KR4248V1

4U 4 GPU AI Server with AMD EPYC™ 7003

KR4248V1 is a 4U GPU server for HPC + AI computational optimization with a balanced design of high performance and high expansion. It supports two AMD EPYC™ 7003 Series processors and four powerful NVIDIA A100 tensor core GPUs connected at high speed with 600GB/s bandwidth through NVLink in 4U space. It delivers a high CPU:GPU performance ratio for HPC applications, robust storage and I/O expansion capabilities. This system provides an easy and cost-effective way for users to leverage the industry-leading SXM4 interface A100 GPU technology to build a powerful HPC/AI computing system.

Performance

2x 7nm AMD EPYC™ 7003 Series processors support advanced AMD 3D V-Cache™ Technology, up to 1536MB L3 Cache.

4 A100 SXM4 GPU via NVLink enabling 600GB/s P2P aggregate communication bandwidth and 320GB memory.


Scalability

Up to 32 DDR4 3200 memory system configurations provide high capacity and wide bandwidth data throughput for HPC and AI computing.

Robust storage and IO expansion can configure multiple high-speed networks and hard disk devices, and GPU Direct technology is supported to realize fast data transmission and network communication.


Reliability

The 19-inch standard rack server with 4U height supports stable operation at 35 ℃ and is more convenient for industrial standard data center deployment.

Support N + N redundancy hot plug titanium/platinum power supply, the power module efficiency is up to 96%, thoroughly ensuring the power supply quality and reliability of the system.


Ecosystem Support

The widely mature X86+CUDA global software ecosystem supports many HPC applications based on CUDA development optimization, such as VASP, GROMACS, and NAMD.

Support Tensorflow, Pytorch, Paddlepadle ,and other industry-leading deep learning frameworks, suitable for CV, transformer, recommendation, and other advanced AI algorithm scenarios.

Technical Specifications

Model KR4248V1
Processor 2x AMD EPYC™ 7003 Series processors, TDP up to 280W
GPU 4x SXM4 NVIDIA A100 Tensor Core GPU
Memory 32x DDR4 RDIMMs / LRDIMM, up to 3200MT / s
Storage 8x 2.5” SSD (up to 4 U.2 NVME)
2x NVMe/SATA M.2, rear 4* NVMe M.2
Front I/O 2x USB 3.0 port , 1* VGA port, 1* head phone port
Rear I/O 2x RJ45 Management port
I/O Expansion 5x PCIe 4.0 x16 slot
RAID 1x RAID Card, support RAID0/1/5/6/10, support Cache super capacitor protection
Fan N+1 Redundant hot swap fans
PSU 4x 3000W_80Plus Titanium/Platinum PSU, N+N Redundant
Management Built-in BMC remote management module, support Redfish/IPMI/SOL/KVM
OS Red Hat 8.3 64bit, CentOS 8.3, Ubuntu 20.04, Vmware 7.0u2 (or high edition)
Dimensions 448mm x 175.35mm x 835.39mm (W x H x D)
Weight Net weight 54kg (Cross weight: 82kg)
Environmental Parameters Working temperature: 10℃~35℃
Storage temperature: -40℃~70℃
Working humidity: 10%~80% R.H.
Storage humidity: 10%~93% R.H.
Working temperature at 0 to 1000 meters (3,300 feet): 0 ℃ to 40 ℃
Working temperature at 1000 to 3050 meters (10,000 feet): 5 ℃ ~ 32 ℃