Specifications
Form Factor
Suitcase-sized mobile data center in a ruggedized carbon-fibre case
Dimensions (with wheels)
9.00 x 14.00 x 24.50 in (228.6 x 355.6 x 622.3 mm)
Dimensions (without wheels)
9.00 x 14.00 x 22.00 in (228.6 x 355.6 x 558.8 mm)
Weight
Under 55 lbs (24.95 kg) maximum
Sled Slots
6 universal slots for compute, accelerator, storage and network sleds in any position
Compute Sled CPU
1 x AMD EPYC 7003 family (16-core 7313 tested at 155 W TDP; up to 64-core planned)
Compute Sled Memory
4 x 64 GB DDR4-3200 (512 GB total); 1 TB via 4 x 256 GB planned
Compute Sled OS Storage
1 x 512 GB NVMe M.2 SSD
Accelerator Sled
1 x NVIDIA L40S 48 GB; single or double-wide PCIe FHFL up to 350 W
Storage Sled
8 x 30 TB NVMe E1.L SSD — 246 TB per sled
AI/M-L Config Sled Mix
2 compute + 1 accelerator + 2 storage + 1 network sled
Internal Fabric
128 Gb/s PCIe board-to-board AI memory fabric
Fabric Expansion
8 x FabreX Mini-SAS-HD-32G (256 Gb/s total) for Gryf daisy-chaining or home-base offload
Network Sled Ports
2 x QSFP56 100GbE plus 6 x SFP28 25GbE, copper or optical
Compute Sled Networking
2 x QSFP56 / QSFP28 / QSFP+ 100GbE, copper or optical
Management Network
5 x SFP+ 10GbE for FabreX Fabric Manager and out-of-band BMC/IPMI
Single-Unit Performance
91.6 TFLOPS FP32, 362 TFLOPS FP16, 733 TFLOPS FP8
SWARM Performance (5 units)
458 TFLOPS FP32, 1,815 TFLOPS FP16, 3,665 TFLOPS FP8
Scaled CPU and Memory
32 cores / 512 GB in 1 unit, scaling to 160 cores / 2.56 TB across 5 units
Scaled Storage
492 TB in 1 unit, scaling to 2.5 PB across 5 units
Power
Dual AC/DC 2,500 W 1+1 hot-swap supplies; IEC-320-C20 inlet, 100-240 VAC 50/60 Hz
Cooling
6 x 60 mm fans with workload-optimised control and removable 45 PPI filters
Operating Temperature
10 °C to 32 °C (50 °F to 90 °F)
OS Support
Linux Rocky 8/9 or Ubuntu 20/24; Windows Server 2019/2022 planned
Compliance
FCC Class A, CE, TAA / Made in the USA, IP55 in transport
Field Replaceable Units
Sleds, power supplies, fan tray with filtration, case top and bottom covers
Overview
GigaIO Gryf is the only AI platform that arrives as carry-on luggage. Co-designed by GigaIO and SourceCode, it packages a reconfigurable mobile data center into a ruggedized carbon-fibre case measuring 9 x 14 x 24.5 inches and weighing under 55 lbs — small enough for a TSA-friendly carry-on, with a detachable top carrying a folding handle and a detachable bottom with wheels so the whole system can be rolled to where the data actually is.
The design is sled-based rather than fixed. Six universal slots accept compute, accelerator, storage and network sleds in any position, so the same chassis can be loaded as an AI inference node, a general HPC compute node, or a petabyte-scale storage node. In the AI/ML launch configuration it carries two compute sleds, one accelerator sled, two storage sleds and one network sled.
The AI/ML configuration delivers 91.6 TFLOPS FP32, 362 TFLOPS FP16 and 733 TFLOPS FP8 from an NVIDIA L40S 48 GB accelerator paired with an AMD EPYC 7003 compute sled holding 512 GB of DDR4-3200. Storage sleds each take eight 30 TB NVMe E1.L drives, for 246 TB per sled and 492 TB in a single unit. Five Gryfs chained as a SWARM scale that to 458 TFLOPS FP32, 3,665 TFLOPS FP8, 160 cores, 2.56 TB of memory and 2.5 PB of storage.
What makes the scaling possible is FabreX, GigaIO's AI memory fabric. Internally the system uses a 128 Gb/s PCIe board-to-board fabric, and eight FabreX Mini-SAS-HD-32G links (256 Gb/s aggregate) daisy-chain multiple Gryf units or offload data at a home base. A five-port SFP+ 10GbE management network carries the FabreX Fabric Manager plus out-of-band BMC and IPMI.
Deployment realities are covered explicitly: dual 2,500 W 1+1 hot-swap power supplies on an IEC-320-C20 inlet, six workload-controlled 60 mm fans with removable 45 PPI filtration, IP55 protection in transport, and sleds, supplies, fan tray and case covers all field-replaceable. The platform is FCC Class A, CE and TAA compliant and Made in the USA.
Key Benefits
Datacenter-class AI inference in carry-on luggage — 91.6 TFLOPS FP32 at under 55 lbs. Six universal sled slots let one chassis serve AI, HPC or storage missions. Scales to 3,665 TFLOPS FP8 by chaining five units over FabreX. 246 TB of NVMe per storage sled for petabyte-scale field storage. Field-replaceable everything with IP55 transport protection and TAA/US-made compliance.
Applications
Tactical and defence edge deployments, disaster response and field command posts, ISR and sensor-data analytics where communications are unreliable, remote scientific research stations, mobile media and sports production, and on-premise AI inference at sites where the cloud cannot reach.
Request a Quote — GIGAIO GRYF — SUITCASE-SIZED PORTABLE AI SUPERCOMPUTER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →