NEC
Request a quoteQuote
NEC SX-Aurora Vector Cards

Vector computing

NEC SX-Aurora Vector Cards

Ideal for AI, big data, and system embedded applicationsApplication fields: product inspection, medical image processing, demand forecasting, weather forecasting, radar analysis, etc.

  • Type 20B-PModel Name
  • 1.600Operating frequency (GHz)
  • 307Theoretical operation performance (GFLOPS) Double precision
Request a quoteContact

VE-Card-Data

Data Center Model

Model Overview

Ideal for large scale processing in data centers and compute centers

Application Field: Large-scale AI/big data, large-scale simulation, etc.< 2U Rackmount Model

Model Image
Model Name B401-8
Number of Vector Engines (VE) installed 8
Form Factor 2U rack mount

Vector Engine (VE)

Mainly performs application arithmetic processing.

Number of Vector Engines (VE) that can be installed 8
Model Name B401-8
Card Type 20A/20B
Maximum total VE Operation Performance (TFLOPS) Double Precision 24.57
Maximum total VE Operation Performance (TFLOPS) Single Precision 49.15
Maximum total VE Memory Bandwidth (TB/s) 12.24
Maximum VE Memory Capacity (GB) 384

Vector Host (VH)

Mainly performs OS processing.

Number of x86 Processors/VH 1
x86 Processor AMD EPYC™ processor
Memory Configuration/VH DDR4 DIMM x 8 / CPU
Maximum Memory Capacity/VH 512GB
Maximum disk capacity SATA SSD 1.92TB
InfiniBand HDR
Number of HCAs 2
Bandwidth per HCA 200Gbps
OS Click here for supported OS

Installation Specifications

Power 4.0kW
Number of power supplies 2
Power cable connector shape AC200V power cable IEC C13 IEC C14
Cooling Water cooling + Air cooling
Operating Temperature 10°C to 40°C
Recommended Operating Temperature 10°C to 40°C
Volume (sound pressure level) 68.1 dBA

Software

By using NEC's original compiler the Vector Engine is easy to use.

Standard Software Vector Engine control software and driver
Software Development Support NEC Software Development Kit for Vector Engine, NEC MPI, NEC MPI/Scalar-Vector Hybrid

VE-Card-Onsite

On-Site Model

Model Overview

Ideal for simulations, AI, and big data applications in manufacturing and other industries

Application Field: Large-scale demand forecasting, manufacturing simulation, etc.

Model Image
Model Name A311-4
Number of Vector Engines (VE) installed 2, 4
Form Factor 1U rack mount

Vector Engine (VE)

Mainly performs application arithmetic processing.

Number of Vector Engines (VE) that can be installed 2,4
Card Type 10BE
Maximum total VE Operation Performance (TFLOPS) Double Precision 8.65
Maximum total VE Operation Performance (TFLOPS) Single Precision 17.30
Maximum total VE Memory Bandwidth (TB/s) 5.40
Maximum VE Memory Capacity (GB) 192

Vector Host (VH)

Mainly performs OS processing.

Installation Specifications

Number of x86 Processors/VH 2
x86 Processor Intel Xeon Scalable Processor®
Memory Configuration/VH DDR4 DIMM x 6 / CPU
Maximum memory capacity/VH 192GB
Maximum Disk Capacity SATA SSD 1.92TB
InfiniBand HDR100
Number of HCAs 2
Bandwidth per HCA 100Gbps
OS Click here for supported OS
Number of x86 Processors/VH 2
x86 Processor Intel Xeon Scalable Processor®
Memory Configuration/VH DDR4 DIMM x 6 / CPU
Maximum memory capacity/VH 192GB
Maximum Disk Capacity SATA SSD 1.92TB
InfiniBand HDR100
Number of HCAs 2
Bandwidth per HCA 100Gbps
OS Click here for supported OS

Software

By using NEC's original compiler the Vector Engine is easy to use.

Standard Software Vector engine control software and driver
Software Development Support NEC Software Development Kit for Vector Engine, NEC MPI, NEC MPI/Scalar-Vector Hybrid

NEC SX-Aurora TSUBASA - Vector Engine

NEC Vector Engine Models

NEC SX Aurora Card

The latest generation of the Vector Engine Processor now has 16 cores and a memory bandwidth of 2.45 TB/s and a DP peak performance of 4.91 TeraFLOPS.

NEC Vector Engine Type 30A
Core specfication   
Clock speed (GHZ) 1.600
Peak performance (GigaFLOPS) DP 307
Peak performance (GigaFLOPS) SP 614
Processor Specifications   
# of cores / processor 16
Peak performance (TeraFLOPS) DP 4.91
Peak performance (TeraFLOPS) SP 9.83
Memory bandwidth (TB/s) 2.45
Cache capacity (MB) 64
Memory capacity (GB) 96

The second generation of the Vector Engine Processor has a higher memory bandwidth of 1.53 TB/s and a DP peak performance of 3.07 TeraFLOPS.

