Narrow down by specifying conditions
現在1888件がヒットしています。check
InfiniBand provided by Mellanox, a network solution for supercomputing that we have been involved in for a long time. With the announcement of NVIDIA's acquisition of Mellanox, it is expected that the number of solutions that combine GPUs will increase in the future.
One of the reasons for this is the rapid increase in the use of GPUs in the field of AI, which has been used in various situations in recent years. We provide optimal hardware for AI/deep learning development by combining Supermicro's GPU server and Mellanox's InfiniBand, which we have supported for many years.
Proposal example
The number of businesses utilizing AI, deep learning, etc. is increasing. The advent of GPU servers has made it possible to easily perform AI, deep learning, and scientific calculations, which once could only be performed by supercomputers a decade ago.
We are receiving increasing requests from companies that develop their own AI algorithms and provide AI services to their customers to create their own AI infrastructure. At Macnica, we have prepared a proposal that can meet our customers' diverse GPU server specification requirements and provide low-latency and wide-bandwidth network infrastructure.
For GPU servers, we offer Supermicro products, which have a rich lineup to meet the different specifications of each customer. As for the network infrastructure, Macnica will propose Mellanox, which has the know-how accumulated through many years of supporting InfiniBand in the supercomputing market.
Also, by proposing GPU-Direct together, we can fully demonstrate the performance of the GPU server.
Supermicro GPU Server
- Rich GPU server portfolio
- 1U to 9U size, models available with 2 to 20 GPUs
- High speed fabric interface support (100-200Gbps InfiniBand)
Mellanox high-speed InfiniBand network
- Industry-leading broadband
- Low latency performance ・EDR (100Gbps), HDR (200Gbps) switches, adapter cards, cables, etc.
System level support from Macnica
- One-stop provision of maintenance and technical support including servers and networks
- Abundant technical know-how, construction know-how
Mellanox technology features
Only Network Vendor to Offer InfiniBand, the Most Adopted Network in the Supercomputing Market
We develop our own hardware and provide low-latency and high-bandwidth networks.
network switch
Overwhelming performance (HDR)
- 90ns latency
- 390M packets per sec (64B)
- 16Tb/s full non-blocking
Supports up to the latest HDR (200Gbps) *As of December 2019
- HDR switch
- 1U 200G x40 port
- Supports 100G x80 port expansion
- EDR switch
- 1U 100G x36 port
adapter card
- 0.6us latency
- 215Mpps packet transfer performance
- storage acceleration
- Supports RDMA, NVMeof
- High-speed communication between GPU nodes GPU-Direct compatible
GPU-Direct Overview and Benefits
By adopting GPU Direct technology to eliminate the need for CPU involvement and buffer copies in the communication loop, GPU-to-GPU communication can achieve high throughput and low latency.
Difference in latency with or without GPU-Direct
Red line: no GPU-Direct
Blue line: with GPU-Direct
Bandwidth difference with or without GPU-Direct
Red line: no GPU-Direct
Blue line: with GPU-Direct
Solve these problems!
- I want to create an AI infrastructure, but it is difficult to select hardware...
- I don't know how to use InfiniBand...
One-stop provision of hardware, software, and integration for AI platforms
In addition to system configuration proposals and technical support for GPU servers, network adapter cards, and network switches, we can also provide one-stop maintenance services.
