Windows tweaks for increasing RAM FS performance
6 posts
Are there any recommended additional Windows tweaks for improving SoftPerfect RAM File System performance on Microsoft Windows 11 even further?
When I used SoftPerfect RAM Disk, I disabled memory compression and the paging executive, and enabled Large System Cache. However, I believe Large System Cache is no longer relevant for the RAM File System. I've also read that the IOPageLockLimit registry key is deprecated in Windows 11 and has no effect.
Thanks in advance
Your understanding in correct, and the short answer is that there are no Windows registry keys worth touching for the RAM File System.
The reason is in how it works. RAM FS keeps all file data and metadata in nonpaged kernel memory, which Windows never pages out and never runs through memory compression, so disabling the paging executive or memory compression changes nothing for it. It also bypasses the Windows file cache entirely and serves all reads and writes directly from its own memory, so LargeSystemCache is indeed irrelevant, and you are right that IOPageLockLimit has been deprecated in Windows years ago.
In practice, performance gains come from what you put on the volume, not from tuning the system around it. Point anything temporary and write-heavy at it: Windows TEMP and TMP folders (there is a built-in tool for this under Tools - Set Windows Temporary Folder), browser caches, downloads, build output and other scratch data.
Would you recommend disabling memory compression and the paging executive, and enabling Large System Cache when running RAM Disk on Windows 11 64-bit?
The same applies to RAM Disk, at it also keeps its data in locked physical memory, so none of those three keys would change anything for it, and the advice above about what you put on the volume is the same.
One thing that is more effective than any registry key: if you are on a multi-socket machine (with several CPU sockets), set the NUMA node under Advanced in the Add/Edit Disk dialog, so the disk memory stays on the same node as the cores using it.
Thanks. I only have a single (physical) CPU and a 0 option under the NUMA node setting, so it won't make any difference.
The NUMA node section in the settings could do with a brief explanation, as I suspect I'm not alone in not knowing what it meant, or was for. Perhaps, if you have a multi-CPU machine you would know, and thus those who need it would understand what that setting was for.
NUMA (Non-Uniform Memory Access) is an architecture used in multi-socket systems. Instead of all CPUs sharing one equally accessible pool of RAM, the machine is divided into NUMA nodes, typically one node per CPU socket, with each socket having its own directly attached RAM. In these terms, memory access can be:
- Local - CPU accesses RAM attached to its own NUMA node, which means lower latency and higher bandwidth.
- Remote - CPU accesses RAM attached to a node of another socket, which means higher latency and additional inter-socket traffic.
For NUMA, the important distinction is physical CPU sockets, not simply the number of cores, as one physical CPU can contain multiple CPU cores.
Selecting a particular NUMA node for the RAM disk tells the OS to preferentially allocate memory for that RAM disk from the specified node. Making the CPU cores that are most actively using your RAM disk and the RAM disk itself run on the same node can give you a more optimal sequence
Applications > CPU > local RAM > RAM disk
instead of
Applications > CPU > inter-socket link > remote RAM > RAM disk
This can improve performance because RAM disk operations are ultimately memory accesses. Keeping the data physically close to the CPU cores that are using it reduces memory latency and avoids consuming inter-socket bandwidth. The gain is most noticeable on workloads that perform high numbers of small or latency-sensitive accesses. Whereas, as you correctly noted, on a single-socket system the NUMA setting makes no difference because there is only one option.