Share This Article
Artificial intelligence testing: how Primate Labs’ solution work
Primate Labs has launched a new solution – Geekbench AI, based on a machine learning model. The product enables artificial intelligence testing and extends the capabilities of Geekbench cross-platform.
AI algorithms produce different results depending on the type and characteristics of the device. With Geekbench AI, developers can see how AI performs under different conditions. Testing on the platform allows you to evaluate the productivity of the solution on different types of hardware, including:
- processors;
- graphics cards;
- neural processing units (NPUs).
Geekbench AI algorithms help evaluate the processing speed of new-generation processors. Users can compare the results and analyse the hardware’s productivity.
Primate Labs representatives point out that AI testing is a more complex process than GPU testing. In this case, there are several metrics to analyze. For GPUs, on the other hand, experts usually measure only pure computing power.
When evaluating AI performance on GPUs, testers must consider a number of factors.. These include integration with frameworks and API support. These and other features should be captured in the benchmark, which is what Geekbench AI provides. The solution consists of various variations and subtests, but it is difficult to evaluate them on the same level as games. The latter also analyses scaling and frame generation.
Features of the testing process
As for the performance of Geekbench AI, the test results are divided into three groups:
- full accuracy;
- semi-accuracy;
- quantum estimates.
This approach allows users to gain a full understanding of how artificial intelligence works. This allows more changes to be made depending on the goals and capabilities of the devices. In general, the concept of AI testing is similar to that used for processors. In both cases, software use multi-threaded metrics.
Based on the results of the Geekbench AI analyses, it ranks software according to its performance. The score is a system of points awarded to the hardware for each feature. This list is subject to change regularly and is always up to date.
Geekbench AI is a valuable tool for analysing AI performance. It can help developers and solution providers improve their products.
In addition, Geekbench AI is one of the few applications of its kind to test new frameworks, such as Intel OpenVINO and Qualcomm QNN, which maximise hardware performance. The benchmark includes a variety of models run based on AI capabilities. Each reflects a specific consumer workload, such as image classification or segmentation, object evaluation and recognition, and more. Three different types of data are used for each type of test, increasing the accuracy of the results.