Intel® Select Solutions for AI Inferencing are turnkey platforms that provide pre-bundled, verified, and optimized solutions for low-latency, high throughput inference performed on a CPU, not on a separate accelerator card.
Ingredient |
Intel Select Solutions for AI Inferencing v2 Base Configuration |
Intel Select Solutions for AI Inferencing v2 Plus Configuration |
|---|---|---|
Number of Nodes |
Single-node configuration |
Single-node configuration |
Processor |
2 x Intel® Xeon® Gold 6248 processor (2.50 GHz, 20 cores, 40 threads), or a higher number Intel® Xeon® Scalable processor |
2 x Intel® Xeon® Platinum 8268 processor (2.90 GHz, 24 cores, 48 threads), or a higher number Intel Xeon Scalable processor |
Memory |
192 GB or higher (12 x 16 GB 2,666 MHz DDR4 ECC RDIMM) |
384 GB (12 x 32 GB 2,934 MHz DDR4 ECC RDIMM) |
Boot Drive |
1 x 256 GB Intel® SSD DC P4101 Series (M.2 80 mm PCIe* 3.0 x4, 3D2, TLC) or higher |
1 x 256 GB Intel SSD DC P4101 Series (M.2 80 mm PCIe* 3.0 x4, 3D2, TLC) or higher |
Storage |
Data drive: 1.6 TB NVM Express* (NVMe*) Intel® SSD DC P4510 Series Cache drive: 375 GB Intel® Optane™ SSD DC P4800X U.2 NVMe* SSD |
Data drive: 1.6 TB NVMe* Intel SSD DC P4510 Series Cache drive: 375 GB Intel Optane SSD DC P4800X U.2 NVMe* SSD |
Data Network |
1 x Intel® Ethernet Converged Network Adapter (Intel® Ethernet CNA) XXV710-DA2 SFP28 DA Copper PCIe* x 8 dual-port 25/10/1 GbE |
1 x Intel Ethernet Converged Network Adapter (Intel Ethernet CNA) XXV710-DA2 SFP28 DA Copper PCIe* x 8 dual-port 25/10/1 GbE |
Management Network |
Integrated 1 GbE port 0/RMM port |
Integrated 1 GbE port 0/RMM port |
Software |
||
Linux OS |
CentOS Linux release 7.6.1810/Red Hat Enterprise Linux* (RHEL) 7 |
CentOS Linux release 7.6.1810/Red Hat Enterprise Linux* (RHEL) 7 |
Intel® Math Kernel Library (Intel® MKL) |
Intel Math Kernel Library (Intel MKL) version 2019 Update 4 |
Intel Math Kernel Library (Intel MKL) version 2019 Update 4 |
Intel® Distribution of OpenVINO™ Toolkit |
2019 R1.0.1 |
2019 R1.0.1 |
OpenVINO™ Model Server |
0.4 |
0.4 |
TensorFlow |
1.14 |
1.14 |
PyTorch |
1.2.0 |
1.2.0 |
MXNet |
1.3.1 |
1.3.1 |
Intel® Distribution for Python* |
2019 Update 1 |
2019 Update 1 |
Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN) |
0.19 (implied by the OpenVINO™ toolkit) |
0.19 (implied by the OpenVINO toolkit) |
Deep Learning Reference Stack (DLRS) |
v4.0 |
v4.0 |
Source-to-Image |
1.1.14 |
1.1.14 |
Docker |
18.09 |
18.09 |
Kubernetes |
v1.15.3 |
v1.15.3 |
Kubeflow |
v0.6.1 |
v0.6.1 |
Helm |
2.14.3 |
2.14.3 |
Seldon Core |
0.3.2 |
0.3.2 |
Ceph |
v14.2.1 |
v14.2.1 |
Min.io (Rook v1.0) |
RELEASE.2019-04-23T23-50-36Z |
RELEASE.2019-04-23T23-50-36Z |
Rook |
1.0.5 |
1.0.5 |
Other |
||
Trusted Platform Module (TPM) |
TPM 2.0 |
TPM 2.0 |
Minimum Performance Standards Verified to meet or exceed the following minimum performance capabilities: |
||
Classification Using ResNet-50 on OpenVINO Toolkit |
