Qwen3.5 9B Instruct (Q8)
Local OKOKB · 26
Best for: Quality, Coding, Reasoning·Pop: 86/100
Perf: est. 69 tok/s, typical error 5% · first token ~1.0s
Fits in 16 GB VRAM with room to spare. Best for quality, coding, reasoning on RTX 4080 SUPER.
Best Local AI Models for RTX 4080 SUPER (16GB)
The RTX 4080 SUPER delivers near-flagship performance with 16GB GDDR6X. It runs 8B models at est. 102 tok/s, typical error 5% and 14B models at around est. 63 tok/s, typical error 15%. Its high memory bandwidth makes it one of the fastest 16GB cards available.
Quick answer
The best local LLM for the RTX 4080 SUPER is Qwen3.5 9B Instruct (Q8) (est. 69 tok/s, typical error 5%) on its 16GB VRAM. It uses ~10.7GB of VRAM; the RTX 4080 SUPER handles up to 14B parameter models at Q4. A 14B model at Q4 runs at est. 63 tok/s, typical error 15%. (ModelFit, 2026)
Sizing rule: a Q4 model needs about 0.6 GB of VRAM per billion parameters, and ModelFit budgets 90% of this card's 16GB for weights, context, and KV-cache. The per-size table below uses that same budget. Other strong fits: Gemma 4 12B (12B, ~8GB) and Gemma 4 E4B (Q8) (4.5B, ~7.5GB). On this card Qwen3.5 9B Instruct (Q8) runs at est. 69 tok/s, typical error 5%. What it will not run: a 32B model does not fit in VRAM alone, so it spills to system RAM over PCIe and slows sharply.
$ollama run qwen3.5:9b-q8_0
- Top pick
- Qwen3.5 9B Instruct (Q8)
- Decode speed
- est. 69 tok/s, typical error 5%
- VRAM needed
- ~10.7 GB
Speeds are ModelFit estimates from memory bandwidth and model size, not measured benchmarks. Typical error: the median gap between the ModelFit speed model and measured runs on hardware the model was not fitted on, for this backend and model size.
Cite this page: ModelFit, RTX 4080 SUPER 16GB Local LLM (2026): 14B at est. 63 tok/s, https://modelfit.io/gpu/rtx-4080-super/, updated October 2026, CC BY 4.0.
Specs
VRAM
16 GB GDDR6X
Speed (8B Q4)
est. 102 tok/stypical error 5%
Bandwidth
736 GB/s
Architecture
Ada Lovelace
Max model size
Up to 14B parameter models
Compatibility
10 excellent, 0 workable
Table 1
RTX 4080 SUPER estimated tokens/sec by model size
Q4_K_M · ModelFit estimate| Model size | Est. speed | Fit on 16GB |
|---|---|---|
| 7B | est. 116 tok/stypical error 5% | Fits in VRAM |
| 14B | est. 63 tok/stypical error 15% | Fits in VRAM |
| 20B MoE (3.6B active) | est. 134 tok/stypical error 15% | Fits in VRAM |
| 32B | est. 6 tok/stypical error 5% | CPU offload (slow) |
| 35B MoE (3B active) | est. 54 tok/stypical error 15% | CPU offload (slow) |
| 70B | est. 1 tok/stypical error 10% | CPU offload (slow) |
| 120B MoE (5.1B active) | est. 35 tok/stypical error 15% | CPU offload (slow) |
Fig. 1RTX 4080 SUPER: estimated speed by model size
Decode speed, generic Q4_K_M model of each size at 4K context, 736 GB/s, typical error 5-15%
Fits in 16 GBCPU offload (slow)
7B116
14B63
20B MoE (3.6B active)134
32B6
35B MoE (3B active)54
70B1
120B MoE (5.1B active)35
tok/s est.ModelFit estimates, not measured benchmarks: decode speed of a generic Q4_K_M model of each size at 4K context on the RTX 4080 SUPER's 736 GB/s bandwidth. MoE rows scale by active parameters (decode reads only the active experts), so a 35B MoE runs far faster than a dense 32B. "CPU offload" sizes exceed the 16GB VRAM; dense models slow to a crawl there, MoE models degrade less because hot experts stay GPU-resident. What typical error means.
Fig. 2
Context costs VRAM too.
