Not logged in » Login
Aug 02 2021

The FUJITSU Server PRIMERGY GX2570 M6 – a brand-new design to lead the GX portfolio

The FUJITSU Server PRIMERGY GX2570 M6 is an advanced dual socket rack server designed around a dedicated new generation dense GPU (8x) configuration, combined with high-performance computing and memory.  Optimized for accelerating high-grade AI (especially deep learning), data science, and HPC workloads, it is designed for enterprises looking to replace or supplant standard server units with a dedicated, no-compromise GPU-centric workload accelerator.

Image

 

In today’s hyper-competitive business workspace, success belongs to the firms which can run advanced workloads such as data science and analytics and artificial intelligence to derive business insights from their data. However, legacy IT infrastructures cannot easily manage the parallel processing that some of these complex software programs and algorithms require. GPU accelerated servers are the answer – they can utilize both traditional high-performance computing and memory architectures to “direct” the overall sequential data processing, while parceling out the workload’s parallel processing tasks to GPUs.

Fujitsu’s brand new PRIMERGY GX line of graphics processing unit (GPU) accelerated servers is designed around this complex, yet rewarding architectural approach. Within the all-new design of the GX line, the FUJITSU Server PRIMERGY GX2570 M6 takes GPU acceleration to all new levels!

 

A server design truly meant for true no-compromise performance

The GX2570 M6 delivers no-compromise performance, and its use cases include, but are not limitedImage to, high-grade AI, data science, and HPC workloads.

The advanced design delivers the highest levels of application performance via 2x 3rd Generation Intel® Xeon® Scalable Processors plus 2 TB memory (32x DIMM, 3200 MT/s) to drive 8x NVIDIA A100 Tensor Core GPUs (40 GB, with 80 GB on the roadmap), along with PCIe Gen4 technology and NVIDIA® NVLink® for high-throughput data transfer.

Multi-GPU capability particularly benefits deep learning workloads such as natural language processing and computer vision.

The PCIe Gen4 technology used by the server’s subsystems for data-exchange and for expansion is twice as fast as previous generation PCIe Gen3 systems. Furthermore, each 8x A100 GPU baseboard has 8x A100 Tensor-Core GPUs with 6x NVSwitch™ GPU-GPU Interconnect nodes, and each GPU has 12x NVLink™ ports. The NVSwitch™ node is a NVLink™ switch that connects to all 8x A100 GPUs. This mesh topology allows any A100 GPU to talk to any other A100 GPU at a bi-directional speed which is many times that of the fastest PCIe Gen4 x16 bus.

The GPU board is also connected directly to the 8x PCIe Gen4 expansion slots (the server has 10x total) which can field advanced HDR Network Interface Cards like the Infiniband 100/200 Gb cards capable of exchanging data at high speeds with GPUs from other systems without CPU overhead.

The PCIe Gen4 slots can also be used for high-cache memory equipped RAID/SAS controllers for NVMe and SAS, SATA HDD or SSD drives. The server can then field up to 10x NVMe drives for high-speed storage, or up to 6x NVMe/SAS/SATA devices in the front and 4x NVMe devices located at its rear.

To sum it up, the GX2570 M6 fields a powerful and fast design architecture so it can be a “no-compromise” node in a wider computing network with other servers including other GPU-accelerated GX2570 M6s or act as the “high-end” system to back up conventional units.

 

Tested and certified to deliver on application performance and customer ease-of-use

ImageThe GX2570 M6 is carefully designed to ensure that its applications can run without constraint. The “NGC-ready,” “NVQual” certifications ensure that the server is already tested for multiple use-cases.

Investment protection is enabled via the provision for N+1 PSU redundancy (in this case 3+1) – it allows a standby PSU to replace a failed PSU so that the server continues working without loss of valuable data or hardware damage. The server’s 4x 2200W Platinum Power supplies not only meet its demanding power requirements, but their high energy efficiency also optimizes costs over the server lifecycle.

For easy and effective server administration, the server has FUJITSU ISM whose functions include server status and event monitoring, update, inventory and archive management, logging and auditing, floor layout and rack-view – all via an easy-to-use user interface.

In summary, the GX2570 M6 offers true high-end GPU-accelerated performance, without compromising on any of the flexible and customer-friendly design features that the PRIMERGY server line is reputed for.

And FUJITSU will continue to enhance the server with further technology upgrades to make sure it remains at the cutting edge!

 

Nitin Sitaram

 

About the Author:

Nitin Sitaram

Specialist Marketing Manager, Product Data Center Server, Global Marketing

SHARE

Comments on this article

No comments yet.

Please Login to leave a comment.

X

Please login

Please log in with your Fujitsu Partner Account.

Login


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now