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: Smooth Fit 12.29 GB / 15.2 GB Usable

Qwen3-14B on 摩尔线程 MTT S80 16GB: VRAM Requirement & Speed Estimate

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Qwen3-14B requires 12.29 GB of VRAM (8.97 GB weights + 2.68 GB KV cache + 0.63 GB overhead), fitting within 摩尔线程 S80 16G's 15.2 GB usable memory at ~31.3 tokens/s.

Concurrency (Batch):
Dense 14.8B·40L (d=5120)·40Q/8KV·Ctx 128K
Runnable Models →
GPUs
16 GB VRAM·448 GB/s·Safe Usable 15.2 GB
Smooth Fit12.29 / 15.2 GB~31.3 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
12.29/ 15.2 GB usable

Physical 16 GB · Load 80.8%

Theoretical Decode Speed
~31.3tokens/s

Actual speed varies ~±25% by runtime

Max Context at Batch=1
33K/ 128K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 8.97 GBKV Cache (1×16K): 2.68 GBOverhead: 0.63 GB
🏃 Smooth & Snappy~16s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3:14b

摩尔线程 S80 16G Fits Qwen3-14B · Cost & TCO

Approx. $239 (~$15/GB VRAM) · 255W TDP

📊 All Quantization Levels for Qwen3-14B on 摩尔线程 S80 16G

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 33.33 GB Out of VRAM (OOM) — —
FP8 / Q8_0 19.17 GB Partial RAM Offload (Slow) ~10.6 tok/s —
GGUF Q6_K 15.51 GB Partial RAM Offload (Slow) ~22.3 tok/s 14K
GGUF Q5_K_M 13.87 GB Tight Fit (Watch Context) ~25 tok/s 23K
Q4_K_M / AWQ 12.29 GB Smooth Fit ~31.3 tok/s 33K
GGUF IQ4_XS 11.15 GB Smooth Fit ~35 tok/s 40K
GGUF Q3_K_M 10.51 GB Smooth Fit ~37.6 tok/s 43K
GGUF Q2_K 9.08 GB Smooth Fit ~45 tok/s 52K

💡 Recommended Alternative GPUs & Macs for Qwen3-14B

Frequently Asked Questions

Can 摩尔线程 MTT S80 16GB run Qwen3-14B locally?

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Qwen3-14B requires 12.29 GB of VRAM (8.97 GB weights + 2.68 GB KV cache + 0.63 GB overhead), fitting within 摩尔线程 S80 16G's 15.2 GB usable memory at ~31.3 tokens/s.

What is the best quantization for Qwen3-14B on 摩尔线程 S80 16G?

The highest-quality quantization that fits cleanly on 摩尔线程 S80 16G is GGUF Q5_K_M (High Balanced Quality), using 13.87 GB VRAM at ~25 tokens/s with up to 23K context.

How much VRAM does Qwen3-14B need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), Qwen3-14B requires 33.33 GB VRAM (29.6 GB for weights alone).