2 posts)→
More from Infra
- KV cache, not parameter count, may be the real bottleneck for long-context local models — jonejy · 2026-09-03
- RX 6800 XT vs RX 9060 XT for ComfyUI: is the $105 premium worth it for LTX and Wan? — Nice-Regret-9207 · 2026-09-03
- Developers warn a new Cloudflare feature could hurt your SEO — turn it off — gaganghotra_ · 2026-09-03
- If Amazon Trainium is any good, why weren't they at Hot Chips? — firstadopter · 2026-09-03
- Dragonfly rethinks Redis with sharded multi-threading to scale across modern multi-core servers — techNmak · 2026-09-03
- Zeiss exec: China is about 15 years behind in cutting-edge chipmaking tools — broodsugar · 2026-09-03