Qwen3.5 9B Instruct (Q8) loads ~10.7 GB of weights; at 16k context the KV cache adds ~0.5 GB (still fits the ~14 GB usable VRAM), and at 64k it adds ~2.0 GB (still fits).
Weights 11 GBKV cache · 16k context 0.5 GBHeadroom 3.2 GBRuntime reserve 1.6 GB
The model leaves about 3.2 GB of the usable VRAM budget free after weights and 16k context.
KV-cache figures assume an fp16 cache, the llama.cpp/Ollama default. Standard GQA models use a size-class estimate (8 KV heads x 128 head dim class); hybrid linear-attention models (Qwen3.5/3.6, Qwen3-Next) use the exact per-token cost from their published config, since only their sparse full-attention layers cache KV. A q8_0 KV cache roughly halves either figure. Estimates, not measurements.
Get told when a model beats Qwen3.5 9B Instruct (Q8) on your RTX 4080 SUPER
One email when a new open-weight model is a better fit for 16 GB of VRAM, plus the Thursday weekly.
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Where to buy the RTX 4080 SUPER
Storage & accessories for your model library
- Internal NVMe SSD · 2TBA Gen4 M.2 drive keeps your whole GGUF and quant collection on fast local storage, loading models straight off NVMe.Check price on Amazon
- USB4 NVMe Enclosure40Gbps external storage fast enough to run models from. Pair it with an M.2 drive for a portable model vault.Check price on Amazon
ModelFit may earn a commission on purchases through these links, at no extra cost to you.
RTX 4080 SUPER VRAM for AI: what actually fits?
16GB GDDR6X at 736 GB/s puts the 4080 SUPER near the top of 16GB cards in bandwidth. It handles 14B models at Q4 with ease (est. 63 tok/s, typical error 15%), and 8B models fly at est. 102 tok/s, typical error 5%. The 736 GB/s bandwidth sits between the 5070 Ti (896) and 4070 Ti SUPER (672). Even at its $999 launch MSRP the value proposition was weak against the RTX 5070 Ti; 2026 memory-shortage pricing has widened that gap rather than closed it. Best considered by existing owners or on the used market.
Table 2
RTX 4080 SUPER vs similar GPUs
| Hardware | Memory | Speed | Bandwidth |
|---|---|---|---|
| RTX 4070 Ti SUPER | 16 GB | est. 94 tok/stypical error 5% | 672 GB/s |
| RTX 4080 SUPER | 16 GB | est. 102 tok/stypical error 5% | 736 GB/s |
| RTX 5080 | 16 GB | est. 132 tok/stypical error 5% | 960 GB/s |
| RTX 4090 | 24 GB | est. 133 tok/stypical error 5% | 1008 GB/s |
Recommended models
registry-verified10 modelsGemma 4 12B
Local OKExcellentA · 44
Best for: Chat, Coding, Multimodal·Pop: 80/100
Perf: est. 73 tok/s, typical error 15% · first token ~0.3s
Fits in 16 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 4080 SUPER.
Gemma 4 E4B (Q8)
Local OKExcellentA · 48
Best for: On-device, Mobile, Chat·Pop: 82/100
Perf: est. 109 tok/s, typical error 5% · first token ~0.9s
Fits in 16 GB VRAM with room to spare. Best for on-device, mobile, chat on RTX 4080 SUPER.
Gemma 4 12B (Q8)
Local OKOKB · 11
Best for: Chat, Coding, Multimodal·Pop: 80/100
Perf: est. 46 tok/s, typical error 15% · first token ~1.4s
Fits in 16 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 4080 SUPER.
Ministral 3 14B
Local OKExcellentA · 37
Best for: Chat, Reasoning, Vision·Pop: 74/100
Perf: est. 63 tok/s, typical error 15% · first token ~0.3s
Fits in 16 GB VRAM with room to spare. Best for chat, reasoning, vision on RTX 4080 SUPER.
Qwen3.5 9B Instruct
Local OKExcellentA · 51
Best for: Quality, Coding, Reasoning·Pop: 86/100
Perf: est. 109 tok/s, typical error 5% · first token ~0.2s
Fits in 16 GB VRAM with room to spare. Best for quality, coding, reasoning on RTX 4080 SUPER.
