G9v3 39A5B — Hardware Requirements & GPU Compatibility
Chat
G9v3 39A5B is a 39.0B-parameter open language model from ai9stars. It supports a context window of up to 131,072 tokens. At BF16 it needs about 78.27 GB of VRAM — see which GPUs and Macs can run it below.
G9v3 39A5B runs on a 96 GB+ Mac or a workstation GPU
Best local pick → Mac Studio (M3 Ultra, 96GB) · BF16 · 78.6 GB · ~38 tok/s
Macs from 96 GB (BF16 · ~38 tok/s)
Apache 2.0
Not set up for it?
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run G9v3 39A5B
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve ai9stars/G9v3-39A5B --max-model-len 16384 --tensor-parallel-size 8vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX PRO 6000 (96 GB) (17 GB headroom) · BF16
- Generation speed
- ~79 tok/s
- generation speed
- Cost per 1M output tokens
- $5.39
- per 1M output tokens
How Much VRAM Does G9v3 39A5B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | Runs on | |
|---|---|---|---|---|
| 78.3 GB | 78.9 GB | 16-bit · bfloat — Brain floating point 16 — preferred for training | Runs on H100 SXM 80 GB |
Selected BF16
Brain floating point 1678.6GBat 16K context
- Weights
- 77.9 GB
- KV cache
- 0.3 GB
- Overhead
- 0.3 GB
Server-class: ~78.6 GB at 16K context · one H100 SXM 80 GB
Which Hardware Can Run G9v3 39A5B?
BF16 · 78.3 GBG9v3 39A5B (BF16) requires 78.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 102+ GB is recommended. Using the full 131K context window can add up to 2.5 GB, bringing total usage to 80.8 GB. No consumer GPU has enough memory. 30 devices with unified memory can run G9v3 39A5B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Studio (M3 Ultra, 96GB).
96 GB
fits · 17.4 GB free · up to 128K ctx
3 GPUs · 2 Mac classes
RTX PRO 6000 Blackwell Max-Q Workstation Edition~87Simulate output on RTX PRO 6000 Blackwell Max-Q Workstation EditionRTX PRO 6000 Blackwell Workstation Edition~87Simulate output on RTX PRO 6000 Blackwell Workstation EditionRTX PRO 6000 Blackwell Server Edition~79Simulate output on RTX PRO 6000 Blackwell Server EditionM3 Ultra*96 GB~38Simulate output on Mac Studio (M3 Ultra, 96GB)M3 Max*96 GB~17Simulate output on MacBook Pro 14" M3 Max (96 GB)128–144 GB
fits · 49.5 GB free · up to 128K ctx
4 GPUs · 2 Mac classes · 4 device families
GH200 Grace Hopper Superchip~181Simulate output on GH200 Grace Hopper SuperchipH200 NVL~179Simulate output on H200 NVLH200 SXM~179Simulate output on H200 SXMM2 Ultra*128 GB~37Simulate output on Mac Pro M2 Ultra (128 GB)M5 Max128 GB~30Simulate output on Mac Studio M5 Max (128 GB)GB10~16Simulate output on NVIDIA DGX SparkJetson T5000 (Thor)~16Simulate output on NVIDIA Jetson AGX Thor Developer Kit
By VRAM tier · typical GPUs and Macs at BF16
Consumer GPUs (up to 32 GB) and Macs up to 64 GB: too small for this model (needs ≥ 78.6 GB)8 GB tier →12 GB tier →16 GB tier →24 GB tier →Macs →
Similar models
4| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Intern S2 PreviewInternLM | — | 73.2 GB | —needs > 24 GB | same size |
| Seed OSS 36B BaseByteDance-Seed | — | 22.8 GB | ~31 | same size |
| Phi 3.5 MoE InstructMicrosoft | — | 23.9 GB | ~141 | same size |
| Falcon 40B InstructTII UAE | — | 23.8 GB | ~33 | same size |
Specifications
- Publisher
- ai9stars
- Parameters
- 39.0B
- Active parameters
- 5.4B per token
- Architecture
- G9v3ForCausalLM
- Context Length
- 131,072 tokens
- Vocabulary Size
- 130,560
- Release Date
- 2026-07-21
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does G9v3 39A5B need?
G9v3 39A5B requires 78.3 GB of VRAM at BF16. Full 131K context adds up to 2.5 GB (80.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 39.0B × 16 bits ÷ 8 = 77.9 GB
KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 2.9 GB (at full 131K context)
VRAM usage by quantization
Can NVIDIA GeForce RTX 5090 run G9v3 39A5B?
No — G9v3 39A5B requires at least 78.3 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
Can I run G9v3 39A5B on a Mac?
Yes — Mac Studio (M3 Ultra, 96GB) and 20 other Macs can run G9v3 39A5B. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 78.3 GB of usable unified memory (RAM minus macOS overhead).
Can I run G9v3 39A5B locally?
Yes — G9v3 39A5B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is G9v3 39A5B?
At BF16, G9v3 39A5B can reach ~101 tok/s on AMD Instinct MI350X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = 1000 ÷ (active GB ÷ (bandwidth GB/s × efficiency) × 1000 + layers × routing ms)
Mixture-of-Experts: just the active experts get read for each token, along with a fixed routing cost per layer.
Example: NVIDIA B200 → 10.9 GB active ÷ (8000 × 0.65) = 2.104 ms, plus 38 layers × 0.055 ms = 2.090 ms, so 1000 ÷ 4.194 ms = ~238 tok/s
Estimated speed at BF16 (78.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of G9v3 39A5B?
At BF16, the download is about 77.93 GB.
Which GPUs can run G9v3 39A5B?
No single consumer GPU has enough VRAM to run G9v3 39A5B at BF16 (78.3 GB). Multi-GPU or professional hardware is required.
Which devices can run G9v3 39A5B?
30 devices with unified memory can run G9v3 39A5B at BF16 (78.3 GB), including ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB), GMKtec EVO-X2 (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.