Efrain Garay (@efraingaray)

Software architect, 20+ years building systems. I test new models, databases and tools on a 16 GB GPU and a home lab, and publish the numbers, failures included, at efraingaray.com.

Efrain Garay

About

Available for

Efrain Garay's blogs

NVFP4 vs FP8 vs GGUF on a 16 GB Ada GPU: NVFP4 decodes 2.3x faster and ties BF16 on GSM8K

Kandinsky 6.0 on a 16 GB GPU: the official recipe runs out of memory, and only the 29B model passed my Spanish test

Qwen-Image 2.1 on 16 GB of VRAM, quantized to NF4: a real benchmark against FLUX

Efrain Garay

About

Available for

Efrain Garay's blogs

NVFP4 vs FP8 vs GGUF on a 16 GB Ada GPU: NVFP4 decodes 2.3x faster and ties BF16 on GSM8K

Kandinsky 6.0 on a 16 GB GPU: the official recipe runs out of memory, and only the 29B model passed my Spanish test

Qwen-Image 2.1 on 16 GB of VRAM, quantized to NF4: a real benchmark against FLUX

Efrain Garay

About

Available for

Efrain Garay's blogs

NVFP4 vs FP8 vs GGUF on a 16 GB Ada GPU: NVFP4 decodes 2.3x faster and ties BF16 on GSM8K

Kandinsky 6.0 on a 16 GB GPU: the official recipe runs out of memory, and only the 29B model passed my Spanish test

Qwen-Image 2.1 on 16 GB of VRAM, quantized to NF4: a real benchmark against FLUX

Search Hashnode

Efrain Garay

About

Available for

Efrain Garay's blogs

NVFP4 vs FP8 vs GGUF on a 16 GB Ada GPU: NVFP4 decodes 2.3x faster and ties BF16 on GSM8K

Kandinsky 6.0 on a 16 GB GPU: the official recipe runs out of memory, and only the 29B model passed my Spanish test

Qwen-Image 2.1 on 16 GB of VRAM, quantized to NF4: a real benchmark against FLUX