will it local
No tensor cores; CUDA-core-only inference (CC 6.1) Pascal support deprecated in CUDA 13+ — pin older llama.cpp builds or use Vulkan backend (Vulkan TG can be faster than CUDA on this card per llama.cpp issue #19817) 6GB fits 7B Q4_K_M with modest context; 8B dense needs tight KV cache or partial offload MoE models (35B-A3B class) are surprisingly usable (~15-17 tok/s) with expert offloading, --no-mmap and --mlock Requires 1x 6-pin PCIe power (120W TDP) Beware the 3GB variant — different card with 1152 CUDA cores and much worse LLM usability; verify 6GB before buying
computed band 10.0-19.0 tok/s for 8B Q4_K_M (roofline, 192.2 GB/s VRAM); community: 19.1 tok/s DeepSeek R1 Distill Qwen 7B Q4_K_M (https://www.localscore.ai/accelerator/1289); community: 12 tok/s Qwen3-8B (dense, fully on GPU) Q4_K_M (https://www.youtube.com/watch?v=9vY4-Z-tkHs (Codacus, llama.cpp)); community: 15.0 tok/s Qwen3.6-35B-A3B (MoE, expert offload) UD-IQ2_M (https://localmaxxing.com/en/hardware/DISCRETE_GPU:nvidia+geforce+gtx+1060+6gb)
Effective decode window: 0.28–0.50 of 192.2 GB/s nominal → ~54–96 GB/s effective (llama.cpp decode, Q4_K_M basis; arch window, community-calibrated)
→ 13b-q4 no 9.9 GB vs 6 GB usable — does not fit
→ 3b-q4 full 2.8 GB needed of 6 GB usable — headroom for context
→ 4b-q4 full 3.7 GB needed of 6 GB usable — headroom for context
→ 8b-q4 tight 6.1 of 6 GB usable — barely over; real with q8 KV cache (halves KV) or shorter context
→ 19.1 tok/s · DeepSeek R1 Distill Qwen 7B Q4_K_M · source
→ 12 tok/s · Qwen3-8B (dense, fully on GPU) Q4_K_M · source
→ 15.0 tok/s · Qwen3.6-35B-A3B (MoE, expert offload) UD-IQ2_M · source
→ 25-28 tok/s · falcon-h1 7B Q4_K_S · source
Nominal bandwidth: 192.2 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)
→ whisper:medium on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) medium · 3.1GB · needs ~5.1GB (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 on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) distill 7B · 4.7GB · needs ~4.9GB (4k ctx)
→ mistral:7b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 7B · 4.1GB · needs ~4.6GB (4k ctx)
→ qwen3:8b on Tesla P100 (16GB, used) ($135) 8B · 5.2GB · needs ~5.8GB (4k ctx)
→ qwen2.5vl:7b on Tesla P100 (16GB, used) ($135) 7B · 5.6GB · needs ~5.8GB (4k ctx)
→ llama3.1:8b on Tesla P100 (16GB, used) ($135) 8B · 4.9GB · needs ~5.4GB (4k ctx)
→ deepseek-r1:8b on Tesla P100 (16GB, used) ($135) distill 8B · 4.9GB · needs ~5.4GB (4k ctx)
→ llava:7b 7B — 6.7 GB vs 6 GB usable — partial CPU offload, expect large speed loss
⚠ No tensor cores; CUDA-core-only inference (CC 6.1)
⚠ Pascal support deprecated in CUDA 13+ — pin older llama.cpp builds or use Vulkan backend (Vulkan TG can be faster than CUDA on this card per llama.cpp issue #19817)
⚠ 6GB fits 7B Q4_K_M with modest context; 8B dense needs tight KV cache or partial offload
⚠ MoE models (35B-A3B class) are surprisingly usable (~15-17 tok/s) with expert offloading, --no-mmap and --mlock
⚠ Requires 1x 6-pin PCIe power (120W TDP)
⚠ Beware the 3GB variant — different card with 1152 CUDA cores and much worse LLM usability; verify 6GB before buying