NEC Vector Engine Type 20A Type 20B
Core specfication   
Clock speed (GHZ) 1.600 1.600
Peak performance (GigaFLOPS) DP 307 307
Peak performance (GigaFLOPS) SP 614 614
Processor Specifications   
# of cores / processor 10 8
Peak performance (TeraFLOPS) DP 3.07 2.45
Peak performance (TeraFLOPS) SP 6.14 4.91
Memory bandwidth (TB/s) 1.53 1.53
Cache capacity (MB) 16 16
Memory capacity (GB) 48 48

The NEC Vector Engine Processor was developed using 16 nm FinFET process technology for extreme high performance and low power consumption. The Vector Engine Processor had the world's first implementation of one processor with six HBM2 memory modules using Chip-on-Wafer-on-Substrate technology, leading to an outstanding memory bandwidth of 1.35 TB/s at that time.

NEC Vector Engine Type 10AE Type 10CE
Core specfication   
Clock speed (GHZ) 1.584 1.400
Peak performance (GigaFLOPS) DP 304 268
Peak performance (GigaFLOPS) SP 608 537
Processor Specifications   
# of cores / processor 8 8
Peak performance (TeraFLOPS) DP 2.43 2.15
Peak performance (TeraFLOPS) SP 4.86 4.30
Memory bandwidth (TB/s) 1.35 1.00
Cache capacity (MB) 16 16
Memory capacity (GB) 48 24

The NEC Vector Engine Processor follows the tradition of NEC Vector Processors providing optimal sustained performance.

NEC SX-Aurora TSUBASA Platforms

NEC provides hardware platforms that can be configured in different ways.

For more detailed information to the systems

VH-Selection

This is a list of servers equipped with Vector Engine that have been confirmed to work.

OEM/System Vendor Server model Server PCIe Specs VE type Number of VE's Test server Evaluation Company
Configuration

Recommended temperature during test configuration
Dell Technologies Japan Inc. PowerEdge R7525 Rack Server Gen4（VE works with Gen3） 20B-P 3 3x Vector Engine Type 20B-P
2x AMD EPYC 7H12
16x 32GB DDR4-32002x SSD SAS 1.92TBGPU Power Cable Install KitiDRAC9, Datacenter		 32℃ Dell Technologies Japan Inc.
Dell Technologies Japan Inc. PowerEdge R750 Rack Server Gen4（VE works with Gen3） 20B-P 2 2x Vector Engine Type 20B-P
2x Intel Xeon Platinum 8358P
16x 32GB RDIMM6x SSD SAS 1.6TBGPU Power Cable Install KitiDRAC9, Datacenter		 32℃ Dell Technologies Japan Inc.
Skymind Holdings Bhd. Axon 3 Gen3 20B-P 1 1x Vector Engine Type 20B-P
1x NVIDIA 2080 Ti
2x Intel Xeon Silver 42106x 16GB DDR4-24001x NVMe SSD 970 GB1x 4TB SATA HDD2x Cooling Fan Adapter for Vector Engine		 27℃ Skymind Holdings Bhd.
Skymind Holdings Bhd. Axon 4 Gen3 20B-P 2 2x Vector Engine Type 20B-P
2x Xeon Silver 4216
4x 32GB DDR4-29331x NVMe SSD 1TB(Micron)1x 2TB SATA HDD (Seagate Exos E2000)		 29℃ Skymind Holdings Bhd.
Super Micro Computer, Inc. SYS-7049GP-TRT Gen3 20B-P 4 4x Vector Engine Type 20B-P
2x Intel Xeon Gold 6226
12x 16GB DDR4-29331x SATA HDD 1TB2x additional rear fan for passive cooling GPU		 29℃ ViSUAL TEChNOLOGY, Inc.
GIGA-BYTE Technology Co., Ltd. G242-Z11 Gen4（VE works with Gen3） 20B-P 2 1x AMD EPYC 7702P
8x 16GB DDR4-3200
1x 960GB SATA SSD 2x Vector Engine Type 20B-P		 25℃ NEC Corporation
GIGA-BYTE Technology Co., Ltd. G492-Z50 Gen4 (VE works with Gen3） 20B-P 8 2x AMD EPYC 7402 24-Core Processor
32x 32GB DDR4-3200
2x 960GB SATA SSD8x Vector Engine Type 20B-P		 32℃ NEC Corporation
Super Micro Computer, Inc. SYS-1029GQ-TRT Gen3 20B-P 4 4x Vector Engine Type 20B-P
2x Intel Xeon Gold 6248
12x 16GB DDR4-29332x SATA SSD 960GB2x IB HCA card HDR100		 30℃ NEC Corporation
Super Micro Computer, Inc. SYS-4029GP-TRT2 Gen3 20B-P 8 8x Vector Engine Type 20B-P
2x Intel Xeon Gold 6226
12x 16GB DDR4-29332x SATA SSD 960GB2x NVMe U.2 1.6TB 2x IB HCA card HDR100		 30℃ NEC Corporation

