检测到您已登录华为云国际站账号，为了您更好的体验，建议您访问国际站服务网站 https://www.huaweicloud.com/intl/zh-cn
不再显示此消息
检测到您已登录华为云国际站账号，为了您更好的体验，建议您访问国际站服务网站 https://www.huaweicloud.com/intl/zh-cn
不再显示此消息
AI inference-accelerated ECSs use self-developed Ascend 310 processors for AI inference acceleration.
Kunpeng AI Inference-accelerated ECSs Kunpeng AI inference-accelerated ECSs are designed to provide acceleration services for AI services. These ECSs are provided with the Ascend AI Processors and Ascend AI Software Stack.
System kernel upgrade may cause AI software exceptions. Check the system update logs and prevent the server from restarting. The AI software may be unavailable. NPU environment command detection NpuToolsWarning Major The hccn_tool was unavailable.
inference-accelerated ai Ascend Inference Table 4 Additional capabilities Suffix Example Description ne c3ne Network Enhanced s c6s Standard v p2v NVLink h c6h High performance t c7t Trust Instance sizes The instance sizes can be small, medium, large, xlarge, or Nxlarge, as shown
Inference-accelerated Pi1 Overview Pi1 ECSs use NVIDIA Tesla P4 GPUs dedicated for real-time AI inference. Working with P4 INT8 calculators, Pi1 ECSs have shortened the inference latency by 15 times.
Scenarios High-performance computing and computer simulation Big data applications AI training and inference Specifications Table 13 C6h ECS specifications Flavor vCPUs Memory (GiB) Max./Assured Bandwidth (Gbit/s) Max. PPS (10,000) Max. NIC Queues Max.
encoding and decoding Hyper-threading (Enabling or Disabling Hyper-Threading) Computing-accelerated P2s NVIDIA V100 5,120 14 TFLOPS of single-precision floating-point computing 7 TFLOPS of double-precision floating-point computing 112 TFLOPS Tensor Cores for deep learning acceleration AI
/Machine learning AI training Compatible with NVIDIA smart NICs for deep learning training, scientific computing, computational fluid dynamics, computational finance, seismic analysis, molecular modeling, and genomics.
Kunpeng ECS Specifications and Types A Summary List of Kunpeng ECS Specifications Kunpeng General Computing-plus ECSs Kunpeng Memory-optimized ECSs Kunpeng Ultra-high I/O ECSs Kunpeng AI Inference-accelerated ECSs Parent Topic: ECS Types and Specifications
Existing ECSs cannot be added to any ECS group if they have local disks attached (such as disk-intensive or ultra-high I/O ECSs), GPU cards attached (GPU-accelerated ECSs), FPGA cards attached (FPGA-accelerated ECSs), or AI cards attached (AI-accelerated ECSs).
Core Usage The AI core usage of the NPU Linux: Obtain the metric value from the libdcmi.so library file of the NPU card. 0-100 % N/A ECS ECS - NPU 1 minute npu_util_rate_ai_cpu NPU AI CPU Usage The AI CPU usage of the NPU Linux: Obtain the metric value from the libdcmi.so library
Commercial use Memory-optimized ECSs 4 GPU-accelerated PI2 ECSs PI2 ECSs use NVIDIA Tesla T4 GPUs dedicated for real-time AI inference. These ECSs use the T4 INT8 calculator for up to 130 TOPS of INT8 computing. The PI2 ECSs can also be used for light-workload training.
Commercial use Memory-optimized ECSs 4 GPU-accelerated PI2 ECSs PI2 ECSs use NVIDIA Tesla T4 GPUs dedicated for real-time AI inference. These ECSs use the T4 INT8 calculator for up to 130 TOPS of INT8 computing. The PI2 ECSs can also be used for light-workload training.
KVM ki1.4xlarge.4 16 64 12/6 140 6 4 2 × 3,200 GiB KVM ki1.6xlarge.4 24 96 15/8.5 200 6 8 3 × 3,200 GiB KVM ki1.8xlarge.4 32 128 18/10 260 6 8 4 × 3,200 GiB KVM ki1.12xlarge.4 48 192 25/16 350 6 16 6 × 3,200 GiB KVM ki1.16xlarge.4 64 228 30/20 400 6 16 8 × 3,200 GiB KVM Kunpeng AI
Kunpeng ECSs are classified into the following types: Kunpeng general computing-plus, Kunpeng memory-optimized, Kunpeng ultra-high I/O, and Kunpeng AI inference-accelerated ECSs. Displayed on the management console.
When ECSs are used as compute resources for gPaaS & AI DaaS services, the ECSs that meet the recycle bin policy will be moved to the recycle bin. This may lead to the failure of clearing resources of gPaaS & AI DaaS services.
