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) 19.83 GB / 15.2 GB Usable

Qwen3.8-27B on 摩尔线程 MTT S80 16GB: VRAM Requirement & Speed Estimate

A single 摩尔线程 S80 16G (15.2 GB usable VRAM) cannot fully fit Qwen3.8-27B at Q4_K_M (19.83 GB required). Consider multi-GPU tensor parallelism, a smaller parameter variant, or cloud GPU / API deployment.

Concurrency (Batch):
Dense 27.8B·64L (d=5120)·24Q/4KV·Ctx 256K·Text+Image+Video
Runnable Models →
GPUs
16 GB VRAM·448 GB/s·Safe Usable 15.2 GB
Partial RAM Offload (Slow)19.83 / 15.2 GB~9.8 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)
19.83/ 15.2 GB usable

Physical 16 GB · Load 130.5%

Theoretical Decode Speed
~9.8tokens/s

Actual speed varies ~±25% by runtime

Max Context at Batch=1
0K/ 256K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 16.85 GBKV Cache (1×16K): 1.34 GBOverhead: 1.64 GB
🚶 Readable Pace~51s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run qwen3-8-27b

Insufficient Local VRAM? Ways to Run Qwen3.8-27B

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 Qwen3.8-27B on 摩尔线程 S80 16G

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 59.35 GB Out of VRAM (OOM) — —
FP8 / Q8_0 32.77 GB Out of VRAM (OOM) — —
GGUF Q6_K 25.89 GB Out of VRAM (OOM) — —
GGUF Q5_K_M 22.81 GB Partial RAM Offload (Slow) ~7 tok/s —
Q4_K_M / AWQ 19.83 GB Partial RAM Offload (Slow) ~9.8 tok/s —
GGUF IQ4_XS 17.7 GB Partial RAM Offload (Slow) ~13.7 tok/s —
GGUF Q3_K_M 16.5 GB Partial RAM Offload (Slow) ~17.5 tok/s —
GGUF Q2_K 13.8 GB Tight Fit (Watch Context) ~25.5 tok/s 36K

💡 Recommended Alternative GPUs & Macs for Qwen3.8-27B

Frequently Asked Questions

Can 摩尔线程 MTT S80 16GB run Qwen3.8-27B locally?

A single 摩尔线程 S80 16G (15.2 GB usable VRAM) cannot fully fit Qwen3.8-27B at Q4_K_M (19.83 GB required). Consider multi-GPU tensor parallelism, a smaller parameter variant, or cloud GPU / API deployment.

What is the best quantization for Qwen3.8-27B on 摩尔线程 S80 16G?

The highest-quality quantization that fits cleanly on 摩尔线程 S80 16G is GGUF Q2_K (Extreme Compression), using 13.8 GB VRAM at ~25.5 tokens/s with up to 36K context.

How much VRAM does Qwen3.8-27B need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), Qwen3.8-27B requires 59.35 GB VRAM (55.6 GB for weights alone).