과학 연산, 데이터 분석, 머신러닝 추론, 비디오 트크랜스코딩과 같은 난이도 높은 연산 처리가 부각되면서 데이터를 신속/정확하게 산출하여 다변화된 비즈니스 환경에 대응이 가능한 컴퓨팅 자원으로 2개의 GPU부터 10개 GPU까지 병렬 서버로 연결하는 최고의 자원을 선택할 수 있습니다.
HGX-ADG4U A100 40GB4
| HPC Server
| 2U DP SXM4 A100 4 GPU Server
| Application : AI , AI Training , AI Inference & HPC
● Supports NVIDIA HGX™ A100 with 4 x SXM4 GPU
● Supports NVIDIA® NVLink® technology
● Up to 600GB/s GPU to GPU interconnection
● Dual AMD EPYC™ 7002 series processor family
● 8-Channel RDIMM/LRDIMM DDR4 per processor, 16 x DIMMs
● 2 x 1Gb/s LAN ports (Intel® I350-AM2)
● 1 x dedicated management port
● 4 x 2.5" Gen4 U.2 NVMe/SATA/SAS hot-swappable HDD/SSD bays
● Ultra-Fast M.2 with PCIe Gen4 x4 interface
● 6 x Low profile Gen4 x16 expansion slots
● 1 x OCP 3.0 Gen4 x16 mezzanine slot
● 3000W 80 PLUS Platinum redundant power supply
AMD EPYC™ 7002 Series Processor (Rome)
The next generation of AMD EPYC has arrived, providing incredible compute, IO and bandwidth capability – designed to meet the huge demand for more compute in big data analytics, HPC and cloud computing.
● Built on 7nm advanced process technology, allowing for denser compute capabilities with lower power consumption
● Up to 64 core per CPU, built using Zen 2 high performance cores and AMD’s innovative chiplet architecture
● Supporting PCIe® 4.0 with a bandwidth of up to 64GB/s, twice of PCIe 3.0
● Embedded security protection to help defend your CPU, applications, and data
Per Socket & IO Bandwidth
4X Floating Point
vs AMD EPYC Naples
vs AMD EPYC Naples
System Block Diagram
High Density HPC
Supports NVIDIA HGX™ A100 4-GPU
Massive datasets, exploding model sizes, and complex simulations require multiple GPUs with extremely fast interconnections. NVIDIA HGX A100 combines NVIDIA A100 Tensor Core GPUs with high-speed interconnects to form the world’s most powerful servers. With A100 80GB GPUs, a single HGX A100 has up to 1.3 terabytes (TB) of GPU memory and over 2 terabytes per second (TB/s) of memory bandwidth, delivering unprecedented acceleration.
PCIe® 4.0 Ready
AMD EPYC Rome is ready to support PCIe® 4.0 with a bandwidth of 64GB/s, twice that of PCIe 3.0. This doubles the bandwidth available from the CPU to peripheral devices such as graphics cards, storage devices and high speed network cards. GIGABYTE’s AMD EPYC 7002 Series server platforms are ready to be used with a new generation of PCIe 4.0 devices such as AMD’s Radeon MI50 GPGPU.
* The PCIe 4.0 standard supports a 16 GT/s bit rate, (roughly 2GB/s per single lane) as opposed to 8GT/s bit rate for PCIe 3.0 (1GB/s).
Higher Memory Speed
AMD EPYC’s Rome 7002 Series processors feature faster 8 channel DDR4 memory lanes, supporting RDIMM or LRDIMM memory modules with a speed of up to 3200MHz (1 DIMM per channel)
More M.2 Storage
GIGABYTE’s AMD EPYC 7002 Series server platforms feature more M.2 drive capacity for ultra-fast NVMe storage – both onboard M.2 slots and extra capacity via optional riser cards.
OCP 3.0 Add-On Card Ready
GIGABYTE’s AMD EPYC Rome Server Platforms feature an onboard OCP 3.0 mezzanine slot for the next generation of PCI Gen 4.0 add on cards.
Compared to previous OCP 2.0 type cards, advantages of this new type include:
● Easier Serviceability : simply slot in / pull out the card without needing to open the server chassis; tool less design
● Larger Thermal Envelope : more space for heat sink provides an increased power budget for new & emerging capabilities
TPM 2.0 Module
GIGABYTE’s AMD EPYC Server is designed to support Trusted Platform Modules (TPM - discrete cryptographic on-board processors).
Carefully Selected Components
GIGABYTE’s passive components are carefully selected to guarantee a stable operating environment and allow the processors and the memory to deliver their maximum performance. GIGABYTE’s server motherboards are built with long lasting components for a high degree of reliability and durability.
Design Features for Easy Maintenance
GIGABYTE’s AMD EPYC 7002 Series server platforms contain many tool less design features for easy installation, maintenance and removal – never worry about forgetting your toolbox or losing a screw again!
Tool-less Rail Kits
Rail kits attach to the side of the server chassis via clipping mechanism.
Tool-less 2.5” & 3.5” Drive Bays
Clipping mechanism secures the drive in place. Install or replace a new drive in seconds.
Efficient, Reliable Power Supply
GIGABYTE’s AMD EPYC 7002 Series server are designed to guarantee high conversion efficiency regardless of load. Redundant 80 Plus ‘Titanium’ or ‘Platinum’ PSUs, combined with GIGABYTE’s precision engineering, translates to super efficient power usage which maximizes the system’s power : performance ratio and reduces OPEX for the user.
Intelligent Power Management Features
SCMP (Smart Crises Management / Protection)
SCMP is a GIGABYTE patented feature that will automatically force the system’s CPU to enter ULFM (ultra-low frequency mode for minimum power consumption) when the BMC is alerted to a PSU fault or error (such as power loss, power surge, overheating or a fan problem). This feature will prevent shutdown in systems with less than 1 + 1 PSU redundancy when one PSU is lost.
Automatic Fan Speed Control
To achieve the best cooling and power efficiency, GIGABYTE servers are enabled with automatic fan speed control. Individual fan speed will be automatically adjusted according to CPU, DIMM, M.2, HDD, and GPU temperature sensors. When BMC detects a temperature change at a particular location, the corresponding fan speed will be changed accordingly.
Certified Ready with Software Partners
Being a member of key software alliance partner programs enables GIGABYTE to rapidly develop and validate joint solutions, enabling our customers to modernize their data centers and implement IT infrastructure and application services with speed, agility, and cost optimization.