ApexMark › CPU + GPU stress test
CPU and GPU stress test together
Run CPU ray tracing and the Oméga 3D scene at the same time. Total torture gives each workload its own FPS graph and controls, with one button to stop both. Compare each scene on its own, then observe what changes when they share your computer.
Open Total tortureTwo scenes running at the same time
The CPU renderer completes ray-traced images in browser workers while the GPU renderer completes its WebGPU scene. Both run concurrently after preparation.
Each scene keeps its own FPS, 1% low and recent history. The test does not add their frame rates or turn them into a combined score.
Explore what changes under concurrent load
Open Live test in the menu and choose Total torture. Rendering starts automatically after preparation while the page is visible. If your browser requests reduced motion or the page is hidden when opened, choose Start both once it is visible.
Start with the individual CPU and GPU live tests if you want a reference for each scene. Then run Total with the same respective settings, camera views, browser and power conditions.
A brief session is useful for showing both renderers at work. A longer visible session lets you observe variation over time.
Control each scene independently
CPU resolution and samples control the ray-traced image. GPU resolution and complexity control the graphics workload. CPU complexity offers five levels: 1, 2, 4, 8 or 16 samples per pixel.
Change one setting at a time to keep the comparison readable. The pair prepares again when settings change.
- Keep one scene’s settings fixed while adjusting the other.
- Use independent camera controls, or fullscreen to inspect the pair.
- Stop both workloads together with the shared Stop control.
Read each graph on its own terms
Each large FPS reading averages its own completed frame times over the latest second. The CPU count includes full image delivery; the GPU count waits for actual queue completion.
Each 1% low needs at least 100 completed frames and uses up to the latest 1,000 durations for that scene. One scene may collect enough frames before the other.
Track a change across both workloads
Changing either scene’s resolution or complexity preserves both recent histories and the session duration. Orange markers show the applied CPU or GPU change on both graphs, with its time and before-and-after settings.
Current FPS, average, minimum and 1% low restart. Stop freezes both traces. Hiding or leaving the page, renderer failure or GPU device loss stops both workloads and releases their resources. Returning requires an explicit restart.
What this combined test can tell you
Total runs these CPU and GPU scenes together. It adds no dedicated RAM, storage or internet test, and reports no physical utilization percentage.
The session runs in your browser without an account or saved public score. For fixed, verified measurements, use the separate ApexMark benchmark. Continuous rendering does not certify hardware health or thermal stability, and the two live readings must remain separate.
Frequently asked questions
What does the Total live test run?
It runs the CPU ray tracer and WebGPU graphics scene concurrently, with two visible surfaces. Each keeps its own settings and statistics. A shared control starts or stops the pair.
Are the CPU and GPU frame rates added together?
No. They measure different completed workloads and remain separate. There is no combined FPS, Total score or hardware utilization percentage.
How can I compare Total with the individual live tests?
Use the same browser, power conditions, respective settings and camera views. Run each individual mode, then the pair, and repeat if needed. Interpret any difference as an observation under those conditions.
Does it stress-test RAM and storage too?
It adds no dedicated memory or storage workload. The renderers naturally need resources, but that does not make this a RAM diagnostic, storage benchmark or full-system stability test.
Why does changing one scene restart both sets of statistics?
A settings change replaces the renderer pair. Current readings and 1% low windows start again, while both recent graphs and session duration remain visible. Markers show which scene’s settings changed.
Can this identify a CPU or GPU bottleneck?
It can show how these two specific workloads behave together. It cannot establish a universal bottleneck, diagnose failing hardware or predict commercial-game FPS. Use matched comparisons and keep those limits in mind.