Disclaimer Notice

  • – This list only provides the information of servers which tun with Vector Engine Type 20B-P.
  • – This list and information are provided solely for your reference purpose only and NEC may make changes to any contents at any time, without notice.
  • – This list and any information are provided "as is" without warranty of any kind, including but not limited to, the warranty for merchantability, fitness for a particular purpose and non-infringement.
  • – NEC disclaims any and all warranties for the contents of this list and any information.
  • – NEC makes no warranty that the Vector Engine will operate on the servers listed in this table and shall not be responsible or liable for any damages arisisng out from such combination.
  • – You shall bear the entire liability and cost for any problem or trouble relating to this list and any information.
  • – In no event shall NEC be liable for any direct, indirect, consequential or special damages or loss of profit, or damages related to the use or the result of use of this list and any information, even if NEC should have predicted, or has been advised of, the possibility of such damages.
  • – Unauthorized modification or reproduction of content or the complete list is prohibited.

Vector-Engine-Types

The cores follow the big-core design philosophy of the previous SX, achieving outstanding computing performance and memory bandwidth per single core. With up to 12 vector processor cores, 4.9TF of computing performance and an outstanding memory bandwidth of 2.45TB/s per processor, applications can be executed at ultra-high speeds.

  • Type 30A
  • Type 20A
  • Type 20B
  • Type 10AE
  • Type 10CE

Type 30A

Basic Core Specifications

Operating Frequency (GHz) 1.600
Theoretical Operation Performance (GFLOPS) Double Precision 307
Theoretical Operation Performance (GFLOPS) Single Precision 614

Basic Processor Specifications

Number of Cores/Processor 16
Theoretical Operation Performance (TFLOPS) Double Precision 4.915
Theoretical Operation Performance (TFLOPS) Single Precision 9.83
Memory Bandwidth (TB/s) 2.45
Cache Capacity (MB) 64
Memory Capacity (GB) 96

Type 20A

Basic Core Specifications

Operating Frequency (GHz) 1.600
Theoretical Operation Performance (GFLOPS) Double Precision 307
Theoretical Operation Performance (GFLOPS) Single Precision 614

Basic Processor Specifications

Number of Cores/Processor 10
Theoretical Operation Performance (TFLOPS) Double Precision 3.07
Theoretical Operation Performance (TFLOPS) Single Precision 6.14
Memory Bandwidth (TB/s) 1.53
Cache Capacity (MB) 16
Memory Capacity (GB) 48

Type 20B

Basic Core Specifications

Operating Frequency (GHz) 1.600
Theoretical Operation Performance (GFLOPS) Double Precision 307
Theoretical Operation Performance (GFLOPS) Single Precision 614

Basic Processor Specifications

Number of Cores/Processor 8
Theoretical Operation Performance (TFLOPS) Double Precision 2.45
Theoretical Operation Performance (TFLOPS) Single Precision 4.91
Memory Bandwidth (TB/s) 1.53
Cache Capacity (MB) 16
Memory Capacity (GB) 48

Type 10AE

Basic Core Specifications

Operating Frequency (GHz) 1.584
Theoretical Operation Performance (GFLOPS) Double Precision 304
Theoretical Operation Performance (GFLOPS) Single Precision 608

Basic Processor Specifications

Number of Cores/Processor 8
Theoretical Operation Performance (TFLOPS) Double Precision 2.43
Theoretical Operation Performance (TFLOPS) Single Precision 4.86
Memory Bandwidth (TB/s) 1.35
Cache Capacity (MB) 16
Memory Capacity (GB) 48

Type 10CE

Basic Core Specifications

Operating Frequency (GHz) 1.400
Theoretical Operation Performance (GFLOPS) Double Precision 268
Theoretical Operation Performance (GFLOPS) Single Precision 537

Basic Processor Specifications

Number of Cores/Processor 8
Theoretical Operation Performance (TFLOPS) Double Precision 2.15
Theoretical Operation Performance (TFLOPS) Single Precision 4.30
Memory Bandwidth (TB/s) 1.00
Cache Capacity (MB) 16
Memory Capacity (GB) 24

Interested in this product?

Request a quote: our team will confirm availability, compatibility and a deployment plan.

Request a quote
✉
WhatsAppCall usRequest a quote