will it local
Still the best used price/VRAM point in NVIDIA-land — a used 3090 spans ~$900–1,300 for 24GB of fast GDDR6X; a used 4090 asks about $2,000 for the same capacity. See live buy links for today's number. 350W + GDDR6X pads on mining history - repad before trusting. In 2026 the smart 24GB play is right-sized 20-35B models, not the biggest 70B quant you can cram. This IS the upgrade target for most legacy owners.
computed band 84.0-140.0 tok/s for 8B Q4_K_M (roofline, 936 GB/s VRAM); community: 115.3 tok/s Qwen3 8B Q4_K (https://www.hardware-corner.net/gpu-llm-benchmarks/rtx-3090/); community: 92 tok/s Meta Llama 3.1 8B Q4_K_M (https://www.kunalganglani.com/llm-benchmarks); community: 95.7 tok/s Meta Llama 3.1 8B Q4_K_M (https://www.localscore.ai/model/1)
Effective decode window: 0.46–0.76 of 936 GB/s nominal → ~431–711 GB/s effective (llama.cpp decode, Q4_K_M basis; arch window, community-calibrated)
Community check — localmaxxing.com medians (5 buckets): localmaxxing.com
→ ~130 tok/s · 3-4B Q4 · 11 runs · just below our band (22% under the low edge)
→ ~85 tok/s · 8-9B Q4 · 20 runs · inside our computed band
→ ~49 tok/s · 12-15B Q4 · 34 runs · inside our computed band
→ ~26 tok/s · 20-27B Q4 · 20 runs · inside our computed band
→ ~26 tok/s · 30-33B Q4 · 20 runs · inside our computed band
→ qwen3 8B — ~83–140 tok/s · the speed pick: the fastest useful quality on the card (class rows 92–115)
→ gemma3 12B — ~53–88 tok/s · the everyday step-up: 8.1GB of 24GB, context stays cheap
→ qwen3 30B-A3B MoE — ~230–380 tok/s (active-expert estimate) · the MoE play: 18.6GB file, ~3B experts active
→ qwen3 32B — ~22–36 tok/s · the ceiling pick: tight but real — 21GB of 24GB at 4k ctx, keep context modest
→ devstral 24B — ~29–49 tok/s · the coding pick: 14.6GB leaves headroom for long context
→ 13b-q4 full 14.6 GB needed of 24 GB usable — headroom for context
→ 3b-q4 full 2.8 GB needed of 24 GB usable — headroom for context
→ 4b-q4 full 3.7 GB needed of 24 GB usable — headroom for context
→ 8b-q4 full 6.1 GB needed of 24 GB usable — headroom for context
→ 92 tok/s · Meta Llama 3.1 8B Q4_K_M llama.cpp · source
→ 95.7 tok/s · Meta Llama 3.1 8B Q4_K_M LocalScore median · source
→ 95 tok/s · Qwen 2.5 7B Q4_K_M llama.cpp b3520, 2K ctx · source
→ 108.0 tok/s · Meta Llama 3.1 8B Q4_0 Ollama 0.3.9 (3-4 run avg, runpod sheet) · source
→ 115.3 tok/s · Qwen3 8B Q4_K llama.cpp llama-bench -fa 1, 4K ctx · source
short-context (≤4k) numbers; measured down-scaling at longer contexts: ×0.75 at 16k, ×0.58 at 32k, ×0.40 at 64k, ×0.24 at 128k
Nominal bandwidth: 936 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 on Tesla P40 (24GB, used) ($290) 20B MoE · 14GB · needs ~14.2GB (4k ctx)
→ devstral:24b on Tesla P40 (24GB, used) ($290) 24B · 14.6GB · needs ~15.2GB (4k ctx)
→ qwen3.8:27b on Mac mini M4 (32GB) ($999) 27B · 16.5GB · needs ~18.5GB (4k ctx)
→ gemma3:27b on Mac mini M4 (32GB) ($999) 27B · 17GB · needs ~18.9GB (4k ctx)
→ flux:schnell on Mac mini M4 (32GB) ($999) schnell fp8 · 17GB · needs ~19GB (4k ctx)
→ qwen3:30b-a3b on Mac mini M4 (32GB) ($999) 30B-A3B MoE · 18.6GB · needs ~19GB (4k ctx)
→ qwen3-coder:30b-a3b on Mac mini M4 (32GB) ($999) 30B-A3B MoE · 18.6GB · needs ~19GB (4k ctx)
→ qwen3:32b on Mac mini M4 (32GB) ($999) 32B · 20GB · needs ~21GB (4k ctx)
→ deepseek-r1:32b on Mac mini M4 (32GB) ($999) distill 32B · 20GB · needs ~21GB (4k ctx)
→ qwen2.5vl:32b on Mac mini M5 Pro (64GB) ($1669) 32B · 22GB · needs ~23GB (4k ctx)
→ mixtral:8x7b 8x7B MoE — 26.5 GB vs 24 GB usable — partial CPU offload, expect large speed loss
→ flux:dev dev fp8 — 25.0 of 24 GB usable — barely over; real with q8 KV cache (halves KV) or shorter context
⚠ 350W — verify PSU and case airflow
⚠ GDDR6X pads + mining history — repad before trusting
⚠ In 2026 the smart 24GB play vs a used 3090 is price-dependent