SuperGemma4 31B Abliterated MLX 4bit — Hardware Requirements & GPU Compatibility
ChatReasoning
SuperGemma4 31B Abliterated MLX 4bit is a 30.7B-parameter open language model from Jiunsong in the Gemma family. It supports a context window of up to 262,144 tokens. At BF16 it needs about 63.02 GB of VRAM — see which GPUs and Macs can run it below.
2.8K downloads
41 likesSpecifications
- Publisher
- Jiunsong
- Family
- Gemma
- Parameters
- 30.7B
- Architecture
- Gemma4ForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 262,144
- Release Date
- 2026-04-15
- License
- Gemma Terms
Get Started
How Much VRAM Does SuperGemma4 31B Abliterated MLX 4bit Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| BF16 | 16.00 | 63.0 GB | 230.8 GB | 61.39 GB | Brain floating point 16 — preferred for training |
Which GPUs Can Run SuperGemma4 31B Abliterated MLX 4bit?
BF16 · 63.0 GBSuperGemma4 31B Abliterated MLX 4bit (BF16) requires 63.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 82+ GB is recommended. Using the full 262K context window can add up to 167.8 GB, bringing total usage to 230.8 GB. No single GPU has enough memory — multi-GPU or cluster setups are needed.
Which Devices Can Run SuperGemma4 31B Abliterated MLX 4bit?
BF16 · 63.0 GB8 devices with unified memory can run SuperGemma4 31B Abliterated MLX 4bit, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Studio M4 Max (64 GB).
Runs great
— Plenty of headroom
Decent
— Enough memory, may be tightRelated Models
Frequently Asked Questions
- How much VRAM does SuperGemma4 31B Abliterated MLX 4bit need?
SuperGemma4 31B Abliterated MLX 4bit requires 63.0 GB of VRAM at BF16. Full 262K context adds up to 167.8 GB (230.8 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 30.7B × 16 bits ÷ 8 = 61.4 GB
KV Cache + Overhead ≈ 1.6 GB (at 2K context + ~0.3 GB framework)
KV Cache + Overhead ≈ 169.4 GB (at full 262K context)
VRAM usage by quantization
BF1663.0 GBBF16 + full context230.8 GB- Can NVIDIA GeForce RTX 5090 run SuperGemma4 31B Abliterated MLX 4bit?
No — SuperGemma4 31B Abliterated MLX 4bit requires at least 63.0 GB at BF16, which exceeds the NVIDIA GeForce RTX 5090's 32 GB of VRAM.
- Can I run SuperGemma4 31B Abliterated MLX 4bit on a Mac?
SuperGemma4 31B Abliterated MLX 4bit requires at least 63.0 GB at BF16, which exceeds the unified memory of most consumer Macs. You would need a Mac Studio or Mac Pro with a high-memory configuration.
- Can I run SuperGemma4 31B Abliterated MLX 4bit locally?
Yes — SuperGemma4 31B Abliterated MLX 4bit can run locally on consumer hardware. At BF16 quantization it needs 63.0 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is SuperGemma4 31B Abliterated MLX 4bit?
At BF16, SuperGemma4 31B Abliterated MLX 4bit can reach ~46 tok/s on AMD Instinct MI300X. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: AMD Instinct MI300X → 5300 ÷ 63.0 × 0.55 = ~46 tok/s
Estimated speed at BF16 (63.0 GB)
~46 tok/s~35 tok/s~29 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of SuperGemma4 31B Abliterated MLX 4bit?
At BF16, the download is about 61.39 GB.
- Which GPUs can run SuperGemma4 31B Abliterated MLX 4bit?
No single consumer GPU has enough VRAM to run SuperGemma4 31B Abliterated MLX 4bit at BF16 (63.0 GB). Multi-GPU or professional hardware is required.
- Which devices can run SuperGemma4 31B Abliterated MLX 4bit?
8 devices with unified memory can run SuperGemma4 31B Abliterated MLX 4bit at BF16 (63.0 GB), including Mac Pro M2 Ultra (192 GB), Mac Studio M2 Ultra (192 GB), Mac Studio M4 Max (128 GB), Mac Studio M4 Max (64 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.