RigRoute LabsRigRoute ↗
Compare

Apple M1 Pro (32 GB unified memory) vs. AMD Radeon RX 9070 (16 GB dedicated VRAM)

Both systems ran the identical standard-ai-v0.1 methodology: the same byte-identical model files (verified by SHA-256), the same llama.cpp build, the same context depth and repetition count. That is what makes this comparison meaningful — see Methodology.

llama 8B Q4_0

8b-q4

Prompt processing (tok/s)

Apple M1 Pro (32 GB unified memory)
190.39 tok/s
AMD Radeon RX 9070 (16 GB dedicated VRAM)
2511.81 tok/s

AMD Radeon RX 9070 (16 GB dedicated VRAM) measured 13.19× the prompt-processing throughput of Apple M1 Pro (32 GB unified memory) in this specific workload.

Generation (tok/s)

Apple M1 Pro (32 GB unified memory)
27.92 tok/s
AMD Radeon RX 9070 (16 GB dedicated VRAM)
102.31 tok/s

AMD Radeon RX 9070 (16 GB dedicated VRAM) measured 3.67× the generation throughput of Apple M1 Pro (32 GB unified memory) in this specific workload.

qwen3 14B Q4_K - Medium

14b-q4

Prompt processing (tok/s)

Apple M1 Pro (32 GB unified memory)
97.09 tok/s
AMD Radeon RX 9070 (16 GB dedicated VRAM)
1177.22 tok/s

AMD Radeon RX 9070 (16 GB dedicated VRAM) measured 12.13× the prompt-processing throughput of Apple M1 Pro (32 GB unified memory) in this specific workload.

Generation (tok/s)

Apple M1 Pro (32 GB unified memory)
11.73 tok/s
AMD Radeon RX 9070 (16 GB dedicated VRAM)
56.38 tok/s

AMD Radeon RX 9070 (16 GB dedicated VRAM) measured 4.81× the generation throughput of Apple M1 Pro (32 GB unified memory) in this specific workload.

Other dimensions
Apple M1 Pro (32 GB unified memory)AMD Radeon RX 9070 (16 GB dedicated VRAM)
Memory architectureunified-memorydedicated-vram (see hardware page for provenance)
BackendBLAS,MTLVulkan
Methodologystandard-ai-v0.1standard-ai-v0.1
Accelerator verification33/33, 41/41 layers offloaded to GPU33/33, 41/41 layers offloaded to GPU
Telemetry (power/temp)UNAVAILABLEUNAVAILABLE

This is not a single score. RigRoute Labs does not name an overall winner — the numbers above describe two specific, measured dimensions of one specific workload, nothing broader.