llmrun
GoogleGemma

Medgemma 27B IT — Hardware Requirements & GPU Compatibility

Vision

MedGemma 27B IT is Google's instruction-tuned medical model built from the base Gemma 3 27B architecture. Hugging Face metadata marks it for tasks involving medical text and imaging, including radiology report generation, chest X-ray analysis, pathology review, dermatology checks, and fundus image interpretation; it accepts both textual and visual inputs. Designed as a foundation for health-related tools rather than a general-purpose assistant, the model contains roughly 28.8 billion parameters when counting its vision encoder. Once quantized, it runs on a single high-memory GPU, though consumer cards require heavy compression to function. Like Gemma 3 27B, this variant supports a context window of up to 128K tokens. Its release falls under the Health AI Developer Foundations terms instead of a standard open-source license, meaning users must acknowledge these conditions on Hugging Face before access is granted. Published in July 2025, it succeeds Gemma 3 and includes smaller MedGemma siblings as part of its family.

27.4B params · denseOther37.5K downloads492 likes14.9K quant downloads
B

Decent on a 24 GB GPU

Runs from12 GB IQ2_XXS · 16KBest on24 GB Q4_K_M · ~36 tok/s · 16KMacs from16 GB IQ2_XXS · ~17 tok/s

Other

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Medgemma 27B IT

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

Ollama
one command · needs Ollama installed
$ollama run medgemma:27b

Q4_K_M fits fully on the RTX 4090 · 2.3 GB free · 16K context

Download: medgemma-27b-it-Q4_K_M.gguf16.55 GBunsloth

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 5090 (32 GB) (10 GB headroom) · Q4_K_M

Generation speed
~64 tok/s
generation speed
Cost per 1M output tokens
$2.32
per 1M output tokens

How Much VRAM Does Medgemma 27B IT Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
12.8 GB20.6 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 32K context · ~51 tok/s
13.2 GB21 GB3-bit small — 3-bit small quantizationFits: 32K context · ~50 tok/s
14.7 GB22.5 GB3-bit medium — 3-bit medium quantizationFits: 32K context · ~45 tok/s
15.1 GB22.9 GB4-bit · legacy — 4-bit legacy quantizationFits: 32K context · ~43 tok/s
18.1 GB25.9 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 16K context · ~36 tok/s
21.5 GB29.3 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 8K context · ~31 tok/s
24.9 GB32.7 GB6-bit, very good — 6-bit quantization, very good qualityDoes not fit: needs 28.5 GB
30.2 GB37.9 GB8-bit, near-lossless — 8-bit quantization, near-losslessDoes not fit: needs 33.8 GB

Selected Q4_K_M

4-bit medium quantization

21.7GBat 16K context

Weights
17.3 GB
KV cache
4.1 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 2.3 GB free · ~36 tok/s

≥50 tok/s needs ≥1.39 TB/s · RTX 5090 ~64

Which Hardware Can Run Medgemma 27B IT?

Q4_K_M · 18.1 GB

Medgemma 27B IT (Q4_K_M) requires 18.1 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 24+ GB is recommended. 8 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 137 devices with unified memory can run Medgemma 27B IT, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB), Mac Mini M5 Pro (24 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Doesn't fit — smallest quant needs 11.9 GB8 GB tier →
12 GB · RTX 3060 12GBIQ2_XXS · 11.9 GB · ~28 tok/s · 16K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ2_K_S · 15.7 GB · ~24 tok/s · 16K16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 21.7 GB · ~36 tok/s · 16K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 21.7 GB · ~5 tok/s · 32KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 21.7 GB · ~21 tok/s · 32KMacs →

Alternatives to Medgemma 27B IT

Gemma family · 17 models →

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
Qwen3.8 27BAlibaba121.917.9 GB~38same size
Qwen3.6 27BAlibaba111.217.9 GB~38same size
Qwen3.5 27BAlibaba103.817.9 GB~38same size
ERNIE 4.5 VL 28B A3B PTBaidu—60 GB—needs > 24 GBsame size
JiRackDeltaNet 27BCMSManhattan—17.6 GB~38same size
Granite 4.1 30BIBM—21.9 GB~36same size

Specifications

Publisher
Google
Family
Gemma
Parameters
27.4B
Release Date
2025-07-09
License
Other

Frequently Asked Questions

How much VRAM does Medgemma 27B IT need?

Medgemma 27B IT requires 18.1 GB of VRAM at Q4_K_M, or 60.4 GB at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 27.4B × 4.8 bits ÷ 8 = 16.5 GB

KV Cache + Overhead ≈ 1.6 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.

VRAM usage by quantization

18.1 GB

Learn more about VRAM estimation →

Can NVIDIA GeForce RTX 4090 run Medgemma 27B IT?

Yes, at Q5_K_L (21.9 GB) or lower. Higher quantizations like Q6_K (24.9 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 (18.1 GB) offers the best balance of quality and VRAM usage. Q4_K_L (18.5 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 8.3 GB.

VRAM requirement by quantization

IQ2_XXS
8.3 GB
IQ3_S
12.8 GB
Q3_K_L
15.5 GB
Q4_K_M ★
18.1 GB
Q5_0
18.9 GB
BF16
60.4 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Medgemma 27B IT on a Mac?

Yes — Mac Mini M2 Pro (16 GB) and 145 other Macs can run Medgemma 27B IT, the smallest at IQ2_XXS. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 18.1 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M5 Pro (24 GB).

Can I run Medgemma 27B IT locally?

Yes — Medgemma 27B IT can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Medgemma 27B IT?

At Q4_K_M, Medgemma 27B IT can reach ~265 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~36 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 ÷ 18.1 × 0.65 = ~287 tok/s

Estimated speed at Q4_K_M (18.1 GB)

~287 tok/s
~36 tok/s
~287 tok/s
~265 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of Medgemma 27B IT?

At Q4_K_M, the download is about 16.46 GB. The full-precision BF16 version is 54.86 GB. The smallest option (IQ2_XXS) is 7.54 GB.

Which GPUs can run Medgemma 27B IT?

8 consumer GPUs can run Medgemma 27B IT at Q4_K_M. Top options include NVIDIA GeForce RTX 5090, AMD Radeon RX 7900 XT, AMD Radeon RX 7900 XTX. 1 GPU has plenty of headroom for comfortable inference.

Which devices can run Medgemma 27B IT?

137 devices with unified memory can run Medgemma 27B IT at Q4_K_M (18.1 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.