How TenClass Saved 80% on GPU Costs with TensorFusion ? â
Background â
TenClass is a vocational education company based in Shenzhen, China, providing digital literacy and professional skills courses to help users enhance their digital skills.
Problem â
To provide a convenient and high-quality hands-on learning environment, TenClass developed a virtual machine manager mvisor, but in the construction of hands-on learning environments for AI drawing and AI video courses, due to the need for direct GPU access in online laboratories, due to the immaturity of GPU virtualization technology, providing a dedicated AI drawing environment for each learner is very difficult and expensive.
Solution â
After learning that TensorFusion can achieve GPU virtualization and pooling, after a period of evaluation and implementation, TenClass became the first customer of TensorFusion.
TensorFusion provided the following solution to TenClass, solving the cost problem and GPU availability of cloud hands-on learning environments:
- Build a GPU pool of T4 GPUs in Linux environment
- Each T4 is cut into 2 or 3 virtual GPUs, cooling the GPU memory to host memory when not active, suspending the virtual GPU power
- Each user's exclusive Windows virtual machine hands-on learning environment, built-in TensorFusion Client, automatically connecting to virtual GPU when starting or executing drawing workflow, and getting AI computing power
Results â
| Before using TensorFusion | After using TensorFusion |
|---|---|
| đ Have to buy GPU from cloud vendor for every learner | đ Save >80% costs |
| đ GPU not available sometimes, impacting user experience | đ With GPU pooling, AI computing availability increased to 99.9% |
| đ Long set up time of new hands-on lab environment | đ Reduced 45% setup time for new hands-on lab |