Medgemma 27B IT — Hardware Requirements & GPU Compatibility
Vision
MedGemma 27B IT is Google's instruction-tuned medical model, built on Gemma 3 27B. The Hugging Face metadata tags it for medical text and imaging uses such as radiology report generation, chest X-ray, pathology, dermatology and fundus images, and it takes both text and images. It is intended as a starting point for health-related applications and is not a general-purpose assistant. The listed size is about 28.8 billion parameters including the vision encoder. It fits a single high-memory GPU once quantized, but consumer cards need aggressive quantization. Like Gemma 3 27B, it supports a 128K-token context window. It is released under the Health AI Developer Foundations terms of use rather than a standard open-source license, and access requires acknowledging those terms on Hugging Face. Published in July 2025, it follows Gemma 3 and has smaller MedGemma siblings.
81.9K downloads
448 likesSpecifications
- Publisher
- Family
- Gemma
- Parameters
- 28.8B
- Release Date
- 2025-07-09
- License
- Other
Get Started
HuggingFace
How Much VRAM Does Medgemma 27B IT Need?
Select a quantization to see compatible GPUs below.
| Quantization | Bits | VRAM | + Context | File Size | Quality |
|---|---|---|---|---|---|
| Q2_K | 3.40 | 13.5 GB | — | 12.26 GB | 2-bit quantization with K-quant improvements |
| Q3_K_S | 3.50 | 13.9 GB | — | 12.62 GB | 3-bit small quantization |
| Q3_K_M | 3.90 | 15.5 GB | — | 14.06 GB | 3-bit medium quantization |
| Q4_0 | 4.00 | 15.9 GB | — | 14.42 GB | 4-bit legacy quantization |
| Q4_K_M | 4.80 | 19.0 GB | — | 17.31 GB | 4-bit medium quantization — most popular sweet spot |
| Q5_K_M | 5.70 | 22.6 GB | — | 20.55 GB | 5-bit medium quantization — good quality/size tradeoff |
| Q6_K | 6.60 | 26.2 GB | — | 23.79 GB | 6-bit quantization, very good quality |
| Q8_0 | 8.00 | 31.7 GB | — | 28.84 GB | 8-bit quantization, near-lossless |
Which GPUs Can Run Medgemma 27B IT?
Q4_K_M · 19.0 GBMedgemma 27B IT (Q4_K_M) requires 19.0 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 25+ GB is recommended. 8 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti.
Runs great
— Plenty of headroom
Which Devices Can Run Medgemma 27B IT?
Q4_K_M · 19.0 GB41 devices with unified memory can run Medgemma 27B IT, including NVIDIA DGX H100, NVIDIA DGX A100 640GB, Mac Mini M4 Pro (24 GB).
Runs great
— Plenty of headroomDecent
— Enough memory, may be tightWhere to Download Medgemma 27B IT
Community quantizations of this model — GGUF for llama.cpp, Ollama, and LM Studio, plus AWQ/MLX variants where available.
Related Models
Frequently Asked Questions
- How much VRAM does Medgemma 27B IT need?
Medgemma 27B IT requires 19.0 GB of VRAM at Q4_K_M, or 63.5 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 28.8B × 4.8 bits ÷ 8 = 17.3 GB
KV Cache + Overhead ≈ 1.7 GB (at 2K context + ~0.3 GB framework)
VRAM usage by quantization
Q4_K_M19.0 GB- Can NVIDIA GeForce RTX 4090 run Medgemma 27B IT?
Yes, at Q5_K_M (22.6 GB) or lower. Higher quantizations like Q6_K (26.2 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
- What's the best quantization for Medgemma 27B IT?
For Medgemma 27B IT, Q4_K_M (19.0 GB) offers the best balance of quality and VRAM usage. Q5_K_S (21.8 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 8.7 GB.
VRAM requirement by quantization
IQ2_XXS8.7 GBIQ3_XS13.1 GBQ4_015.9 GBIQ4_NL17.9 GBQ4_K_M ★19.0 GBBF1663.5 GB★ Recommended — best balance of quality and VRAM usage.
- Can I run Medgemma 27B IT on a Mac?
Medgemma 27B IT requires at least 8.7 GB at IQ2_XXS, 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 Medgemma 27B IT locally?
Yes — Medgemma 27B IT can run locally on consumer hardware. At Q4_K_M quantization it needs 19.0 GB of VRAM. Popular tools include Ollama, LM Studio, and llama.cpp.
- How fast is Medgemma 27B IT?
At Q4_K_M, Medgemma 27B IT can reach ~252 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~34 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 19.0 × 0.65 = ~273 tok/s
Estimated speed at Q4_K_M (19.0 GB)
~273 tok/s~34 tok/s~273 tok/s~252 tok/sReal-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
- What's the download size of Medgemma 27B IT?
At Q4_K_M, the download is about 17.31 GB. The full-precision BF16 version is 57.68 GB. The smallest option (IQ2_XXS) is 7.93 GB.
- Which GPUs can run Medgemma 27B IT?
8 consumer GPUs can run Medgemma 27B IT at Q4_K_M (19.0 GB). Top options include NVIDIA GeForce RTX 5090, AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX. 1 GPU have plenty of headroom for comfortable inference.
- Which devices can run Medgemma 27B IT?
41 devices with unified memory can run Medgemma 27B IT at Q4_K_M (19.0 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Asus ROG Flow Z13 (2025, Ryzen AI Max+ 395, 128 GB), Beelink GTR9 Pro (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.