Doubling Server Capacity for Gaming

For gaming infrastructure teams, one of the fastest ways to hit a scaling wall is memory. As workloads grow and server fleets expand, adding more capacity usually means buying more DRAM—which is often one of the most expensive parts of the server.

But what if the path to 2X more memory capacity did NOT require 2X the DRAM spend?

That is the opportunity MEXT created for UltraServers, a gaming hosting company. They were running a server configured with 256GB of DRAM. They then reduced the physical DRAM by 50% in this server, purchased a new server with little to no DRAM, added the removed DRAM into the new server, and added MEXT. The result? 2X the original memory capacity: 512GB of DRAM total with 2 servers (each with 128GB of physical DRAM and 128GB of flash+MEXT appearing as memory). 

UltraServers plans to scale this effort across their entire fleet. They are running 5 servers with 8 nodes each, with each node configured with 256GB of DRAM (1TB per server). They will purchase 5 net-new servers with little to no DRAM, remove 50% of the physical DRAM from their existing servers, and re-populate this DRAM into the new servers. 

The result would be: 10 servers total, no changes to the existing DRAM footprint, but 2X the effective memory capacity with MEXT (10TB)! And the only purchases that would be required would be the net-new empty-DRAM servers (very cost-effective) plus MEXT software licensing ($126K for 3-years, including enterprise support). This is a relatively small investment compared with the alternative of purchasing 5 additional TB of new DRAM!

At rack scale, the same logic becomes even more compelling. A standard deployment with 240 nodes at 512GB per node delivers 120TB of DRAM across 12 racks. If each node’s DRAM is cut in half and those recovered DIMMs are used to populate additional systems, the organization can theoretically double node count to 480 and double total deployed memory capacity to 240TB—without needing to buy all-new memory for every added server.

The bottom line: MEXT changes the economics of infrastructure expansion. Instead of treating DRAM as a hard scaling limit, organizations can redistribute existing memory, reduce per-node DRAM requirements, and use MEXT to maintain performance while dramatically increasing the number of usable servers in the fleet.