This topic answers frequently asked questions about Intelligent Computing Lingjun.
Q: How do I create and delete node groups in a cluster?
You can create a node group for a Lingjun cluster in two ways:
Create a node group when you create the cluster. For more information, see Cluster and group configurations.
Create a node group for an existing cluster.
In the left-side navigation pane, choose Resources & Nodes > Cluster Management.
Click the target Cluster ID/Name.
Click the Node Groups tab.
Click Create Node Group. Enter the node group name, default instance type, and other information.
(Optional) After creating a node group, you can edit its name or delete it.
To delete a node group:
In the left-side navigation pane, choose Resources & Nodes > Cluster Management.
Click the target Cluster ID/Name.
Click the Node Groups tab.
In the Actions column of the target node group, click Delete.
In the dialog box that appears, click OK.
Q: Why must I delete all nodes before deleting a cluster?
You must first scale in the cluster to zero nodes before you can delete it. For detailed instructions, see Delete a cluster.
Q: Why do deep learning and neural networks require GPUs?
The following table compares GPUs and CPUs.
Item | GPU | CPU |
Arithmetic logic unit (ALU) | Contains numerous arithmetic logic units, making it excellent for large-scale parallel computations. | Contains a few powerful arithmetic logic units. |
Control unit | Has a simpler control unit. | Features a complex control unit. |
Cache | Has a smaller cache that is used for servicing threads, not for storing accessed data. | Has a large cache to store data, which improves access speed and reduces latency. |
Response method | Processes tasks in batches. | Provides real-time responses and processes individual tasks quickly. |
Scenarios | Ideal for compute-intensive, highly similar, and multithreaded parallel workloads that require high throughput. | Suitable for logic-intensive serial computing workloads that require low latency. |
GPUs excel at parallel computations. You can use parallel programming methods to accelerate computations with GPUs. A neural network is highly parallel, making it perfectly suited for GPU computation. A typical example is convolution, where each calculation is independent. Many computations in a neural network can be easily broken down into smaller, independent calculations.
Q: How does Intelligent Computing Lingjun differ from standard GPU hosting?
Intelligent Computing Lingjun clusters use a system architecture and multi-layered performance optimization technologies designed specifically for large-scale AI computing. In highly parallel computing scenarios, such as natural language processing, autonomous driving model training, and recommendation engines, Intelligent Computing Lingjun can reduce training time and costs and help you build larger, more complex models compared to a standard GPU hosting service.
Q: Do I need to install GPU drivers after creating a cluster?
The operating system image for Lingjun compute nodes already includes GPU drivers. You can run the nvidia-smi command to verify the GPU driver installation and check the graphics card status.
Q: How do I view GPU details?
The method for viewing GPU information depends on the node's operating system:
On Linux nodes, run the
nvidia-smicommand to view detailed GPU information.For metrics such as GPU idle rate, utilization, temperature, and power, see the Data Dashboard.
Q: How do I use the eGPU toolkit?
Lingjun nodes come with a pre-installed, three-month trial of the eGPU toolkit. To use it for a longer period, submit a ticket. Currently, long-term licenses for the eGPU toolkit are available only to users who have completed enterprise certification.