A server with 128 GB of RAM: what it is used for and how the memory is laid out
128 GB is the most popular size for servers running databases, caches and virtual machines. As of 3 November 2026 the selection holds 1,323 plans from 79 hosts with 128 GB of memory or more. The lowest price is about $94 a month, the median about $440. Roughly two thirds of the plans have exactly 128 GB; the rest have 192, 256, 384 GB or more. The cheap end is dominated by dual-processor servers on Intel Xeon E5-2620 v3, E5-2670 v3, E5-2680 v4 and Xeon Silver 4110 and 4116, the middle by AMD EPYC 7502, Ryzen 9 5950X and 7950X. Most plans are in the US, Germany, Canada, Russia and France.
Workloads that need this much memory
- Databases. The more of the working set fits in memory, the less often the database goes to disk. The PostgreSQL documentation suggests starting with 25 % of memory for
shared_buffers; the database uses the rest through the operating system cache. For MySQL on a dedicated server the InnoDB buffer pool is usually set to 60–80 % of memory. With 128 GB that lets you keep an 80–100 GB database entirely in memory. The selection for this task is servers for databases. - Caches and queues. Redis and Memcached keep everything in memory, so the server's RAM directly caps the cache size.
- Search and analytics. Elasticsearch, OpenSearch and ClickHouse lean heavily on memory for indexes and sorting.
- Virtual machines. 128 GB means about ten machines with 8 GB each, with room left for the hypervisor. Ready-made configurations are in the servers for virtualization selection.
- Java applications and builds. Large JVM services and parallel compilation of big projects easily take tens of gigabytes.
The same 128 GB can run at different speeds
Capacity is not the only property of memory. Speed depends on the number of channels and how many modules sit in them.
Server processors have many channels: six on the first two generations of Xeon Scalable, eight on AMD EPYC 7002 and 7003, twelve on EPYC 9004. If 128 GB on an EPYC 7002 is made of eight 16 GB modules, every channel is populated and bandwidth is at its maximum. If it is two 64 GB modules, only two channels out of eight are in use, and for a database or analytics that is noticeably slower.
Desktop platforms (Ryzen, Core) have only two channels. 128 GB on them is usually four 32 GB modules, and with two modules per channel DDR5 often runs at a lower frequency. For a web project that does not matter; for a database it does.
Old dual-processor Xeon E5 servers give you 128 GB for little money, but it is DDR3 or DDR4 at 1600–2400 MHz, split between two processors. Such a server is fine for caching and backups, less so for a busy database.
What memory costs and how to save
Prices in the selection differ several times over, and nearly all of the difference comes from the processor and disks, not the memory itself. A quarter of the plans cost less than about $280 a month: mostly dual-processor Xeon E5 and Silver with SATA SSD or HDD. The middle is previous-generation EPYC and Ryzen 9 with NVMe. Above about $795 you find new EPYC 9004 and Xeon Gold with dozens of cores.
If you need memory for caching or analytics and the CPU load is light, an old dual-processor server gives the same capacity at a fraction of the price. If the database writes a lot, paying more for a new platform with NVMe and fast memory pays off in response time. Do not compare price per gigabyte of memory; compare the price of a server that will handle your load with a sensible margin.
What to check with the host
- Module layout. How many modules, of what size, and at what frequency the memory runs. The host can tell you even if the plan does not say.
- ECC and memory type. Xeon and EPYC servers use registered ECC memory. On Ryzen, ECC depends on the motherboard, and Core i7 and i9 before the 12th generation do not support it at all.
- Whether memory can be added later. If the database is growing, it is easier to agree on an upgrade to 256 GB without moving. Ask whether there are free slots and what the upgrade costs.
- Disks for the database. Memory speeds up reads, but writes still go to disk. For a database choose NVMe: it is fitted in about 40 % of the plans in the selection.
If 128 GB is already too little, see servers with 256 GB or more. If less will do, compare VPS with 64 GB of memory: they are cheaper, but the physical server's resources are shared with other customers.