TraverseWork with us
Menu

Traverse Research

Evolve

Our GPU and AI benchmark. Traverse’s own product, launched in March 2025, run in-house by Qualcomm and Samsung and independently evaluated by LTT Labs.

Watch the Evolve trailer
A workload render from Evolve, our GPU and AI benchmark.
Built by
Traverse Research
Disciplines
Benchmarking, GPU architecture
Platforms
PC, Android

The product

Evolve is our cross-platform GPU and AI benchmark, and the one thing on this site you can buy and run yourself rather than commission. It packages the workloads, measurement harness and reporting we built for hardware partners into software any team can license, run and compare.

Who runs it

Evolve launched in March 2025. Qualcomm and Samsung run it in-house. It ships on PC and Android through its own website and storefronts.

What it measures

Its public technical guide covers compute, ray tracing, rasterization, driver performance, energy consumption, acceleration structures and work graphs. Each workload is configurable and repeatable, so a result points at an architectural trade-off rather than a single headline score.

Independent evaluation

In Evolve Benchmark Exploration, LTT Labs tested Evolve 13.1 across five Android devices, including the REDMAGIC 11 Pro, OnePlus 15 and Samsung Galaxy S25 Edge. Its article publishes repeated runs, category scores and comparisons with other benchmarks. The authors discuss repeatability and the need to continue investigating relevance to real workloads.

Read the LTT Labs evaluation.

How Evolve stays neutral

A benchmark is only worth running if nobody can buy its result. Traverse does paid engineering for several of the vendors Evolve measures, and Qualcomm and Samsung run it in-house, so the question deserves a direct answer rather than a reassurance.

Evolve publishes its own neutrality commitments: it is independent by design, owned and steered by no vendor; there is no pay-to-win, so performance cannot be bought; its methods are public and its results reproducible; and it serves users, not sponsors. Read the neutrality statement. The technical guide sets out the rules for what is and is not allowed in a scored run.

The engineering backs that up. Every category is scored separately, so a vendor strong in one place cannot hide a weakness in another behind a single number. Every workload is configurable and repeatable, and the measurement conditions ship with the result. The same build runs on retail hardware through public storefronts, so anyone can reproduce a figure a vendor quotes. LTT Labs evaluated it independently, and our six years as a Khronos member and our open-source record are there for anyone to check.

Why we build it

Benchmarking is not a side project here. It is how we learn a GPU: the same workload engineering behind Evolve is what we bring to pre-silicon programmes, driver investigations and performance research for vendors. Evolve is that capability, productised.

Evolve has its own brand and website. For current features, supported configurations, downloads and business offerings, visit Evolve.

Visit Evolve

Traverse Research + Traverse Games

Let’s turn a good question into working software.

Start a conversation