java-benchmarks

GC Comparison: G1 vs ZGC

Comparison of throughput between G1 and ZGC collectors. Higher scores (ops/s) are better.

Benchmark	G1 Score	ZGC Score	Ratio (ZGC/G1)	Winner	Improvement
allocateLargeObjects (listSize:1000)	9817.914 ops/s	9461.231 ops/s	0.964x	G1	+3.770%
allocateLargeObjects (listSize:10000)	9987.585 ops/s	9719.426 ops/s	0.973x	G1	+2.759%
allocateShortLived (listSize:1000)	95660.123 ops/s	95228.884 ops/s	0.995x	G1	+0.453%
allocateShortLived (listSize:10000)	8837.170 ops/s	9309.619 ops/s	1.053x	ZGC	+5.346%

Summary

Overall, G1 performed better in this suite, winning 3 out of 4 tests.

Key Differences

• G1 GC: Traditional generational collector, balanced throughput and latency. Default in most JDKs.
• ZGC: Low-latency scalable collector, designed for sub-millisecond pauses even with large heaps.

Benchmarks Description

allocateShortLived (Throughput)

Allocates many short-lived String objects to a List. This benchmark tests how efficiently the GC handles high allocation rates of small objects. High throughput here means the GC can keep up with rapid allocations without excessive stalling.

allocateLargeObjects (Throughput)

Allocates 1MB byte arrays repeatedly. This benchmark tests how the GC handles large object allocations (Humongous objects in G1). G1 might struggle more with these as they require contiguous regions.

This site is open source. Improve this page.