will it local
The 2060 family bandwidth trap: base 2060 and 2060 12GB run 192-bit/336 GB/s; the Super runs 256-bit/448 GB/s — do not mix them up when comparing Build llama.cpp with CMAKE_CUDA_ARCHITECTURES=75 or silently lose ~19% to PTX JIT 8GB makes it the Turing pick over the 6GB cards: 8B Q4 fits fully with context room
computed band 42.0-63.0 tok/s for 8B Q4_K_M (roofline, 448.0 GB/s VRAM); community: 52.8-62.5 tok/s 8B Q4 (kb-audit-0903)
Effective decode window: 0.48–0.72 of 448 GB/s nominal → ~215–323 GB/s effective (llama.cpp decode, Q4_K_M basis; arch window, community-calibrated)
→ 13b-q4 no 14.6 GB vs 8 GB usable — does not fit
→ 3b-q4 full 2.8 GB needed of 8 GB usable — headroom for context
→ 4b-q4 full 3.7 GB needed of 8 GB usable — headroom for context
→ 8b-q4 full 6.1 GB needed of 8 GB usable — headroom for context
→ 52.8-62.5 tok/s · 8B Q4 measured (kb-audit-0903) · kb-audit-0903
Nominal bandwidth: 448.0 GB/s — see the effective decode window above
→ qwen3:0.6b on Tesla M60 (16GB, used) ($35) 0.6B · 0.6GB · needs ~1GB (4k ctx)
→ qwen3:1.7b on Tesla M60 (16GB, used) ($35) 1.7B · 1.4GB · needs ~1.8GB (4k ctx)
→ qwen3:4b on Tesla M60 (16GB, used) ($35) 4B · 2.6GB · needs ~3.2GB (4k ctx)
→ llama3.2:1b on Tesla M60 (16GB, used) ($35) 1B · 1.3GB · needs ~1.4GB (4k ctx)
→ llama3.2:3b on Tesla M60 (16GB, used) ($35) 3B · 2.0GB · needs ~2.4GB (4k ctx)
→ gemma3:1b on Tesla M60 (16GB, used) ($35) 1B · 1.0GB · needs ~1.1GB (4k ctx)
→ phi4:mini on Tesla M60 (16GB, used) ($35) mini 3.8B · 2.5GB · needs ~3GB (4k ctx)
→ moondream:2b on Tesla M60 (16GB, used) ($35) 2B · 1.7GB · needs ~3.7GB (4k ctx)
→ whisper:tiny on Tesla M60 (16GB, used) ($35) tiny · 0.1GB · needs ~2.1GB (4k ctx)
→ whisper on Tesla M60 (16GB, used) ($35) base · 0.3GB · needs ~2.3GB (4k ctx)
→ whisper:small on Tesla M60 (16GB, used) ($35) small · 0.9GB · needs ~2.9GB (4k ctx)
→ whisper:medium on Tesla M60 (16GB, used) ($35) medium · 3.1GB · needs ~5.1GB (4k ctx)
→ kokoro:82m on Tesla M60 (16GB, used) ($35) 82M · 0.3GB · needs ~2.3GB (4k ctx)
→ gemma3:4b on Tesla M60 (16GB, used) ($35) 4B · 3.3GB · needs ~3.8GB (4k ctx)
→ deepseek-r1:7b on Tesla M60 (16GB, used) ($35) distill 7B · 4.7GB · needs ~4.9GB (4k ctx)
→ mistral:7b on Tesla M60 (16GB, used) ($35) 7B · 4.1GB · needs ~4.6GB (4k ctx)
→ llava:7b on Tesla M60 (16GB, used) ($35) 7B · 4.7GB · needs ~6.7GB (4k ctx)
→ qwen3:8b on Tesla M60 (16GB, used) ($35) 8B · 5.2GB · needs ~5.8GB (4k ctx)
→ qwen2.5vl:7b on Tesla M60 (16GB, used) ($35) 7B · 5.6GB · needs ~5.8GB (4k ctx)
→ llama3.1:8b on Tesla M60 (16GB, used) ($35) 8B · 4.9GB · needs ~5.4GB (4k ctx)
→ deepseek-r1:8b on Tesla M60 (16GB, used) ($35) distill 8B · 4.9GB · needs ~5.4GB (4k ctx)
→ mistral-nemo:12b on Tesla M60 (16GB, used) ($35) 12B · 7.1GB · needs ~7.7GB (4k ctx)
→ qwen3:14b 14B — 9.9 GB vs 8 GB usable — partial CPU offload, expect large speed loss
→ gemma3:12b 12B — 9.6 GB vs 8 GB usable — partial CPU offload, expect large speed loss
→ deepseek-r1:14b distill 14B — 9.8 GB vs 8 GB usable — partial CPU offload, expect large speed loss
→ phi4:14b 14B — 9.9 GB vs 8 GB usable — partial CPU offload, expect large speed loss
→ whisper:large-v3 large-v3 — 8.2 of 8 GB usable — barely over; real with q8 KV cache (halves KV) or shorter context
→ sdxl SDXL base — 8.9 GB vs 8 GB usable — partial CPU offload, expect large speed loss
⚠ The 2060 family bandwidth trap: base 2060 and 2060 12GB run 192-bit/336 GB/s; the Super runs 256-bit/448 GB/s — do not mix them up when comparing
⚠ Build llama.cpp with CMAKE_CUDA_ARCHITECTURES=75 or silently lose ~19% to PTX JIT
⚠ 8GB makes it the Turing pick over the 6GB cards: 8B Q4 fits fully with context room