Can you run Qwen3.8 27B (Q8) on RTX 3090?

Qwen3.8 27B (Q8) Q8_0 on the NVIDIA GeForce RTX 3090: verdict, VRAM math and estimated speed.

Yes, but slow
Quick answer

Yes, but slowly: Qwen3.8 27B (Q8) spills past the RTX 3090. 27.1 GB weights at Q8_0 vs 21.6 GB usable VRAM; ~4 tok/s est. (ModelFit, 2026).

$ollama run qwen3.8:27b-q8_0
VERDICT
Partial offload
EST. SPEED
~4 tok/s
WEIGHTS
27.1 GB Q8_0

VRAM math and speed are ModelFit engine estimates, not measurements. Commands are registry-verified Ollama tags.

Cite this page: ModelFit, Qwen3.8 27B (Q8) on RTX 3090, https://modelfit.io/can-i-run/qwen3.8-27b-q8-on-rtx-3090/, updated September 2026, CC BY 4.0.

Last updated: September 24, 2026 · Editor: ModelFit Team

VRAM
24 GB (21.6 usable)
Model weights
27.1 GB Q8_0
Est. speed
~4 tok/s
First token
~1.6s
Fit grade
C · 0

What limits this combo

The four constraints the fit engine checks for Qwen3.8 27B (Q8) on the RTX 3090, in order of what usually breaks first.

Weights fitBlocked27.1 GB of weights against 21.6 GB usable: the model does not load fully in VRAM; the shortfall spills to system RAM.
Context ceilingBlockedThe weights alone exceed the budget, so no usable context fits on top.
SpeedBlocked~4 tok/s (est.): single-digit-to-low token rates make interactive use painful; this combo is capacity, not speed.
Use-case ceilingTightStrongest fit: Chat (D). Agentic coding caps at D, either the speed floor or the model's tuning is the limit; see the workload table for the per-case reason.

Where the RTX 3090 sits for Qwen3.8 27B (Q8)

The RTX 3090 is the tightest way to run Qwen3.8 27B (Q8): it loads with partial offload at ~4 tok/s est.. The cheapest card that runs it without spilling into system RAM is the RTX 5090 (32 GB, ~32 tok/s est.). Context ceiling on this card: none; on the RTX 5090: 16k.

CardVRAMUsableVerdictEst. speedMax context
RTX 40608 GB7.2 GBNo~1 tok/snone
RTX 306012 GB10.8 GBNo~1 tok/snone
RTX 407012 GB10.8 GBNo~1 tok/snone
RTX 4060 Ti16 GB14.4 GBNo~1 tok/snone
RTX 5070 Ti16 GB14.4 GBNo~2 tok/snone
RTX 4070 Ti SUPER16 GB14.4 GBNo~2 tok/snone
RTX 508016 GB14.4 GBNo~2 tok/snone
RTX 3090 (this card)24 GB21.6 GBTight~4 tok/snone
RTX 409024 GB21.6 GBTight~5 tok/snone
RTX 509032 GB28.8 GBYes~32 tok/s16k

Same engine as the verdict above: 90% of VRAM usable, KV-cache at fp16, bandwidth-derived tok/s. Verdicts of the other cards are computed for Qwen3.8 27B (Q8) Q8_0 exactly.

Quant explorer

Switch between the quality-gated builds tracked for this exact combo. Q4_K_M stays the default recommendation; heavier Q6/Q8 builds only appear when their exact Ollama tags are registry-verified.

ModelFit engine estimate
Verdict
Slow
Est. speed
~4 tok/s
First token
~1.6s · Fast
Safe context
n/a
C · 0 Tight fit

Why no 2-bit builds: Q1/Q2-class quants can look attractive in a memory table, but their quality loss is large enough that ModelFit excludes them from rankings instead of inflating the catalog with junk options. Usable VRAM: 21.6 GB.

Workload verdicts

Qwen3.8 27B (Q8) on the RTX 3090, graded per use case from the model's registry-verified tuning and the engine's speed estimate for this exact combo.

ModelFit engine estimate
ChatD~4 tok/s est. vs ~20 needed for chat; partial offload slows everything.
CodingD~4 tok/s est. vs ~15 needed for coding; partial offload slows everything.
Agentic codingD~4 tok/s est. vs ~30 needed for agentic coding; partial offload slows everything.
ReasoningD~4 tok/s est. vs ~12 needed for reasoning; not a reasoning-tuned model; partial offload slows everything.
RAGD~4 tok/s est. vs ~20 needed for rag; 32k context does not fit; partial offload slows everything.

Grades combine tag-verified tuning (a model not built for the workload caps at C) with the tok/s floor each workload needs to feel usable. A combo that partially offloads caps at C; one that does not fit is D everywhere.

Memory math: weights + context vs budget

Weights take 27.1 GB. Context costs extra KV-cache on top; this is where long-context sessions break on cards that technically fit the weights.

Weights 27 GBKV cache · 16k context 1.0 GBRuntime reserve 2.4 GB

This configuration exceeds the comfortable VRAM budget by about 6.5 GB. Expect offload pressure or a shorter safe context.

ContextKV-cacheTotalFits
8k tokens0.5 GB27.6 GBOver
16k tokens1.0 GB28.1 GBOver
32k tokens2.0 GB29.1 GBOver
64k tokens4.0 GB31.1 GBOver
128k tokens8.0 GB35.1 GBOver

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.

See how fast it feels

A deterministic typing simulation for Qwen3.8 27B (Q8): first token ~1.6s (fast prefill), then ~4 tokens/sec.

Start the simulation to preview the response pace.

ModelFit engine estimate, not a measured benchmark. Real speed varies with prompt length, thermals, runtime, and KV-cache settings.

Upgrade path

The cheapest tracked card that runs Qwen3.8 27B (Q8) comfortably is the AMD Ryzen AI Max+ 395 (Strix Halo) (110 GB unified memory).

See the Ryzen AI Max+ 395 page

Qwen3.8 27B (Q8) on RTX 3090: FAQ

Can the NVIDIA GeForce RTX 3090 run Qwen3.8 27B (Q8)?

Barely. Qwen3.8 27B (Q8) (Q8_0) needs about 27.1 GB but the RTX 3090 has 21.6 GB usable VRAM, so part of the model spills to system RAM and speed drops to at roughly 4 tokens/sec (est.).

How much VRAM does Qwen3.8 27B (Q8) need?

About 27.1 GB for the weights at Q8_0, plus KV-cache for context: roughly 1.0 GB extra at 16k tokens. The RTX 3090 budget is 21.6 GB (24 GB x 90%).

What is the best quantization of Qwen3.8 27B (Q8) for the RTX 3090?

The Q4_K_M build is the highest quality that fits (16.5 GB vs 21.6 GB usable). The Q8_0 build at 27.1 GB leaves more room for long context.

What GPU do I need to run Qwen3.8 27B (Q8) comfortably?

The cheapest tracked card that runs Qwen3.8 27B (Q8) (Q8_0) fully in VRAM is the AMD Ryzen AI Max+ 395 (Strix Halo) (110 GB).

How much context can Qwen3.8 27B (Q8) use on the RTX 3090?

None worth having: the 27.1 GB of weights alone exceed the 21.6 GB usable VRAM, so the model only runs with system-RAM offload and short prompts.

Is Qwen3.8 27B (Q8) on the RTX 3090 fast enough for coding agents?

No: the engine estimates ~4 tok/s against the ~30 tok/s an agentic loop needs to feel responsive. Chat will feel fine, but multi-step agent runs will drag; a smaller model or a bigger card is the fix.