A Rack-Scale Composable Infrastructure Solution That Delivers the Unlimited Flexibility and Agility of the Cloud, at a Fraction of the Cost
- Orchestrate ANY Compute, Acceleration (CPUs, GPUs, FPGAs, ASICs), Storage, Memory (3D-XPoint) or Networking resource for any workload using an Enterprise-Class, Easy-To-Use and Open Standards high-performance network.
- Deploy, expand, reduce or replace all rack resources dynamically in real time.
- Fully automate on-premise bare-metal cloud-class infrastructure, delivering higher throughput at a fraction of the cost.
Rack-scale Composable Infrastructure: How it works
Buy only what you need,
when you need it
Choose the best hardware/software
without vendor lock-in
Compose on the fly for
- Increase utilization and eliminate over provisioning of endpoints.
- Run on standard commodity servers.
- Buy only what you need, when you need it.
- Run fewer resources thus reduce cooling and save space.
- Save considerable system admin costs by deploying faster, and completely automating redeployment.
your power bill
- Consume less with fewer power-hungry servers and accelerators for the same performance.
- With less hardware to run, become a greener data center.
Yet Without Any Performance Penalty
Find Out How Much You Can Save With GigaIO
Improve GPU utilization by more than 50%
A typical customer problem we solve:
You just spent $400K on a rack of servers with some expensive GPUs to solve your computational problem. You don’t know it yet, but your GPUs are mostly sitting idle, even with your finely tuned models and lots of developer time in optimization. GigaIO solves this problem with FabreX, the highest performance, lowest latency rack-scale network on the planet. In one side by side test for a customer, a rack built with FabreX, ran the same customer weekly workload with 16 GPUs instead of the 32 initially deployed. The result is a system that costs 30% less to purchase, and to operate, and saves 5kW of energy as a bonus.
Why Pay More and Get Less?
Actual Case Study: 8 workgroups, using GPUs, 150 mixed jobs per week.
FabreX Composable Infrastructure
System Purchase Price: $290K
Power Consumption: 11.7 kW
Total Operation Expense per yr: $18,308
InfiniBand Converged Infrastructure
System Purchase Price: $412K
Power Consumption: 15.9 kW
Total Operation Expense per yr: $24,866
Other Benefits of Composable Infrastructure:
- Ability to use less expensive servers to manage your GPUs
- Protected investment: Separate obsolescence and depreciation schedules for servers and accelerators
- No forklift upgrade to incorporate the latest GPUs or FPGAs
- Future-proofing: Less risk of your entire system becoming obsolete as your AI algorithms change
- Decreased upgrade costs: upgrade only the resources you need to, with no change to applications
- Decreased admin costs: no rewiring/reinstall
The Industry’s Only Open and Comprehensive Composable Infrastructure Solution
Only FabreX PCIe Delivers Cloud-Class™ Composition
Lowest latency and highest rack-level bandwidth
- PCIe throughout
- Bandwidth to 256Gb/sec (Gen 4 half duplex)
- Adapter card latency 100 nsec
- Switch latency 150 nsec
- Total end to end latency 350 nsec
Fully managed environment for full visibility and monitoring and control of all resources
- Full telemetry data throughout the network
- Leading BMC
- Managed enclosures,
- Auto-discovery throughout the network
Standards-based open platform for no vendor lock-in
- Full RedFish APIs
Tight integration with certified ready-to-run enterprise-class orchestration
- Award winning, full enterprise capability
- Full orchestration
- Create virtual servers on the fly
- Military level security
- Sophisticated access control
- Our host software enables server-to-server communication over FabreX for protocols such as NVMe-oF, MPI, Libfabric, and TCP/IP. It is open-source, supports all popular Linux installations and can be readily downloaded from our support portal.
- Our switch software engine drives the performance and dynamic composability of GigaIO’s composable disaggregated infrastructure (CDI) for enterprise data centers and high-performance computing environments.
- Together with our cards and cables, the FabreX Switch is the fundamental building block of the FabreX network for true CDI
- Supports bandwidth up to 512Gb/s per port full duplex
Managed Accelerator Pooling Appliance
This expansion chassis is a rack-mount, disaggregated compute accelerator enclosure with space and power for up to 10 PCIe Gen 4 x16 accelerator cards.
What They Say About Us
Note: This article originally appeared on theregister.com, click here to read the original piece. Regardless, we’re looking forward toRead More
Note: This article originally appeared on hpcwire.com, click here to read the original piece. SAN DIEGO, Calif., June 1,Read More
Note: This article originally appeared on hpcwire.com, click here to read the original piece. SAN DIEGO, May 19, 2022Read More