Gemma 4 26B-A4B
Gemma / 26B / Q4_K_M / ~16 GB
Best for: Chat, Coding, Multimodal·Pop: 86/100
Perf: ~85 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 4090.
The RTX 4090 pairs 24GB of GDDR6X with 1,008 GB/s of memory bandwidth. It runs 8B chat models at an estimated 104 tok/s and 32B reasoning models at an estimated 32 tok/s. Only the RTX 5090 is faster among consumer cards.
The best local LLM for the RTX 4090 is Gemma 4 26B-A4B at ~85 tok/s on its 24GB VRAM. It uses ~16GB of VRAM; the RTX 4090 handles up to 32B parameter models at Q4. A 32B model at Q4 runs at ~32 tok/s.
Sizing rule: a Q4 model needs about 0.6 GB of VRAM per billion parameters, and ModelFit budgets 90% of this card's 24GB for weights, context, and KV-cache. The per-size table below uses that same budget. Other strong fits: Qwen3.8 27B (27B, ~16.5GB) and Qwen3.5 27B Instruct (27B, ~16GB). Gemma 4 26B-A4B runs at an estimated 85 tok/s on this card. What it will not run: a 70B model does not fit in VRAM alone, so it spills to system RAM over PCIe and slows sharply.
Speeds are ModelFit estimates from memory bandwidth and model size, not measured benchmarks.
Cite this page: ModelFit, RTX 4090 24GB Local LLM (2026): Runs 32B, Top Pick ~85 tok/s, https://modelfit.io/gpu/rtx-4090/, updated September 2026, CC BY 4.0.
Last updated: September 3, 2026 · Editor: ModelFit Team
*Launch MSRP was $1,599; the card is now discontinued and current market pricing runs well above that
| Model Size | Est. Speed | Fit on 24GB |
|---|---|---|
| 7B | ~117 tok/s | Fits in VRAM |
| 14B | ~65 tok/s | Fits in VRAM |
| 20B MoE (3.6B active) | ~99 tok/s | Fits in VRAM |
| 32B | ~32 tok/s | Fits in VRAM |
| 35B MoE (3B active) | ~84 tok/s | Fits in VRAM |
| 70B | ~2 tok/s | CPU offload (slow) |
| 120B MoE (5.1B active) | ~14 tok/s | CPU offload (slow) |

ModelFit estimates, not measured benchmarks: anchored to an 8B-class Q4_K_M model at 16K context on the RTX 4090's 1008 GB/s bandwidth, then scaled by model size. 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 24GB VRAM; dense models slow to a crawl there, MoE models degrade less because hot experts stay GPU-resident.
Context costs VRAM too. Gemma 4 26B-A4B loads ~16 GB of weights; at 16k context the KV cache adds ~4.0 GB (still fits the ~22 GB usable VRAM), and at 64k it adds ~16.0 GB (exceeds the budget, use a smaller quant or a q8_0 KV cache).
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.
A Gen4 M.2 drive keeps your whole GGUF and quant collection on fast local storage, loading models straight off NVMe.
Check price on Amazon40Gbps external storage fast enough to run models from. Pair it with an M.2 drive for a portable model vault.
Check price on AmazonModelFit may earn a commission on purchases through these links, at no extra cost to you. Prices shown are approximate street references.
24GB GDDR6X at 1,008 GB/s gives the RTX 4090 real headroom. 32B models at Q4 (~20GB) load fully with 3GB left for KV cache. 14B models at Q5 or Q6 fit easily for higher quality inference. At 104 tok/s with 8B models, the 4090 delivers near-instant responses. The only consumer card faster is the RTX 5090 (145 tok/s, 32GB). Among 24GB cards, its 1,008 GB/s bandwidth is the highest ModelFit tracks.
Gemma / 26B / Q4_K_M / ~16 GB
Best for: Chat, Coding, Multimodal·Pop: 86/100
Perf: ~85 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 4090.
Qwen / 27B / Q4_K_M / ~16.5 GB
Best for: Coding, Agent, Vision, Long context·Pop: 95/100
Perf: ~37 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for coding, agent, vision, long context on RTX 4090.
Qwen / 27B / Q4_K_M / ~16 GB
Best for: Chat, Coding, Complex reasoning·Pop: 82/100
Perf: ~37 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for chat, coding, complex reasoning on RTX 4090.