1,900 images per second (91 percent top-5 accuracy) |
2,650 images per second (91 percent top-5 accuracy) |
| Scaling in Emulated Real-World Scenario from 1 Node to 2 Nodes | Up to 1.91x1 | Up to 1.91x2 |
Business Value of Choosing a Plus Configuration Over a Base Configuration |
The Plus configuration provides up to 39 percent faster inferencing performance.1 |
|
| **Recommended, not required | ||
Testing conducted by Intel on October 9, 2019. Test configuration: Two nodes, 2 x Intel® Xeon® Gold 6248 processor (2.50 GHz, 20 cores, 40 threads), 12 x 16 GB 2,666 MHz DDR4 ECC RDIMM (192 GB total memory), boot drive: 1 x 256 GB Intel® SSD DC P4101 Series (M.2 80 mm PCIe* 3.0 x4, 3D2, TLC), data drive: 1.6 TB NVM Express* (NVMe*) Intel® SSD P4510 Series, cache drive: 375 GB Intel® Optane™ SSD DC P4800X U.2 NVMe* SSD, data network: 1 x 10Gb Intel® Ethernet Converged Network Adapter X722 (Intel® Ethernet CNA X722), management network: Integrated 1 gigabit Ethernet (GbE) port 0/RMM port. Software: CentOS* Linux release 7.6.1810/Red Hat Enterprise Linux* (RHEL) 7, Intel® Math Kernel Library (Intel® MKL) version 2019 update 4, Intel® Distribution of OpenVINO™ toolkit 2019 R1.0.1, OpenVINO™ model server 0.4, Intel® Distribution for Python* 2019 update 1, Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN) 0.19, Deep Learning Reference Stack (DLRS) v4.0, Docker v18.09, Helm v2.14.3, Kubernetes v1.15.3, Kubeflow v0.6.1, Seldon Core v0.3.2, Rook v1.0.5, Ceph v14.2.1, Min.io (Rook v1.0) RELEASE.2019-04-23T23-50-36Z. Scaling in emulated real-world scenario—throughput test: Normalized performance 1 (Intel® Hyper-Threading Technology (Intel® HT Technology): Off).
Testing conducted by Intel on October 9, 2019. Test configuration: Two nodes, 2 x Intel® Xeon® Platinum 8268 processor (2.90 GHz, 24 cores, 48 threads), 12 x 16 GB 2,666 MHz DDR4 ECC RDIMM (192 GB total memory), boot drive: 1 x 256 GB Intel® SSD DC P4101 Series (M.2 80 mm PCIe* 3.0 x4, 3D2, TLC), data drive: 1.6 TB NVM Express* (NVMe*) Intel® SSD P4510 Series, cache drive: 375 GB Intel® Optane™ SSD DC P4800X U.2 NVMe* SSD, data network: 1 x 10Gb Intel® Ethernet Converged Network Adapter X722 (Intel® Ethernet CNA X722), management network: Integrated 1 gigabit Ethernet (GbE) port 0/RMM port. Software: CentOS* Linux release 7.6.1810/Red Hat Enterprise Linux* (RHEL) 7, Intel® Math Kernel Library (Intel® MKL) version 2019 update 4, Intel® Distribution of OpenVINO™ toolkit 2019 R1.0.1, OpenVINO™ model server 0.4, Intel® Distribution for Python* 2019 update 1, Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN) 0.19, Deep Learning Reference Stack (DLRS) v4.0, Docker v18.09, Helm v2.14.3, Kubernetes v1.15.3, Kubeflow v0.6.1, Seldon Core v0.3.2, Rook v1.0.5, Ceph v14.2.1, Min.io (Rook v1.0) RELEASE.2019-04-23T23-50-36Z. Scaling in emulated real-world scenario—throughput test: Normalized performance 1.91 (Intel® Hyper-Threading Technology (Intel® HT Technology): Off).