The hardware behind the future


Dell Calls Lightning the Fastest, and Marks Its Own Homework

Dell has expanded its AI Data Platform with Lightning File System, Exascale Storage and NVIDIA-accelerated data engines. Almost every performance claim is footnoted to Dell’s own internal or preliminary testing.

Dell Calls Lightning the Fastest, and Marks Its Own Homework

Dell Technologies has expanded its AI Data Platform with a parallel file system, a three-engine storage platform and a set of NVIDIA-accelerated data engines, and has supplied the benchmarks for nearly all of it itself.

The loudest claim attaches to Lightning File System. The release calls it the world's fastest parallel file system and hangs a footnote on the superlative: Dell preliminary testing comparing random and sequential throughput per rack unit, dated May 2025. Per rack unit is a density measure rather than an absolute one, preliminary is the release's own adjective, and ten months is a long time in this market.

Travis Vigil, senior vice president of ISG Product Management at Dell, said, "The number one problem enterprises face when moving AI pilots to production is curating the data they already have and putting it to work."

Curating is doing quiet work in that sentence. The engine sold to do it, the Data Orchestration Engine, is powered by technology from the Dataloop acquisition, so the pitch is labeling and enrichment sold as governance, aimed at buyers who have already found that their data lake will not feed a GPU.

The performance numbers move between metrics as they go. Lightning is quoted at up to 150 GB/second per rack, then at up to 20X greater performance than traditional flash-only scale out file competitors, then at up to 2X greater throughput per rack unit than competing parallel file systems. The 20X note rests on internal testing of IOPs per node using FIO over a remote file system, so a throughput headline is supported in part by an IOPS comparison on a different unit of account, against competitors the release does not name.

Exascale Storage is called the only 3-in-1 storage built for extreme-scale AI and HPC, running PowerScale, ObjectScale and Lightning File System on common PowerEdge hardware, with support for NVIDIA CX-8 and CX-9 SuperNICs, planned network connectivity up to 800GbE and read performance up to 6TB/second per rack. The note behind only compares publicly available vendor documentation and excludes single-engine multi-protocol designs, which is the clause doing the work. Exclude the architecture most rivals actually ship and the field of one arrives on schedule.

PowerScale gets a smaller and more honest number. Its pNFS implementation is quoted at up to 6X faster with large files than NFSv3, footnoted to preliminary internal testing of single-client random I/O, which is at least a comparison against a named protocol.

One structural detail matters to anyone checking the methodology. The numbered notes at the foot of the release run 1, 2, 3, 3, 4, giving fourteen entries against thirteen superscripts in the body, so from the second note 3 onward the numbering no longer lines up. Several claims therefore cannot be matched to their stated methodology with confidence, including the 6TB/second Exascale figure, whose adjacent note describes testing of Lightning File System.

Neither product is shipping today. Lightning File System will be available in April 2026, Dell says, and Exascale Storage is targeted for early in the second half of calendar 2026.

Leave a Reply