GPT-OSS / 21B / MXFP4 / ~13.8 GB
Best for: Chat, Coding, Reasoning·Pop: 85/100
Perf: ~87 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for chat, coding, reasoning on RTX 4090.
Qwen / 27B / Q4_K_M / ~18 GB
Best for: Coding, Quality, Long context·Pop: 92/100
Perf: ~37 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for coding, quality, long context on RTX 4090.
LFM2 / 24B / Q4_K_M / ~14 GB
Best for: Local AI agents, privacy-first tool calling, MCP workflows·Pop: 80/100
Perf: ~118 tok/s · first token ~0.3s
Fits in 24 GB VRAM with room to spare. Best for local ai agents, privacy-first tool calling, mcp workflows on RTX 4090.
Gemma / 12B / Q8_0 / ~12.8 GB
Best for: Chat, Coding, Multimodal·Pop: 80/100
Perf: ~46 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for chat, coding, multimodal on RTX 4090.
Qwen / 35B / Q4_K_M / ~20 GB
Best for: Reasoning, Coding, Agent scenarios·Pop: 90/100
Perf: ~84 tok/s · first token ~1.0s
Fits in 24 GB VRAM with room to spare. Best for reasoning, coding, agent scenarios on RTX 4090.
Qwen / 14B / Q4_K_M / ~11 GB
Best for: Coding, Quality·Pop: 84/100
Perf: ~65 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for coding, quality on RTX 4090.
Qwen / 14B / Q8_0 / ~15.9 GB
Best for: Coding, Quality·Pop: 84/100
Perf: ~40 tok/s · first token ~0.4s
Fits in 24 GB VRAM with room to spare. Best for coding, quality on RTX 4090.
The RTX 4090 tops out around up to 32b parameter models. For anything bigger, an hourly rented GPU runs the same open weights with the same Ollama workflow, billed by the hour, no hardware purchase needed.
ModelFit may earn a commission on sign-ups made through these links, at no extra cost to you.
Alibaba Cloud: Widest size range (0.5B to 235B)
LlamaMeta: Most popular open-weight model family
DeepSeekDeepSeek AI: Best-in-class reasoning with R1 models
MistralMistral AI: Excellent performance-per-parameter ratio
GemmaGoogle DeepMind: Excellent quality at small sizes (1B-9B)
PhiMicrosoft: Best quality-per-gigabyte at small sizes
The RTX 4090 has 24GB GDDR6X VRAM with 1,008 GB/s bandwidth. About 23GB is usable for models. It runs 32B models at Q4 with room for 8K+ context windows.
Up to 32B parameter models at Q4 quantization. This includes DeepSeek-R1 32B, Qwen 2.5 32B, and larger reasoning models. For 70B models, you need Q2 or dual GPUs, or step up to the RTX 5090.
Rarely. The card is discontinued and its street price has climbed well past launch MSRP, while a used RTX 3090 offers 83% of the speed with the same 24GB VRAM for a fraction of the money. The 4090 is worth it if you also game at 4K or need the absolute fastest 24GB card. For pure AI value, the 3090 wins.
The RTX 5090 is 39% faster on 8B models (an estimated 145 vs 104 tok/s) with 8GB more VRAM (32 vs 24GB), enabling 70B models. The 5090 also costs more in 2026 now that the 4090 is discontinued and only scalped stock remains. If buying new, the 5090 is the clear choice.
Not at full quality. A 70B Q4 model needs ~42GB VRAM. The 4090 has 24GB, so you would need Q2 quantization (lower quality) or run with partial CPU offloading (much slower). For 70B, the RTX 5090 (32GB) is recommended.
ModelFit estimates a 32B model on the RTX 4090 runs at roughly 32 tok/s at Q4_K_M. The current 27B-class pick in the catalog is Qwen3.8 27B (ollama run qwen3.8:27b).
New open-weight models, real Apple Silicon benchmarks, and the one model worth running on your Mac this week. Free, one email a week, unsubscribe anytime.
By subscribing you agree to our Privacy Policy and to receive the weekly email. Unsubscribe anytime.
Use our interactive wizard to compare models across Apple Silicon and NVIDIA GPUs.