White Papers

ManualsBrandsDell ManualsConverged InfrastructureServers Solution Resources

PowerEdge Product Group

Direct from

Development

Elastic AI Infrastructure using Dell EMC PowerEdge and

Bitfusion FlexDirect

Tech Note by:

Ramesh Radhakrishnan, Dell EMC

Subbu Rama, Bitfusion

SUMMARY

Bitfusion FlexDirect disaggregates

GPU accelerators and re-

aggregates them in real-time over

Ethernet, Infiniband RDMA or

RoCE network, to create an

elastic AI infrastructure.

Just like network attached

storage, FlexDirect allows

customers to do network attached

GPUs.

FlexDirect on Dell EMC

PowerEdge servers offers a

seamless way for any machine in

the network to access any

arbitrary fraction of GPU or

multiple GPUs, anytime.

Application performance demands have increasingly been outpacing Moore’s

Law in a variety of fields, particularly AI and deep learning. Co-processors like

GPUs offer immense speedup to applications in fields like AI and deep learning,

compared to CPUs. AI and deep learning applications requires truly elastic

compute infrastructure from dev-test to model training and inference in order to

achieve high utilization of infrastructure resources. In organizations, GPU

accelerated servers are usually operated as silo-ed, stand-alone assets,

causing increased CAPEX and OPEX as well as slow datacenter

modernization.

The benefit of combining Dell EMC’s powerful portfolio of compute, storage and

networking with Bitfusion’s FlexDirect software allows our customers to

consolidate multiple silo-ed GPU clusters into a single shared resource pool, to

decrease CAPEX and OPEX as well as increase productivity.

Composable Elastic AI Compute Platform

Bitfusion FlexDirect enables GPUs to be available as first-class resource on any

machine in a PowerEdge Cluster that can be abstracted, partitioned, automated

and shared much like traditional compute or storage resource. GPU accelerators

can be partitioned into multiple virtual GPUs of any size and accessed remotely

by any machine, over the network. With this, GPU accelerators are now part of a

common infrastructure resource pool and available for use by anyone in the

environment.

Organizations can scale the operations with policies and business logic (time of

day policies, class of users, permission to access the top performance GPUs per

user class, etc.) for AI development and production use cases. GPUs from

different departments can be pooled to create bigger clusters to increase

compute performance and infrastructure utilization.

Figure 1: FlexDirect on Dell EMC PowerEdge Servers to create Elastic AI Infrastructure

Summary of content (3 pages)

PAGE 1
Direct from Development PowerEdge Product Group Elastic AI Infrastructure using Dell EMC PowerEdge and Bitfusion FlexDirect Tech Note by: Ramesh Radhakrishnan, Dell EMC Subbu Rama, Bitfusion SUMMARY Bitfusion FlexDirect disaggregates GPU accelerators and reaggregates them in real-time over Ethernet, Infiniband RDMA or RoCE network, to create an elastic AI infrastructure. Just like network attached storage, FlexDirect allows customers to do network attached GPUs.
PAGE 2
Increased Productivity across AI Development, Training and Deployment IT regains the ability to assign GPU resources based on organization business priorities, and remotely pool together resources, while attaching them in real-time to the workloads, with known schedule and utilization plan. For example, GPU resources from Department A which completed intensive training and development schedule, can be reassigned to Department B which now experiences peak demand for GPUs for an urgent AI project.
PAGE 3
PowerEdge Product Group Direct from Development Figure 3: Performance comparison of Network Attached Remote GPUs with FlexDirect on Dell EMC PowerEdge Servers vs. Native GPU execution Performance is represented on y-axis (images/sec, higher is better), for training different CNN models using Tensorflow. Figure 4: Fractional GPUs with FlexDirect on Dell EMC PowerEdge Servers vs. using a full GPU.