on

Can the AMD Instinct MI300X run ShieldGemma 27B?

66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.

Yes, it runs well

Runs well. At Q8_0 it needs about 39.8 GB of 192 GB usable and makes one decision in about 151 ms.

Best quant
Q8_0
Time per decision
151 ms
Decisions per second
6.6/s
Input assumed
512 tokens
Returns
Probability or score

Best app to run it on the AMD Instinct MI300X

On Linux

vLLM

Inference server · runs on vLLM

  • The full model fits in GPU memory at 8-bit, so it can batch many requests: best for serving a team or an app.
vllm serve google/shieldgemma-27b
  • Lemonade AMD’s own app: llama.cpp tuned for Radeon on Vulkan and ROCm.
  • LM Studio Point-and-click: search, download and chat in one window.

On Windows

Lemonade

Desktop app · runs on llama.cpp, ONNX Runtime GenAI (Ryzen AI NPU)

  • AMD’s own app: llama.cpp tuned for Radeon on Vulkan and ROCm.
lemonade pull mradermacher/shieldgemma-27b-GGUF:Q8_0
  • LM Studio Point-and-click: search, download and chat in one window.
  • Ollama One command with Ollama’s own build of this model.

How we choose: the files published for this model, your operating system and GPU, whether it fits fully in memory, and published speed tests (MLX vs llama.cpp on Apple silicon, vLLM vs llama.cpp vs Ollama on NVIDIA). Details.

Download ShieldGemma 27B

Recommended for the AMD Instinct MI300X: Q8_0. Every other quantization has its own download link in the table below.

Quickest · Ollama

One command downloads a ready-to-run version and starts a chat.

Get Ollama at ollama.com/download (opens in a new tab) · model page (opens in a new tab)

ollama run shieldgemma:27b

Quantized files · GGUF

Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.

Download the Q8_0 file (opens in a new tab)

huggingface.co/mradermacher/shieldgemma-27b-GGUF (opens in a new tab)

ollama run hf.co/mradermacher/shieldgemma-27b-GGUF:Q8_0

Original weights · Hugging Face

Full-precision files from Google, for fine-tuning, vLLM, Transformers.

huggingface.co/google/shieldgemma-27b (opens in a new tab)

Free Hugging Face account needed: request access on the model page, and Google approves it (usually within minutes to a few hours).

Links checked 3 days ago against Hugging Face, Ollama and GitHub.

How it’s rated

Quality 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking.
How capable the model itself is at Q8_0, from its editorial quality tier scaled by the quantization or precision. It’s the same on any hardware that runs this variant. Scale: S 85+ · A 70+ · B 55+ · C 40+ · D under 40.
Runs Runs well
How well it runs on the AMD Instinct MI300X: speed, capped by how well it fits in memory. Runs great: 30+ tokens/s (or near the image, video or decision speed target) · Runs well: 10+ tokens/s · Slow: 3+ tokens/s · Very slow: under 3 · Won’t run: doesn’t fit even with offload. A tight fit or RAM offload caps it at Runs well; heavy offload caps it at Slow.

Match score

To order lists, each result also gets a match score out of 100 from four parts. It favors the most capable model that still runs well.

PrecisionEditorial precision tier from the model’s published benchmarks, scaled by the quantization or precision used.29.7 / 45
Memory fitFull marks when everything sits in GPU memory; fewer when part spills to system RAM.25.0 / 25
SpeedTime for one decision against a 100 ms target: reading the input plus any tokens the model writes.12.2 / 15
ClarityHow easy the answer is to use in code: a probability or score earns the most, written feedback before a score the least.15.0 / 15
Match score81.9 / 100

The match score orders results; it is not shown as a grade. Quality and Runs are the two ratings in the tables.

Every quantization on the AMD Instinct MI300X

Quant Memory needed Available Time per decision Runs Quality Download
Q4_K_M 27.6 GB 192 GB 146 ms Runs well 62Good quality (B) · 62 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 80/100 (quality, fit, speed and memory use) sets the ranking. GGUF (opens in a new tab)
Q5_K_M 30.3 GB 192 GB 147 ms Runs well 64Good quality (B) · 64 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. GGUF (opens in a new tab)
Q6_K 33.3 GB 192 GB 148 ms Runs well 65Good quality (B) · 65 of 100: how capable the model is at Q6_K, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. GGUF (opens in a new tab)
Q8_0 39.8 GB 192 GB 151 ms Runs well 66Good quality (B) · 66 of 100: how capable the model is at Q8_0, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. GGUF (opens in a new tab)
F16 65.4 GB 192 GB 162 ms Runs well 66Good quality (B) · 66 of 100: how capable the model is at F16, whatever it runs on. Match score 81/100 (quality, fit, speed and memory use) sets the ranking. Original weights (opens in a new tab)

Times are for one request with a 512-token input and about 1 tokens written. Batching servers process many requests at once.

Other hardware that runs it

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.