will it local
Launched March 4 2025 ($549). LocalScore 8B median: 55.9 tok/s gen, LocalScore 709, TTFT 477 ms (accelerator/168). TPU: GB205, 6144 CUDA / 176 Tensor cores, 28 Gbps effective GDDR7, 672 GB/s, 192-bit bus — brief's '~672' confirmed.
computed band 73.0-100.0 tok/s for 8B Q4_K_M (roofline, 672.0 GB/s); community: 55.9 tok/s Llama 3.1 8B Instruct Q4_K_M (https://www.localscore.ai/accelerator/168)
Effective decode window: 0.56–0.80 of 672 GB/s nominal → ~376–538 GB/s effective (llama.cpp decode, Q4_K_M basis; card-specific window from cited community benches)
Community check — localmaxxing.com medians (2 buckets): localmaxxing.com
→ ~180 tok/s · 3-4B Q4 · 3 runs · inside our computed band
→ ~69 tok/s · 12-15B Q4 · 3 runs · just above our band (15% over the high edge)
→ 13b-q4 offload-partial 14.6 GB vs 12 GB usable — partial CPU offload, expect large speed loss
→ 3b-q4 full 2.8 GB needed of 12 GB usable — headroom for context
→ 4b-q4 full 3.7 GB needed of 12 GB usable — headroom for context
→ 8b-q4 full 6.1 GB needed of 12 GB usable — headroom for context
→ 55.9 tok/s · Llama 3.1 8B Instruct Q4_K_M LocalScore median · source
Nominal bandwidth: 672.0 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)
→ llava:7b on Raspberry Pi 5 + AI Kit (Hailo-8L) ($110) 7B · 4.7GB · needs ~6.7GB (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)
→ whisper:large-v3 on Tesla P100 (16GB, used) ($135) large-v3 · 6.2GB · needs ~8.2GB (4k ctx)
→ mistral-nemo:12b on Tesla P100 (16GB, used) ($135) 12B · 7.1GB · needs ~7.7GB (4k ctx)
→ sdxl on Tesla P100 (16GB, used) ($135) SDXL base · 6.9GB · needs ~8.9GB (4k ctx)
→ llava:13b on Tesla P100 (16GB, used) ($135) 13B · 8.0GB · needs ~11.1GB (4k ctx)
→ gemma3:12b on Tesla P100 (16GB, used) ($135) 12B · 8.1GB · needs ~9.6GB (4k ctx)
→ qwen3:14b on Tesla P100 (16GB, used) ($135) 14B · 9.3GB · needs ~9.9GB (4k ctx)
→ deepseek-r1:14b on Tesla P100 (16GB, used) ($135) distill 14B · 9.0GB · needs ~9.8GB (4k ctx)
→ phi4:14b on Tesla P100 (16GB, used) ($135) 14B · 9.1GB · needs ~9.9GB (4k ctx)
→ gpt-oss:20b 20B MoE — 14.2 GB vs 12 GB usable — partial CPU offload, expect large speed loss
⚠ 12GB VRAM is a hard wall for 14B+ Q4 models — Qwen2.5 14B Q4_K_M runs at just 20.8 tok/s with 1.07 s TTFT on this card
⚠ Blackwell '5090-class AI performance with 12GB' marketing does not translate to VRAM-constrained LLM inference
⚠ DLSS 4 Multi Frame Generation is 50-series exclusive but irrelevant to inference workloads
⚠ Raw compute gains over a 4070 SUPER are modest; the 672 vs 504 GB/s bandwidth jump is the real upgrade
⚠ Bench anomaly (documented 0904): LocalScore 8B median (55.9) implies eff ~0.44 on 672 GB/s GDDR7 — below arch window; early small sample. Band floor widened to contain the recorded row.
$900 — 16GB / 896 GB/s — the same-tier fix for the 12GB wall and a real step up in tok/s
If buying used: 24GB + 1008 GB/s beats the 5070 outright for LLM serving, often near the same used price