Compare

Loading community benchmarks…

Community benchmarks

Compare

Put two chips, runtimes, or runtime versions side by side. Everything else is held fixed or called out, so a ratio is worth exactly what the matched runs say.

Add your benchmark

Runs compared: 12
Like-for-like pair: 1
Chips available: 8

Compare

Side A

Side B

Model

Runtime

Runtime version

Quantisation

Backend

Prefill size

Decode size

Cooldown

8 runs vs 4 runs · 1 like-for-like pair

Held fixedModel: Llama 3.2 3B InstructPrefill: PP512Decode: TG128

A side mixes quantisations. Pick a quantisation to compare the same weights.
The sides differ in conditioning, decode workload, harness schema. Ratios may reflect benchmark protocol rather than the chip.

Chip comparison

Apple M5 Pro vs NVIDIA GB10

Ratios read side A relative to side B. Geometric mean over 1 like-for-like pair.

Decode: 1.19×+19.0%
Prefill: 0.29×-70.9%

Side A

Apple M5 Pro

8 runs

1 contributor

Latest 11 Sep 2026

Decode · TG128: 100.4tok/s median
best 117.7 · worst 76.6
Prefill · PP512: 4,068tok/s median
best 4,344 · worst 3,315

Models: Llama 3.2 3B Instruct, Llama-3.2-3B-Instruct
Runtimes: BaseRTllama.cpp
Versions: BaseRT 0.2.4, llama.cpp b10809 (5266f24da)
Quantisations: Q4, Q4_K_M, Q8_0
Backends: BLAS + Metal, Metal
Cooldown: off
Conditioning: runtime_native_warmup, warmup_only
Decode workload: TG128
Harness schema: basert-benchmark-harness/1, computearena-measurements/1

Side B

NVIDIA GB10

4 runs

1 contributor

Latest 19 Sep 2026

Decode · TG128: 79.6tok/s median
best 110.5 · worst 68.7
Prefill · PP512: 14,761tok/s median
best 14,988 · worst 14,073

Models: Llama-3.2-3B-Instruct-cuda-q4mix, Llama-3.2-3B-Instruct-cuda-q8, Llama-3.2-3B-Instruct-Q4, Llama-3.2-3B-Instruct-Q8
Runtimes: BaseRT
Versions: BaseRT 0.2.6
Quantisations: Q4
Backends: CUDA
Cooldown: off
Conditioning: warmup_only
Decode workload: TG128 @ 1 ctx
Harness schema: basert-benchmark-harness/1

What differs besides the chip

Facet	Apple M5 Pro	NVIDIA GB10
Model	Llama 3.2 3B Instruct, Llama-3.2-3B-Instruct	Llama-3.2-3B-Instruct-cuda-q4mix, Llama-3.2-3B-Instruct-cuda-q8, Llama-3.2-3B-Instruct-Q4, Llama-3.2-3B-Instruct-Q8
Runtime	BaseRT, llama.cpp	BaseRT
Runtime version	BaseRT 0.2.4, llama.cpp b10809 (5266f24da)	BaseRT 0.2.6
Quantisation	Q4, Q4_K_M, Q8_0	Q4
Backend	BLAS + Metal, Metal	CUDA
Conditioning	runtime_native_warmup, warmup_only	warmup_only
Decode workload	TG128	TG128 @ 1 ctx
Harness schema	basert-benchmark-harness/1, computearena-measurements/1	basert-benchmark-harness/1

Like-for-like pairs

Each row is a configuration present on both sides. Values are per-cell medians; ratios read Apple M5 Pro relative to NVIDIA GB10.

Configuration	Decode A	Decode B	Ratio	Prefill A	Prefill B	Ratio	Runs A / B
Llama-3.2-3B-Instruct-cuda-q8 BaseRTQ4	94.6	79.6	1.19×	4,297	14,761	0.29×	5 / 4

Side A · Apple M5 Pro

Top 8 of 8 runs by decode throughput.

Model / format	Chip / backend	Decode	Prefill	Contributor	Date	Report
Llama-3.2-3B-Instruct llama.cpp b10809 (5266f24da)Q4_K_M	Apple M5 Pro BLAS + Metal	117.7 TG128	3,318 PP512	arki05	11 Sep 2026	View Benchmark
Llama-3.2-3B-Instruct llama.cpp b10809 (5266f24da)Q4_K_M	Apple M5 Pro BLAS + Metal	115.2 TG128	3,315 PP512	arki05	11 Sep 2026	View Benchmark
basecompute/Llama-3.2-3B-Instruct BaseRT 0.2.4Q4	Apple M5 Pro Metal	106.8 TG128	4,297 PP512	arki05	11 Sep 2026	View Benchmark
basecompute/Llama-3.2-3B-Instruct BaseRT 0.2.4Q4	Apple M5 Pro Metal	106.1 TG128	4,291 PP512	arki05	11 Sep 2026	View Benchmark
basecompute/Llama-3.2-3B-Instruct BaseRT 0.2.4Q4	Apple M5 Pro Metal	94.6 TG128	3,845 PP512	arki05	11 Sep 2026	View Benchmark
basecompute/Llama-3.2-3B-Instruct BaseRT 0.2.4Q4	Apple M5 Pro Metal	79.5 TG128	4,343 PP512	arki05	11 Sep 2026	View Benchmark
basecompute/Llama-3.2-3B-Instruct BaseRT 0.2.4Q4	Apple M5 Pro Metal	79.5 TG128	4,344 PP512	arki05	11 Sep 2026	View Benchmark
Llama 3.2 3B Instruct llama.cpp b10809 (5266f24da)Q8_0	Apple M5 Pro BLAS + Metal	76.6 TG128	3,516 PP512	arki05	11 Sep 2026	View Benchmark

Side B · NVIDIA GB10

Top 4 of 4 runs by decode throughput.

Model / format	Chip / backend	Decode	Prefill	Contributor	Date	Report
Llama-3.2-3B-Instruct-cuda-q4mix BaseRT 0.2.6Q4	NVIDIA GB10 CUDA	110.5 TG128 @ 1 ctx	14,651 PP512	basecompute	19 Sep 2026	View Benchmark
Llama-3.2-3B-Instruct-Q4 BaseRT 0.2.6Q4	NVIDIA GB10 CUDA	89.7 TG128 @ 1 ctx	14,073 PP512	basecompute	19 Sep 2026	View Benchmark
Llama-3.2-3B-Instruct-cuda-q8 BaseRT 0.2.6Q4	NVIDIA GB10 CUDA	69.4 TG128 @ 1 ctx	14,988 PP512	basecompute	19 Sep 2026	View Benchmark
Llama-3.2-3B-Instruct-Q8 BaseRT 0.2.6Q4	NVIDIA GB10 CUDA	68.7 TG128 @ 1 ctx	14,871 PP512	basecompute	19 Sep 2026	View Benchmark