Gemma 4 12B

Q4_K_M running on AMD Radeon RX 6900 XT + AMD Radeon Pro W5700X ×2 with macOS 26.5.2 Tahoe

PERFORMANCE / TOKENS PER SECONDRDNA 2 + RDNA 1

Prompt processing800.04tok/s

#36 in index

Token generation47.1tok/s

#57 in index

Run fingerprint

Enough context to make the number useful.

ModelGemma 4 12BDense architecture

QuantizationQ4_K_MExact submitted variant

GPUAMD Radeon RX 6900 XT + AMD Radeon Pro W5700X ×247.95 GB VRAM

GPU architectureRDNA 2 + RDNA 1Metal backend

Test context512prompt tokens

Model familyDenseSigned ToshLLM app · amd-gpu · Metal

Prompt processing and generation are reported separately so memory bandwidth and decode behavior stay visible.

Quantization, GPU family, VRAM and context travel with the result instead of being hidden behind one score.

This result remains explicitly labeled as a community submission.