Manticore: Datacenter-Class AI Compute for the Near Edge

Manticore: Datacenter-class AI and HPC at the Edge.

Manticore™ brings datacenter-class infrastructure to edge locations once thought impossible. Self-contained, no datacenter facility required, no cloud dependency.

Half rack,
no datacenter Standalone power and cooling

Your hardware,
your choice Open fabric, no vendor lock-in

Near-linear
scaling 99.7% efficiency across GPUs

What is GigaIO’s Manticore?

Manticore is a fully self-contained half-rack system that brings datacenter-class computing out of the data center to enable you to run full AI inference models and advanced machine learning anywhere you want. It delivers data center class compute power closer to where the data is generated, speeding “time to knowledge”. It delivers the compute power of a full GPU cluster in a half-rack you can run on-site, in spaces like a manufacturing floor or back office.

You get the security of your data staying on-site, with greatly improved compute times and dramatic savings over running the workload in the cloud, even where connectivity is limited, intermittent, or nonexistent.

What if the Edge could process data, not just collect it?

For most operations, the edge is where data is gathered, not where it’s understood. Sensors, cameras, and instruments generate far more than a remote site can process, which leaves two options: push it over a network you may not have, or physically truck it to a datacenter and wait. One depends on connectivity that’s limited, costly, or denied. The other costs you days. Either way, the answers arrive long after the moment you needed them.

Manticore changes what the edge is capable of: a single half-rack with the power to run advanced machine learning models and AI inference models on-site, turning the place you collect data into the place you act on it.

What makes Manticore right for the edge?

Data Sovereignty and Security

Your data can stay local. Data custody is provable, not promised.

Operational continuity

Keeps running where outside comms are degraded, intermittent, or denied.

Data Gravity

Bring compute to the data when it’s too large, costly, or sensitive to move.

Real-time Results

Act on insight the moment it’s generated, before it loses value in transit.

Thermal and Power Autonomy

Standalone cooling and power in a single half-rack footprint

Scalability

Start with what the site needs and scale compute as the workload grows.

Manticore FAQs

What is GigaIO Manticore used for?

Manticore runs datacenter-class infrastructure outside of the datacenter, enabling you to run large machine learning and AI inference workloads on-site at the edge, in a self-contained half-rack. It’s built to dramatically reduce the time to knowledge for most operational workloads, where the operations generate more data than they can move or can’t send to the cloud, across defense, energy, mining, media, and remote industrial sites where compute has to happen where the data is.
Who is Manticore a good fit for?

Manticore fits teams that need serious compute where traditional datacenter infrastructure can’t go. It is perfect for the “Near Edge” in remote industrial sites, forward defense operations, broadcast and field production, and any environment where connectivity is limited, costly, or denied, but the workload still demands near real time performance.
Does Manticore require an internet or cloud connection to run?

No. Manticore operates fully on-site and processes machine learning and AI inference workloads with no cloud dependency. It keeps running when connectivity is limited, degraded, or denied, so your data stays local and your results come back in real time, even in disconnected or air-gapped environments.
Can I run Manticore in an environment that also uses connections to the Cloud or to remote datacenters?

Absolutely. The difference with Manticore is that your operational continuity does not have to be dependent on those connections. It has the power to fully run the workloads locally when and if those connections fail or are intermittent — and then transfer data when the connections are restored.
Do I need a traditional datacenter or special facility to run Manticore?

No. Manticore is a single self-contained half-rack with standalone power and cooling, designed to run in ordinary spaces like a manufacturing floor or back office. There’s no datacenter to build and no special facility required to bring datacenter-class computing onsite.
Which GPUs and accelerators does Manticore support?

Manticore runs on FabreX, GigaIO’s patented PCIe-based open fabric technology, so you’re never locked to a single vendor. Use your existing hardware mix or choose new accelerators to fit each workload, rather than committing to one supplier’s roadmap. Use GPUs, FPGAs, ASICs, video capture cards from vendors like NVIDIA, AMD, Intel, d-Matrix, Furiosa, Qualcomm or others.
How is Manticore different from Gryf?

Gryf is portable and deploys to the field in a transportable form factor; Manticore is a fixed-site rack that brings datacenter-class AI and HPC to a fixed edge location. They share the same architecture and software but are two different implementations; solving edge compute problems for two different deployment profiles, mobile versus fixed.
Can I use storage in Manticore?

Yes. There are a couple ways to do this in Manticore. If you purchase servers from GigaIO, we supply servers with 8 2.5” NVME slots. You can fill them with 15 to 60TB drives for up to half a petabyte of storage. Alternately, you can add a PCIe carrier card into the Pooling Appliance that supports U.2 or M.2 storage cards.
Do I have to buy servers from GigaIO?

No. In most cases, you can use servers you already own or go buy them yourself. However, there can be circumstances where you want certain numbers or types of GPUs that many servers will not handle properly, where purchasing servers from GigaIO that are optimized for these purposes may make the most sense.

Book a Briefing
A working conversation about your site and workloads.

Technical Evaluation
Learn about workload, power, and footprint.

Get Up and Running Fast
Faster than you think. Ask our team.

"*" indicates required fields