A plethora of boutique vendors have over the past few months released powerful GPU servers based on AMD’s EPYC 9000 series - these are essentially beefed-up workstations. Velocity Micro is the latest one to join the fray with its ProMagix G480a GPU which can take up to eight dual-slot Nvidia GPUs to accelerate CUDA or Tensor-enabled workflows (no AMD Instinct MI300 yet).
No gaming here (although you should be able to run Cyberpunk 2077 - and Crysis - without much issue should you want to), just sheer number crunching at a scale that only a couple of decades ago, would require a supercomputer.
Housed in a massive 4U chassis, the dual-socket motherboard can take up to two EPYC 9654 CPU totaling 192 cores and paired with 1TB DDR5, much lower than what we’d expect. Purchasers can either rely on an integrated VGA card from Aspeed (remember these) or kit the server with up to eight 48GB PNY Nvidia RTX 6000 Ada generation workstation graphics cards. Powering all this are three (yes, three) 3kW power supply units with an extra redundant one.
Who needs one of these monsters?
Velocity Micro says that these are perfect for “AI or Machine Learning, complex simulations, high demand scientific calculations”. “By putting emphasis on scalability, functionality, and performance, we’ve created a line of server solutions that tie in the legacy of our high-end brand while also offering businesses alternative options for more specialized solutions for the highest demand workflows,” said Randy Copeland, President and CEO of Velocity Micro.
A fully kitted server with 384 threads, 1TB of memory and eight RTX 6000 Ada reached almost $135,000. Velocity Micro offers up to three-year warranty on these with a lifetime upgrade plan available for maintenance, upgrades and tuning. There are cheaper options from the likes of Dihuni (less than $97,000), Xicomputer (lower than $90,000) or Bizon-Tech (less than $92,000); so it does pay to shop around, look at the small prints and after sales support.
from TechRadar - All the latest technology news https://ift.tt/WsXNTjp
No comments:
Post a Comment