ai.canshu.com
中文
VRAM Calculator Advisor & TCO API Pricing Compare Models Leaderboard GPUs

Search models, GPUs, and tools

↑↓ Navigate · Enter Open · ★ Watchlist prioritized ai.canshu.com
Q4_K_M: Partial RAM Offload (Slow) 23.34 GB / 15.7 GB Usable

Qwen3.5-35B-A3B (MoE) on NVIDIA GeForce RTX 4070 Ti Super 16GB: VRAM Requirement & Speed Estimate

A single RTX 4070 Ti Super (15.7 GB usable VRAM) cannot fully fit Qwen3.5-35B-A3B (MoE) at Q4_K_M (23.34 GB required). Consider multi-GPU tensor parallelism, a smaller parameter variant, or cloud GPU / API deployment.

Concurrency (Batch):
MoE 35B / A3B·40L (d=2048)·16Q/2KV·Ctx 256K·Text+Image+Video
Runnable Models →
GPUs
16 GB VRAM·672 GB/s·Safe Usable 15.7 GB
Partial RAM Offload (Slow)23.34 / 15.7 GB~79.7 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Partial RAM Offload (Slow)
23.34/ 15.7 GB usable

Physical 16 GB · Load 148.9%

Theoretical Decode Speed
~79.7tokens/s

MoE: reads only 1.82 GB active weights/step

Max Context at Batch=1
0K/ 256K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 21.22 GBKV Cache (1×16K): 0.4 GBOverhead: 1.72 GB
🚀 Blazing Fast (Instant CoT)~6.3s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3.5:35b-a3b

Insufficient Local VRAM? Ways to Run Qwen3.5-35B-A3B (MoE)

Rent an hourly high-VRAM cloud GPU pod, or check the 12-month Buy vs Cloud vs API break-even curve.

📊 All Quantization Levels for Qwen3.5-35B-A3B (MoE) on RTX 4070 Ti Super

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 73.1 GB Out of VRAM (OOM) — —
FP8 / Q8_0 39.63 GB Out of VRAM (OOM) — —
GGUF Q6_K 30.98 GB Out of VRAM (OOM) — —
GGUF Q5_K_M 27.09 GB Out of VRAM (OOM) — —
Q4_K_M / AWQ 23.34 GB Partial RAM Offload (Slow) ~79.7 tok/s —
GGUF IQ4_XS 20.67 GB Partial RAM Offload (Slow) ~108.9 tok/s —
GGUF Q3_K_M 19.15 GB Partial RAM Offload (Slow) ~137.7 tok/s —
GGUF Q2_K 15.76 GB Partial RAM Offload (Slow) ~339.7 tok/s 12K

💡 Recommended Alternative GPUs & Macs for Qwen3.5-35B-A3B (MoE)

Frequently Asked Questions

Can NVIDIA GeForce RTX 4070 Ti Super 16GB run Qwen3.5-35B-A3B (MoE) locally?

A single RTX 4070 Ti Super (15.7 GB usable VRAM) cannot fully fit Qwen3.5-35B-A3B (MoE) at Q4_K_M (23.34 GB required). Consider multi-GPU tensor parallelism, a smaller parameter variant, or cloud GPU / API deployment.

What is the best quantization for Qwen3.5-35B-A3B (MoE) on RTX 4070 Ti Super?

Qwen3.5-35B-A3B (MoE) (35B parameters) exceeds single-device VRAM even at aggressive 2–3 bit quantization. Cloud GPUs or hosted APIs are recommended.

How much VRAM does Qwen3.5-35B-A3B (MoE) need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), Qwen3.5-35B-A3B (MoE) requires 73.1 GB VRAM (70 GB for weights alone). Because it uses a Mixture-of-Experts (MoE) architecture, all 35B parameters must sit in memory, while only 3B active parameters are read per token.