GLM-4-Voice-9B VRAM & GPU Requirement Calculator
Interactive hardware sizing for GLM-4-Voice-9B (9.4B by Zhipu AI). Estimated 8K context footprint: 6.9 GB (INT4) • 11.7 GB (FP8) • 21.2 GB (FP16).
GPUs
Memory Requirements 21.23 GB
Requires 1 GPUs (based on memory capacity)
20 GB
All model weights
0.31 GB
Conversation history cache
0.8 GB
Concurrent processing cache
0.12 GB
Temporary computation cache
Performance Metrics
Visualize 46 tok/s →Total Throughput
46 tokens/s
Throughput Per User
46 tokens/s/user
Estimated Latency
3291 ms
Max QPS
0.3 queries/s
Estimated using memory bandwidth and tensor compute throughput across 1x NVIDIA RTX 4090 (24GB).
Quick Deploy CLI Command
vllm serve THUDM/glm-4-voice-9b --max-model-len 8192
Flags reflect your selected precision, context length, and tensor parallel GPU count.Open GLM-4-Voice-9B spec page →
GLM-4-Voice-9B Architecture & Memory Profile
Total Parameters
9.4B
Architecture
Dense Transformer
Developer / Provider
Zhipu AI
Recommended Precision
FP16
GLM-4-Voice-9B VRAM Matrix by Precision & Context Length
Precomputed single-user memory requirements (model weights + KV cache + runtime overhead).
| Precision | 8K Context | 32K Context | 128K Context |
|---|---|---|---|
| FP16 | 21.2 GB | 22.2 GB | 25.9 GB |
| FP8 | 11.7 GB | 12.1 GB | 14.0 GB |
| INT8 | 11.7 GB | 12.1 GB | 14.0 GB |
| INT4 | 6.9 GB | 7.1 GB | 8.1 GB |
Recommended GPU Hardware for GLM-4-Voice-9B (FP16)
Minimum GPU count for 8K context at batch size 1 and estimated generation throughput.
| GPU Hardware | Min GPUs | Total VRAM Pool | Est. Speed |
|---|---|---|---|
| NVIDIA RTX 4090 (24GB) | 1× | 24 GB | 46 tok/s |
| NVIDIA RTX 5090 (32GB) | 1× | 32 GB | 81 tok/s |
| Apple M4 Max (128GB) (128GB) | 1× | 128 GB | 19 tok/s |
| NVIDIA A100 (80GB) (80GB) | 1× | 80 GB | 88 tok/s |
| NVIDIA H100 (80GB) | 1× | 80 GB | 149 tok/s |
| NVIDIA H200 (141GB) | 1× | 141 GB | 217 tok/s |
| Huawei Ascend 910B (64GB) | 1× | 64 GB | 54 tok/s |
GLM-4-Voice-9B Deployment FAQ
How much VRAM is needed to run GLM-4-Voice-9B?
At 8K context and batch size 1, GLM-4-Voice-9B uses about 21.2 GB in FP16, 11.7 GB in FP8, and 6.9 GB in INT4/AWQ quantization.
Can I run GLM-4-Voice-9B on RTX 4090 or H100 GPUs?
Running GLM-4-Voice-9B in FP16 precision requires at least 1x RTX 4090 (24GB) or 1x NVIDIA H100 (80GB) GPUs.
What precision should I choose for GLM-4-Voice-9B in production?
For GLM-4-Voice-9B (9.4B by Zhipu AI), FP16 provides the best balance between inference speed, VRAM footprint, and output quality.