Deploy GPU ↗
HARDWARE FIT CHECK • Qwen 3.8 235B-A22B MoE + 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)150.54 GB / 48 GB VRAM

Can I Run Qwen 3.8 235B-A22B MoE on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)?

Full VRAM footprint analysis, tokens/sec generation benchmark, and copy-paste Ollama & vLLM runtime commands for running Qwen 3.8 235B-A22B MoE (235B) on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM) (48 GB).

✕ OOM150.54 GB (314%)0 t/s

Target Hardware & Model

interactive
GPU Topology / Quantity2x Dual-GPU Rig (48GB Pool)
Pre-configured 2x GPU tensor-parallel topology (48GB total VRAM, 1648 GB/s aggregate bandwidth)
Frontier 235B sparse MoE reasoning & repository synthesis for 128GB–256GB unified memory and multi-GPU clusters
Qwen 3.8 235B-A22B MoE on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)MoE (22B Active / 235B Total)

✕ Out of Memory — Exceeds VRAM Capacity

✕ OOM
Required VRAM
150.54 GB
314% of hardware capacity (48 GB)
Est. Decode Speed
0 tok/s
22B active params @ 1648 GB/s
Available Headroom
-102.54 GB
Needs +102.54 GB more VRAM
Memory Allocation & Headroom
150.54 GB allocated of 48 GB capacity
314% utilized
Weights: 141 GBMoE (22B Active)
KV: (32K): 3.15 GB
Overhead: 6.39 GB
Deficit: -102.54 GB
Architecture & Throughput Transparency: Single-user decode speed (~0 tok/s) is memory-bandwidth bound at batch size = 1 on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM) (1648 GB/s). Prompt prefill / TTFT (~0 tok/s) is compute-bound and utilizes GPU Tensor cores. Actual latency varies with FlashAttention-2 and runtime engine (Ollama/vLLM/llama.cpp).
MoE Memory vs Speed Architecture: While decode throughput scales with the active parameter count (22B parameters streamed per token), all 235B parameters must be resident in VRAM so the dynamic router can route tokens to any expert layer.

Local Hardware Configurations & Pricing

Out of VRAM: High-Capacity Hardware Upgrades
High-bandwidth VRAM specs and current market prices
Consumer GPU
32 GB VRAM

NVIDIA GeForce RTX 5090 (32GB GDDR7)

Maximum tokens/sec on 27B–35B models & Blackwell FP4 quantization

1792 GB/s575W TDP
$1,999Amazon Prime →
Multi-GPU Rig
48 GB VRAM

2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)

Best price-to-VRAM homelab rig for Llama 3.3 70B Q4_K_M & Qwen 2.5 Coder 32B Q8_0

1648 GB/s700W TDP
$1,550Amazon Prime →
Apple Silicon↑ Upgrade
54 GB VRAM

Apple Mac Mini M4 Pro (64GB Unified Memory)

Silent 65W desktop running 32B–70B models with large KV context

273 GB/s65W TDP
$1,999Amazon Prime →
Apple Silicon↑ Upgrade
112 GB VRAM

Apple Mac Studio M4 Max (128GB Unified Memory)

Running Mistral Large 2 123B & Llama 3.3 70B Q8_0 locally with 128K context

546 GB/s145W TDP
$3,499Amazon Prime →

Out of VRAM? Don't downgrade model quality.

Out of VRAM

Spin up an on-demand cloud GPU instance in 60 seconds with pre-installed Ollama and vLLM. (Qwen 3.8 235B-A22B MoE @ 150.54 GB VRAM)

RunPod Cloud GPUsBest 1-Click Ollama & vLLM
$0.38/hr (RTX 4090 Spot) – $2.39/hr (H100)
Launch 1-Click GPU Pod →
Vast.ai GPU MarketplaceCheapest Dual RTX 3090 / 4090
$0.18/hr (RTX 3060) – $0.78/hr (2x RTX 3090)
Rent Spot GPU →
Lambda GPU CloudEnterprise 8x H100 Clusters
$2.49/hr (H100) – $19.92/hr (8x H100)
Check H100 Availability →

Quantization Precision Matrix

32K ctx • KV: Q8_0
QuantizationWeightsKV CacheTotal VRAMEst. SpeedFit Status
FP1616 bpw470 GB3.15 GB491.05 GB (999%)0 t/sOOM
Q8_08.5 bpw249.1 GB3.15 GB262.42 GB (547%)0 t/sOOM
Q6_K6.56 bpw192.7 GB3.15 GB204.04 GB (425%)0 t/sOOM
Q5_K_M5.68 bpw166.85 GB3.15 GB177.29 GB (369%)0 t/sOOM
Q4_K_M4.8 bpw141 GB3.15 GB150.54 GB (314%)0 t/sOOM
FP44.25 bpw124.55 GB3.15 GB133.51 GB (278%)0 t/sOOM
IQ3_XS3.3 bpw96.35 GB3.15 GB104.32 GB (217%)0 t/sOOM

Export Full Architecture & Infrastructure Spec

Download an instant Markdown architecture spec with exact memory math, hardware benchmarks, and runtime CLI flags.

FAQ & Reference

Frequently Asked Questions: Qwen 3.8 235B-A22B MoE on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)

How much VRAM does Qwen 3.8 235B-A22B MoE need on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)?

At standard 4-bit quantization (Q4_K_M) with 32K context, Qwen 3.8 235B-A22B MoE requires 150.54 GB of total VRAM (weights + KV cache + CUDA runtime overhead).

What inference speed will Qwen 3.8 235B-A22B MoE achieve on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)?

Based on 2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)'s 1648 GB/s memory bandwidth, expected decode speed is approximately 0 tokens per second.

Can I expand context beyond 32K tokens?

Each additional 32K tokens of context requires additional KV cache memory. For longer context windows, consider 8-bit or 4-bit KV cache quantization.