Data processing units have become an important part of data center infrastructures in recent years as the need to store and move data has increased sharply. So it should come as no surprise that two of the major players in the DPU space are going toe-to-toe with new DPU performance claims.
Nvidia published a blog post this week claiming a new performance record for its Bluefield-2 DPU, stating that a single BlueField-2 data processing unit reached “41.5 million input/output operations per second (IOPS) — more than 4x more IOPS than any other DPU.”
It appears that “any other DPU” is a veiled reference to competitor Fungible, which last month issued a press release saying that it worked with the University of San Diego on a test that “shattered the NVMe over TCP storage initiator performance world record, achieving 10M IOPS.”
Nvidia’s blog further stated, “The BlueField-2 DPU delivered record-breaking performance using standard networking protocols and open-source software. It reached more than 5 million 4KB IOPS and from 7 million to over 20 million 512B IOPS for NVMe over Fabrics (NVMe-oF), a common method of accessing storage media, with TCP networking, one of the primary internet protocols.”
Nvidia also explained its test methodology, saying the latest performance “was achieved by connecting two fast Hewlett Packard Enterprise Proliant DL380 Gen 10 Plus servers, one as the application server (storage initiator) and one as the storage system (storage target). Each server had two Intel “Ice Lake” Xeon Platinum 8380 CPUs clocked at 2.3GHz, giving 160 hyperthreaded cores per server, along with 512GB of DRAM, 120MB of L3 cache (60MB per socket) and a PCIe Gen4 bus.”The blog post added, “To accelerate networking and NVMe-oF, each server was configured with two NVIDIA BlueField-2 P-series DPU cards, each with two 100Gb Ethernet network ports, resulting in four network ports and 400Gb/s wire bandwidth between initiator and target, connected back-to-back using NVIDIA LinkX 100GbE Direct-Attach Copper (DAC) passive cables. Both servers had Red Hat Enterprise Linux (RHEL) version 8.3.” For further details–and there are many more–read the blog post on Nvidia’s website.
Despite the level of detail Nvidia provided, Blocks and Files wrote that some key details and context of the test were not revealed, making for an "incomplete" claim.
All of this is important because DPUs have been positioned by Nvidia and others as necessary to handle growing storage and networking needs in data centers as CPUs increasingly need to offload some of the demands being placed upon them.
RELATED: Nvidia looks to democratize AI with big batch of announcements