LFM2 24B-A2B Instruct
Local OKOKB · 3
Best for: Local AI agents, privacy-first tool calling, MCP workflows·Pop: 80/100
Perf: est. 207 tok/s, typical error 15% · first token ~0.1s
Fits in 16 GB VRAM with room to spare. Best for local ai agents, privacy-first tool calling, mcp workflows on RTX 4080 SUPER.
Ornith 1.0 9B
Local OKExcellentA · 61
Best for: Agentic coding on small machines·Pop: 76/100
Perf: est. 109 tok/s, typical error 5% · first token ~0.2s
Fits in 16 GB VRAM with room to spare. Best for agentic coding on small machines on RTX 4080 SUPER.
Qwen3 14B
Local OKOKB · 24
Best for: Coding, Quality·Pop: 84/100
Perf: est. 63 tok/s, typical error 15% · first token ~0.3s
Fits in 16 GB VRAM with room to spare. Best for coding, quality on RTX 4080 SUPER.
LFM2.5 8B-A1B
Local OKExcellentA · 62
Best for: On-device agents, tool calling, multilingual chat·Pop: 72/100
Perf: est. 255 tok/s, typical error 15% · first token ~0.1s
Fits in 16 GB VRAM with room to spare. Best for on-device agents, tool calling, multilingual chat on RTX 4080 SUPER.
Similar GPUs for local AI
Speeds are estimates for Qwen3 8B Q4_K_M, typical error 5%.
Compatible model families
- QwenAlibaba Cloud: Qwen3.8 27B local flagship (Apache 2.0, vision)
- LlamaMeta: Most popular open-weight model family
- DeepSeekDeepSeek AI: V4.1 Flash: multimodal 552B MoE, API only
- MistralMistral AI: Excellent performance-per-parameter ratio
- GemmaGoogle DeepMind: Gemma 4 E4B: 4.5B effective, runs on 8GB Macs
- PhiMicrosoft: Best quality-per-gigabyte at small sizes
RTX 4080 SUPER FAQ: common questions
How much VRAM does the RTX 4080 SUPER have for LLMs?
The RTX 4080 SUPER has 16GB GDDR6X VRAM with 736 GB/s bandwidth. About 15.5GB is usable. Same capacity as the 4070 Ti SUPER and 5070 Ti, but higher bandwidth for faster inference.
What size LLM can I run on an RTX 4080 SUPER?
Up to 14B parameter models at Q4 quantization. This includes all popular 14B models like DeepSeek-R1 14B and Qwen 2.5 14B. At est. 63 tok/s, typical error 15% on 14B (est. 102 tok/s, typical error 5% on 8B), responses feel fast for chat workloads.
Is the RTX 4080 SUPER good for local AI?
It is excellent for AI performance but weak value even at its $999 launch MSRP: the RTX 5070 Ti costs less and is 23% faster (est. 125 tok/s, typical error 5% vs est. 102 tok/s on 8B). Current street pricing keeps that gap wide. Buy the 4080 SUPER only used, well below the price of a new 5070 Ti.
Should I upgrade from RTX 4080 SUPER to RTX 5080?
Both launched at the same $999 MSRP. The RTX 5080 is 29% faster (est. 132 tok/s, typical error 5% vs est. 102 tok/s on 8B), and the 4080 SUPER's street price has since climbed well above MSRP amid the 2026 memory shortage. Both have 16GB VRAM. If you already own the 4080 SUPER, weigh that gain against the cost of swapping cards. If buying new, the 5080 is the better starting point.
How fast is a 27B-class model on the RTX 4080 SUPER?
The RTX 4080 SUPER's 16GB of VRAM cannot fit a 32B model comfortably. The largest size class it fits is 20B MoE (3.6B active), at est. 134 tok/s, typical error 15%. The largest individual model in the catalog that fits is LFM2 24B-A2B Instruct (24B).
Related guides & benchmarks
Sizing local AI? Start with RAM & VRAM
Browse all NVIDIA GPUs for AI
Real-world calibration
Have this hardware? Share a measured run.
Measure a fixed Ollama reference model on your machine. This separate, opt-in contribution does not change ModelFit's third-party SWE-Bench leaderboard; results appear on the measured-speed leaderboard after the next site update.
$ npx -y @wecko-ai/modelfit bench --submit --cleanup
Want personalized recommendations?
Use our interactive wizard to compare models across Apple Silicon and NVIDIA GPUs.