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 22.47 GB / 61.4 GB Usable

Gemma-4-31B-IT on 华为昇腾 Ascend 910B 64GB (HBM2e): VRAM Requirement & Speed Estimate

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Gemma-4-31B-IT requires 22.47 GB of VRAM (18.61 GB weights + 2.18 GB KV cache + 1.67 GB overhead), fitting within 昇腾 910B 64G's 61.4 GB usable memory at ~43.9 tokens/s.

Concurrency (Batch):
Dense 30.7B·60L (d=5376)·32Q/4KV·Ctx 256K·Text+Image+Video
Runnable Models →
GPUs
64 GB VRAM·1200 GB/s·Safe Usable 61.4 GB
Smooth Fit22.47 / 61.4 GB~43.9 tok/s
3. Weight Quantization
4. Per-Request Context16K (16,384 tok)
Verified: 2026-11-20·Roofline bandwidth/compute estimate (Batch=1)
Smooth Fit
22.47/ 61.4 GB usable

Physical 64 GB · Load 36.6%

Theoretical Decode Speed
~43.9tokens/s

Actual speed varies ~±25% by runtime

Max Context at Batch=1
256K/ 256K

Enable Q8/Q4 KV quantization to scale batch concurrency

Weights: 18.61 GBKV Cache (1×16K): 2.18 GBOverhead: 1.67 GB
🏃 Smooth & Snappy~11.4s for a 500-token response
Reference Launch CLI:
OLLAMA_CONTEXT_LENGTH=16384 ollama run gemma4:31b

昇腾 910B 64G Fits Gemma-4-31B-IT · Cost & TCO

Approx. $13,200 (~$206/GB VRAM) · 310W TDP

📊 All Quantization Levels for Gemma-4-31B-IT on 昇腾 910B 64G

Quantization Total VRAM (16K) Verdict Est. Speed Max Context
FP16 / BF16 66.11 GB Partial RAM Offload (Slow) ~7.6 tok/s —
FP8 / Q8_0 36.75 GB Smooth Fit ~25.6 tok/s 256K
GGUF Q6_K 29.16 GB Smooth Fit ~32.9 tok/s 256K
GGUF Q5_K_M 25.75 GB Smooth Fit ~37.7 tok/s 256K
Q4_K_M / AWQ 22.47 GB Smooth Fit ~43.9 tok/s 256K
GGUF IQ4_XS 20.12 GB Smooth Fit ~49.7 tok/s 256K
GGUF Q3_K_M 18.79 GB Smooth Fit ~53.7 tok/s 256K
GGUF Q2_K 15.81 GB Smooth Fit ~65.6 tok/s 256K

💡 Recommended Alternative GPUs & Macs for Gemma-4-31B-IT

Frequently Asked Questions

Can 华为昇腾 Ascend 910B 64GB (HBM2e) run Gemma-4-31B-IT locally?

Yes. At Q4_K_M 4-bit quantization with a 16K context window, Gemma-4-31B-IT requires 22.47 GB of VRAM (18.61 GB weights + 2.18 GB KV cache + 1.67 GB overhead), fitting within 昇腾 910B 64G's 61.4 GB usable memory at ~43.9 tokens/s.

What is the best quantization for Gemma-4-31B-IT on 昇腾 910B 64G?

The highest-quality quantization that fits cleanly on 昇腾 910B 64G is FP8 / Q8_0 (99.9% Near-Lossless), using 36.75 GB VRAM at ~25.6 tokens/s with up to 256K context.

How much VRAM does Gemma-4-31B-IT need in full FP16 precision?

In unquantized FP16/BF16 precision (16K context), Gemma-4-31B-IT requires 66.11 GB VRAM (61.4 GB for weights alone).