GPU-Ready Performance
Using the GPUDirect and RDMA protocols AI data servers powered by PEAK:AIO can deliver over 80GB/s to GPU servers.
PEAK:AIO AI Data Servers
Today’s AI servers consume and analyse data at much higher rates than many traditional storage solutions can deliver, resulting in low GPU utilisation and dramatically extending training times, decreasing productivity. Most AI projects evolve into small to mid-scale GPU clusters and not the super-clusters found in HPC environments.
However, to gain the performance needed to keep the GPUs busy and fed with data, users often feel the need to choose between feature-rich enterprise-class storage and parallel filesystems designed for HPC. Not any more - PEAK:AIO has developed a software platform from the ground up for AI workloads to complement the NVIDIA DGX, HGX, EGX and MGX range of AI appliances. Delivering ultra-low latency and tremendous bandwidth at a price which allows more investment to be made on GPU resource and less on legacy storage - ensuring your project’s funds are better spent and your team is more productive.
"PEAK:AIO simply allowed KCL and the NHS trials to invest more funds in GPU servers where they provide for the best outcome."
Dr Jorge Cardoso
Kings College London
"A key focus of PEAK:AIO was to not to simply repurpose existing storage technology and shoehorn it into an AI solution, we completely rebuilt the storage stack from the ground up to match the requirements of real-life AI projects. It was very clear, cost of legacy storage is prohibitive for most mid-scale GPU clusters, the balance of funds must be focused on GPU and not storage."
Mark Klarzynski
CEO & Founder
Using the GPUDirect and RDMA protocols AI data servers powered by PEAK:AIO can deliver over 80GB/s to GPU servers.
Using the industry-standard NFS3/4 protocols PEAK:AIO data servers are simple to connect and don’t require any complex drivers.
PEAK:AIO dispenses with the features that slow down legacy storage solutions, but retains RAID protection for your data.
As your AI projects evolve PEAK:AIO AI data servers grow alongside, scaling in capacity from as little as 30TB to up to 720TB per 2U.
PEAK:AIO AI data servers are much more cost-effective than legacy storage, enabling you to spend more on GPU servers.
PEAK:AIO AI data servers are designed for direct connection to GPU servers or small to mid-size clusters of up 10 GPU servers via InfiniBand or Ethernet switches. They can be configured using NVIDIA ConnectX-6 or ConnectX-7 Smart NICs delivering either 40GB/s or 80GB/s throughput over 200Gb or 400Gb connections respectively. Additionally, PEAK:AIO configurations start as small as your project needs and then scale as you need, removing the traditional requirement to over invest in storage at the onset. Selected AI data servers can be configured using Micron 9400 Pro NVMe SSDs featuring a 30.72TB capacity - the largest datacentre-grade SSD available to date.
Scan is direct partner of PEAK:AIO and is able to provide AI data servers specifically configured to your requirements. Scan also provides guided proof-of concept trials of our PEAK:AIO solutions.
Feature | Dell PowerEdge R7625 |
---|---|
Storage Capacity | Up to 720TB raw |
Bandwidth | 80GB/s |
Connectivity | 2 x QSP56 HDR InfiniBand / Ethernet 200Gb/s |
Software Platform | PEAK:AIO |
Export compatibility | NVMe-oF (InfiniBand or Ethernet) NFS (InfiniBand or Ethernet – RDMA or TCP) |
Software Licence | 3 years / 5 years on request |
Form Factor | 2U - 737mm deep |
Warranty | 3 / 5 years |
Support | 3 years 24/7 Premium Support |