Graid’s SupremeRAID know-how combines a GPU with a software-defined storage stack (SDS) to ship insanely excessive storage speeds that surpass what most servers can hit as we speak with standard RAID configurations.
By pairing Graid’s SupremeRAID card with the Nvidia RTX A2000 GPU, Retention evaluate managed to hit 5.6M IOPS in 1.1ms utilizing one of the best RAID efficiency at 4K chunk measurement, whereas Graid landed a results of 28.5M IOPS in the identical check.
The system they used featured a Supermicro AS-2125HS-TNR server with two AMD EPYC 9654 CPUs, 384GB DRAM and 24 3.84TB KIOXIA’s CM7-R Gen5 SSDs. They configured the drives for a RAID5 configuration with a 4K stripe measurement for Graid.
The perfect RAID configuration hit 205k IOPS at 15.01ms with 64K chunk in comparison with Graid hitting 2.02M IOPS at 1.52ms.
RAID vs SupremeRAID
RAID playing cards are restricted to twenty-eight GB/s speeds with Gen4 SSDs, with 4 Gen4 SSDs saturating one RAID card and extra required to maximise utilization of a 24-bay server.
SupremeRAID, however, can help 32 drives in a single system and isn’t restricted by most PCIe slot bandwidth. Utilizing the Graid system can due to this fact work to unravel the bottlenecks in a traditional server configuration.
It additionally licenses the GPU to deal with the drive administration and information safety – offloading this from the CPU and liberating it as much as run purposes.
Storage Evaluation used Nvidia A2000 GPUs in exams, that are comparatively cheap, to realize 28M IOPS together with speeds of as much as 260GB/s. That is regardless of the cardboard utilizing a Gen4 interface, however this doesn’t stop it from reaching efficiency ranges of Gen5 SSDs.
That quickest SSDs seen in isolation can hit speeds approaching 15GB/s, however Phison’s PS5026-E26 Max14um Gen5 SSDwhen e.g. sequential learn speeds of 14,175 MB/s and write of 12,741 MB/s
Though the publication examined this on Nvidia A2000 GPUs, Graid’s software program is in the meantime appropriate with nearly any Nvidia GPU. These are straightforward to put in and don’t require further battery energy, it added.