Gryf Resource Sled: Compute
Enterprise-grade Compute.
Suitcase-sized.
The same processor architecture that runs in hyperscale datacenters, now built for the edge.
Compute Sled
Processing Power That Keeps Pace with the Workload
The Gryf Compute Sled delivers enterprise-grade server performance in a sled you can hot-swap in minutes. Built on the same processor architecture found in hyperscale data centers, it gives applications the headroom to ingest, analyze, and act on massive datasets without bottlenecks — right at the edge.
Up to 64 AMD Epyc Cores
Up to 1,024GB DDR4 Memory
One Integrated Environment
What makes the Compute Sled exceptional isn’t just its processor, it’s how GigaIO’s patented PCIe-based open fabric technology, FabreX, integrates it into the broader Gryf system. FabreX uses PCIe as the backbone throughout Gryf in order to enable a compute sled — or multiple compute sleds — to scale up resources on the server as well as share them between servers.
Gryf’s Compute Sled runs up to 64 AMD EPYC cores and 1,024GB of DDR4 memory, giving your applications the horsepower to ingest, analyze, and act on massive datasets without waiting on the network. And multiple Compute Sleds across a Gryf stack can share accelerator and storage resources — disaggregated, composable, and dynamically allocated to whichever workload needs them most.
If your workload needs more than one kind of processing power, no problem. Pair the Compute Sled with a GPU / Accelerator Sled to feed inference workloads, or add a Storage Sled to keep massive datasets close at hand.
Specifications
AMD EPYC 7313 16-core, or AMD EPYC 7713P 64-core
2x QSFP56 100GbE
Up to 1,024GB DDR4 3200MHz
Linux, Windows, KVM, VMware
512GB NVMe M.2 SSD
Up to 6 Configurable Slots
Simple to Service
Should a component need service, simply swap the bad sled with a good one. No waiting for special technicians to be flown in. No opening boxes, no cables and connectors to be confused by. Swapping a sled is just like swapping a battery.
Every sled includes remote management through IPMI/BMC, so your team can restart, diagnose, or reconfigure it through the chassis RJ45 port, no matter what state the primary OS is in. No hands-on-hardware required, no waiting on someone to be standing next to it.
Complete Your Configuration
Each Gryf can hold six hot-swappable sleds. Add more than one Compute Sled or choose from other options:
GPU / Accelerator Sled
Dedicated acceleration for AI inference and training
View sled →
Network Sled
High-bandwidth connectivity for the edge
View sled →
Storage Sled
Massive, instantly accessible capacity
View sled →
"*" indicates required fields