32GB VRAM cards spike overnight: B70 jumps from $1.4K to $2K, derailing local LLM builds
tableball35 · reddit · 2026-10-25
A Reddit user seeking build advice: they were planning to buy an AMD R9700 ($1.7-1.8K) or B70 ($1.3-1.4K) with 32GB VRAM for local LLM inference, but prices spiked seemingly overnight to $2.3K and $2K, derailing their Black Friday plans. With two PCs (an Ultra 7 + 5070Ti build and an aging i9-14900F + 4070S box they planned to upgrade and link via RPC/Ethernet for dual-machine inference), they're now weighing a downgrade to 24GB options (A4500/B60) or a fiddly V100 setup, while budgeting extra DDR5 and a PSU replacement. The thread reflects the recent supply crunch and price surge in AI inference cards.
Related event: 32GB GPU Prices Surge, Disrupting Local AI Inference Builds(2 posts)→
More from Infra
- Claude Models Now GA on Google Cloud with New Anthropic-Managed Deployment Option — emmanuelvivier · 2026-10-25
- llama.cpp PR fuses FFN gate/up matmuls, up to 10.7% faster inference — jacek2023 · 2026-10-25
- Task-specific inference engines emerging as specialization costs collapse, argue researchers — _AustinCalvert_ · 2026-10-25
- Reverse-engineering Apple's on-device 20B model: MoE in flash, only 2GiB RAM — julsimon · 2026-10-25
- NVIDIA shows off an 8-unit DGX Spark cluster at GTC Berlin — MaziyarPanahi · 2026-10-25
- Reddit: what's the minimum hardware budget to self-host DeepSeek 4.1 Flash at ~40 tok/s — Robert__Sinclair · 2026-10-25