RTX 8000 48GB down to $2,100 while 3090 prices doubled — owner calls it the best homelab all-rounder
triynizzles1 · reddit · 2026-11-03
A local-LLM enthusiast reports two-year price trends: his RTX 8000 48GB, bought for $2,400 in 2024, now sells around $2,100, while a single 3090 has risen to $1,500–1,600 (dual 3090s were $1,600–1,800 in 2024), making the 8000 cheaper per GB of VRAM.
Trade-offs: slower inference bandwidth than a 3090 (672 vs 936 GB/s) but double the memory for larger models/contexts/quants, and far better power draw (265W vs 700W for two 3090s). RTX Pro 6000 doubles memory at 8x the cost; Strix Halo and DGX Spark prices have also climbed.
Two years of trouble-free drivers: runs local-tier models fully in VRAM at 256k context, sometimes q6 (Qwen 27b, Llama 3 70b, QWQ); MoE models like gpt-oss-120b offload layers to 64GB system RAM, and with strata he gets q4xs at 390k context, 850 t/s prompt processing and 60–70 t/s generation (i9-9900k, DDR4). Verdict: $2,100 isn't cheap, but it may be the best do-everything homelab GPU right now.
More from Infra
- Peking University's MAServeBench benchmarks LLM serving with multi-agent workloads — PekingUniversity · 2026-11-03
- ProVe flips speculative decoding: large proposer, small verifier yields 2.5-2.7x speedup — Ziqi Jin · 2026-11-03
- NVFP4 DeepSeek on 4x DGX Spark reproduced: 2.3x vLLM at one user, 1.3x under load — AIFlow_ML · 2026-11-03
- Ubuntu's delayed CVE-2026-98432 fix is in -proposed, likely held back by database regression fixes — bytebot · 2026-11-03
- The Circuit: Anthropic's S-1, Intel's questions, Marvell and custom silicon bets — BenBajarin · 2026-11-03
- Analyst: Palantir earnings could hurt the AI trade more than help it, squeezing rivals alongside Nvidia — pdamodaran · 2026-11-03