ELSA Japan has announced the launch of the GigaIO Gryf portable AI system for the domestic market.
The company highlights that the system supports 100GbE and FabreX. According to the technical specifications of GigaIO, the L40S and H100 NVL options offer different bandwidth capabilities. The H100 NVL configuration delivers up to 3341 TFLOPS FP8 and 30 TFLOPS FP64.
ELSA has not disclosed pricing details, explaining that costs vary based on configuration. Customers are advised to contact the company directly for pricing information.
Previously, we reported that Durabook introduced the Z14I-HG, the world’s first tactical laptop, while Samsung confirmed its plans to release a 256 TB SSD.
The system’s chassis features six internal slots for modules. ELSA informs that clients can mix and match computational, acceleration, storage, and networking modules according to their needs. The compute power is driven by AMD EPYC 7713P and EPYC 7313P processors. For GPU options, NVIDIA L40S with 48 GB or H100 NVL with 94 GB can be used. The storage capacity for data reaches up to 246 TB NVMe E1.L.
+40% to PC Price: Expensive DRAM and NAND coupled with a processor shortage
By browsing this site, you agree to all the site policies on itc.ua.
By browsing this site, you agree to all the site policies on itc.ua.
This computer, designed in a suitcase-like compact form factor, is travel-friendly. Jointly developed by GigaIO, SourceCode, and ELSA, the system is described as a configurable AI platform suitable for on-site deployment without requiring server rack installation.
ELSA notes that the system’s weight does not exceed 25 kg depending on the configuration. The GigaIO Gryf employs the GigaIO FabreX PCIe/CXL architecture, which allows for software-based redistribution of CPU, GPU, and storage resources without the need for physical cabling.
ELSA adds that the SWARM configuration via the FabreX protocol enables the integration of up to 5 Gryf systems. In this setup, the manufacturer claims performance exceeding 3600 TFLOPS in FP16 format, with storage capabilities reaching petabyte levels through NVMe. This positions the system closer to the realm of mobile devices than standalone portable workstations, although the basic unit is sold as a standalone platform.
What do you think about this article? Votes: It’s great, Let it be! Damn! Well, that’s something… Annoying!