will it local
GTX 16-series Turing has NO tensor cores despite being Turing — llama.cpp falls back to dp4a INT8 kernels on CUDA cores 4GB VRAM: 7B/8B Q4 does not fit on-GPU; either heavy CPU offload (very slow) or drop to Q2/Q3 quants or 3-4B models Also sold in a GDDR6 variant (192 GB/s) — the GDDR5 original is 128.1 GB/s; check which one you're buying Only one approved localmaxxing run as of research date — bench depth is thin 75W TDP, many cards have no PCIe power connector — popular for OEM/SFF conversions CC 7.5 still supported by current CUDA/llama.cpp — better driver runway than Pascal
computed band 12.0-18.0 tok/s for 8B Q4_K_M (roofline, 128.1 GB/s VRAM); community: 30.6 tok/s Qwen3.5-4B IQ4_NL (https://localmaxxing.com/en/hardware/DISCRETE_GPU:nvidia+geforce+gtx+1650)
Effective decode window: 0.48–0.72 of 128.1 GB/s nominal → ~61–92 GB/s effective (llama.cpp decode, Q4_K_M basis; arch window, community-calibrated)
→ 3b-q4 full 2.8 GB needed of 4 GB usable — headroom for context
→ 4b-q4 tight 3.7 of 4 GB usable — keep context modest
→ 8b-q4 no 6.1 GB vs 4 GB usable — does not fit
→ 30.6 tok/s · Qwen3.5-4B IQ4_NL · source
Nominal bandwidth: 128.1 GB/s — see the effective decode window above
→ qwen3:0.6b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 0.6B · 0.6GB · needs ~1GB (4k ctx)
→ qwen3:1.7b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 1.7B · 1.4GB · needs ~1.8GB (4k ctx)
→ qwen3:4b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 4B · 2.6GB · needs ~3.2GB (4k ctx)
→ llama3.2:1b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 1B · 1.3GB · needs ~1.4GB (4k ctx)
→ llama3.2:3b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 3B · 2.0GB · needs ~2.4GB (4k ctx)
→ gemma3:1b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 1B · 1.0GB · needs ~1.1GB (4k ctx)
→ phi4:mini on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) mini 3.8B · 2.5GB · needs ~3GB (4k ctx)
→ moondream:2b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 2B · 1.7GB · needs ~3.7GB (4k ctx)
→ whisper:tiny on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) tiny · 0.1GB · needs ~2.1GB (4k ctx)
→ whisper on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) base · 0.3GB · needs ~2.3GB (4k ctx)
→ whisper:small on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) small · 0.9GB · needs ~2.9GB (4k ctx)
→ kokoro:82m on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 82M · 0.3GB · needs ~2.3GB (4k ctx)
→ gemma3:4b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 4B · 3.3GB · needs ~3.8GB (4k ctx)
→ deepseek-r1:7b distill 7B — 4.9 GB vs 4 GB usable — partial CPU offload, expect large speed loss
→ mistral:7b 7B — 4.6 GB vs 4 GB usable — partial CPU offload, expect large speed loss
⚠ GTX 16-series Turing has NO tensor cores despite being Turing — llama.cpp falls back to dp4a INT8 kernels on CUDA cores
⚠ 4GB VRAM: 7B/8B Q4 does not fit on-GPU; either heavy CPU offload (very slow) or drop to Q2/Q3 quants or 3-4B models
⚠ Also sold in a GDDR6 variant (192 GB/s) — the GDDR5 original is 128.1 GB/s; check which one you're buying
⚠ Only one approved localmaxxing run as of research date — bench depth is thin
⚠ 75W TDP, many cards have no PCIe power connector — popular for OEM/SFF conversions
⚠ CC 7.5 still supported by current CUDA/llama.cpp — better driver runway than Pascal
⚠ GTX 1650 Ti LAPTOP sibling measures ~2.8 tok/s on 8B Q4_K_M — 35-50W power-capped silicon, do not confuse with the desktop 1650
⚠ GTX 1650 Ti LAPTOP sibling measured 2.8 tok/s on Llama 3.1 8B Q4_K_M (localscore.ai/accelerator/374) — a DIFFERENT, slower card (lower TGP); do not read that number against this card's band