Bought a $6,279 RTX 5090 Prebuilt: Was it the right choice for local AI?
Sexyvette07 · reddit · 2026-09-02
The author purchased a $6,279 prebuilt PC with an RTX 5090 and 64GB RAM after struggling with VRAM limitations on a 4080, seeking community validation.
Motivation:
- Previous 16GB VRAM was insufficient for LLMs, leading to excessive time spent on quantization optimization.
- Considered Mac Studio M5 Ultra (unified memory benefits) and dual-GPU AMD setups, but rejected the latter due to cost and gaming incompatibility.
Why 5090?:
- 32GB VRAM + 64GB RAM allows running 70B models or high-performance <30B models (like Qwen 27B).
- Retains CUDA ecosystem, Windows, and x86 compatibility for gaming and dev.
- Includes a 4TB Gen4 NVMe SSD.
Trade-offs:
- Significant performance penalty when models spill over to system RAM.
- Much lower power efficiency compared to ARM-based Mac Studios.
More from Embodied
- Opus-assisted open design: a $31k low-field head MRI sketched in hours — DreamFly_13 · 2026-10-18
- DIY robotics build: servo mechanism tested, wide yaw range but limited pitch — _Stocko_ · 2026-10-18
- Brain implants are restoring speech to people with ALS, Science reports — EricMTrautmann · 2026-10-18
- Meta wearables VP runs his AI agent across 3 phones and 23 pairs of glasses — nikola_mr64990 · 2026-10-18
- AMI-EV: spinning prism gives event cameras stable texture, Science Robotics 2024 — RexDouglass · 2026-10-18
- Neuralink patients control robotic arms, drones and games with thought alone — XFreeze · 2026-10-18