All Documents
Current Document
Content is empty
If you don't find the content you expect, please try another search term
Content is empty
If you don't find the content you expect, please try another search term
Last updated：2021-03-19 14:46:28
Kingsoft Cloud provides a variety of GEC instance types for typical application scenarios. Different types of GEC instances use different hardware (GPU, CPU, memory, and disk) and network resource configurations. This topic details the application scenarios, models, and configuration information of GEC instance types.
The following table describes GEC instance types.
| GEC category | Instance type | Application scenario |
|---|---|---|
| Passthrough | Common training and inference scenarios such as deep learning, voice, and graphics and image learning | |
| vGPU | Cloud rendering and small-scale, elastic and flexible AI application scenarios |
GN6I GEC instances are applicable to inference scenarios and simple training scenarios.
Based on NVIDIA Tesla T4, each GPU has 16 GB GDDR6 graphic memory, 8.1 TFLOPS single precision FP32 computing capability, and 130 TOPS INT8 computing capability.
Instance characteristics:
The following table describes the GN6I GEC types and specifications.
| Type | GPU | GPU graphic memory (GDDR6) | vCPUs | Memory (GiB) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) | Queues |
|---|---|---|---|---|---|---|---|
| GN6I.4A1 | T4*1 | 16 GB*1 | 4 | 16 | 50 | 4 | 2 |
| GN6I.8A1 | T4*1 | 16 GB*1 | 8 | 32 | 80 | 5 | 2 |
| GN6I.16A1 | T4*1 | 16 GB*1 | 16 | 64 | 120 | 6 | 4 |
| GN6I.16B2 | T4*2 | 16 GB*2 | 16 | 64 | 120 | 6 | 4 |
| GN6I.32B2 | T4*2 | 16 GB*2 | 32 | 128 | 240 | 8 | 8 |
| GN6I.32C4 | T4*4 | 16 GB*4 | 32 | 128 | 240 | 8 | 8 |
P3 GEC instances are applicable to deep learning training scenarios and inference scenarios.
Based on NVIDIA Tesla P40, each GPU has 24 GB DDR5 GPU memory, 12 TFLOPS single precision FP32 computing capability, and 46 TOPS INT8 computing capability.
Instance characteristics:
The following table describes the P3 GEC types and specifications.
| Type | GPUs (Tesla P40) | GPU graphic memory (GDDR5) | vCPUs | Memory (DDR4) | Data disk (local SSD) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|---|
| P3.28A1 | 1 | 24 GB*1 | 28 | 56 GB | 1,000 GB | 30 | 3 |
| P3.56B2 | 2 | 24 GB*2 | 56 | 112 GB | 2,000 GB | 40 | 6 |
| P3.56C4 | 4 | 24 GB*4 | 56 | 224 GB | 4,000 GB | 40 | 8 |
P3I GEC instances are applicable to inference and prediction scenarios such as speech recognition, speech synthesis, and image recognition.
Based on NVIDIA Tesla P4, each GPU has 8 GB DDR5 GPU memory, 5.5 TFLOPS single precision FP32 computing capability, and 22 TOPS INT8 computing capability. In deep learning inference and prediction scenarios, the latency of a GPU instance is 15 times lower than that of a CPU, and the throughput is 60 times higher.
Instance characteristics:
The following table describes the P3I GEC types and specifications.
| Type | GPU (Tesla P4) | GPU graphic memory (GDDR5) | vCPUs | Memory (DDR4) | Data disk (local SSD) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|---|
| P3I.14B1 | 1 | 8 GB*1 | 14 | 120 GB | 500 GB | 20 | 3 |
| P3I.28C2 | 2 | 8 GB*2 | 28 | 240 GB | 1,000 GB | 30 | 6 |
Instance characteristics:
P3IN GEC instances use the same hardware and are applicable to the same scenarios as P3I GEC instances. The following table describes the P3IN GEC types and specifications.
| Type | GPU (Tesla P4) | GPU graphic memory (GDDR5) | vCPUs | Memory (DDR4) | Data disk (local SSD) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|---|
| P3IN.4A1 | 1 | 8 GB*1 | 4 | 16 GB | 120 GB | 10 | 1.5 |
| P3IN.8B1 | 1 | 8 GB*1 | 8 | 32 GB | 180 GB | 20 | 1.5 |
| P3IN.16C2 | 2 | 8 GB*2 | 16 | 64 GB | 360 GB | 30 | 3 |
| P3IN.32D4 | 4 | 8 GB*4 | 32 | 128 GB | 720 GB | 40 | 6 |
P4V GEC instances are applicable to deep learning training scenarios and inference scenarios.
Based on NVIDIA Tesla V100, each GPU has 16 GB HBM2 GPU memory, 15 TFLOPS single precision FP32 computing capability, and 125 TFLOPS mixed-precision computing capability.
Instance characteristics:
The following table describes the P4V GEC types and specifications.
| Type | GPU (Tesla V100) | GPU graphic memory (HBM2) | vCPUs | Memory (DDR4) | Data disk (local SSD) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|---|
| P4V.8A1 | 1 | 16 GB*1 | 8 | 32 GB | 240 GB | 20 | 1.5 |
| P4V.16B2 | 2 | 16 GB*2 | 16 | 64 GB | 480 GB | 30 | 3 |
| P4V.28C4 | 4 | 16 GB*4 | 28 | 128 GB | 960 GB | 30 | 6 |
| P4V.56D8 | 8 | 16 GB*8 | 56 | 256 GB | 1,920 GB | 40 | 8 |
vGN5 GEC instances are applicable to the following scenarios:
Instance characteristics:
The following table describes the vGN5 GEC types and specifications.
| Type | GPU (Tesla P4) | GPU graphic memory (GDDR5) | vCPUs | Memory (DDR4) | Data disk (local SSD) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|---|
| vGN5.vCS-8B2 | 1/2 | 4 GB | 8 | 48 GB | 400 GB | 20 | 2 |
| vGN5.vPC-4C4 | 1/4 | 2 GB | 4 | 24 GB | 200 GB | 10 | 1 |
| vGN5.vPC-2D8 | 1/8 | 1 GB | 2 | 12 GB | 100 GB | 10 | 1 |
vCS is applicable for CUDA computing such as AI inference, and vPC is applicable for graphics and image processing. For more information about vGN5 configuration, see vGPU user guide.
vGN6 GEC instances are applicable to the following scenarios:
Instance characteristics:
The following table describes the vGN6 GEC types and specifications.
| Type | GPU (Tesla T4) | GPU graphic memory (GDDR6) | vCPUs | Memory (DDR4) | Packet sending/receiving capability (10,000 pps) | Bandwidth (Gbps) |
|---|---|---|---|---|---|---|
| vGN6.vCS-10B2 | 1/2 | 8 GB | 10 | 40 GB | 80 | 3 |
| vGN6.vCS-4C4 | 1/4 | 4 GB | 4 | 20 GB | 50 | 2 |
| vGN6.vPC-2D8 | 1/8 | 2 GB | 2 | 10 GB | 30 | 1 |
vCS is applicable for CUDA computing such as AI inference, and vPC is applicable for graphics and image processing. For more information about vGN6 configuration, see vGPU user guide.
Pure Mode
