Trading 2 RTX 5090s for 4 R9700s: does doubled VRAM pay off?
poofph · reddit · 2026-10-16
The poster currently runs local AI on two RTX 5090s, getting 150 tok/s and 4k prefill on strata and flash-next Q5KXL. They consider selling one 5090, putting the other back in their gaming PC, and buying four AMD R9700s for roughly the price of one 5090 — doubling VRAM. Open questions: how much inference speed they'd lose, whether doubled VRAM compensates, and whether future models will run well on the R9700s.
More from Infra
- MiniCPM-V 4.5 MoE reportedly hits 13+ tokens/s on iPhone with just 3GB memory — heyshrutimishra · 2026-10-16
- Solo-built Vulkan inference engine LaForge hits ComfyUI-level SD1.5 speed: 512px in 0.8s, 402MB container — Ichidan256 · 2026-10-16
- Running PolyStrata + Cline coding agent on a $0-fancy laptop at ~90 tok/s — rpwoerk · 2026-10-16
- TensorFold 1.0.12 runs 120B vision model on 1-2 DGX Sparks, adds Mac image input — HankYeomans · 2026-10-16
- vLLM brings frontier model local deployment to NVIDIA RTX Spark PCs — vllm_project · 2026-10-16
- NVIDIA signals $2B in new European AI deals at GTC Berlin, totaling ~$10B in a year — CurieuxExplorer · 2026-10-16