{"id":1521,"date":"2026-10-29T22:41:23","date_gmt":"2026-10-29T22:41:23","guid":{"rendered":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/?p=1521"},"modified":"2026-10-29T22:41:23","modified_gmt":"2026-10-29T22:41:23","slug":"dedicated-server-numa-node-binding-game-workloads","status":"publish","type":"post","link":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/","title":{"rendered":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers"},"content":{"rendered":"\n

On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket dedicated server, that assumption breaks. Each socket owns a slice of the installed RAM, and a core reading memory attached to the other socket pays a measurable penalty \u2014 typically tens of percent higher latency and lower achievable bandwidth.<\/p>\n\n\n\n

For most workloads this is invisible. For a game server tick loop touching thousands of entities every few milliseconds, it shows up as tail latency: ticks that should take a fraction of the budget occasionally take noticeably longer, players feel it as intermittent stutter, and the cause is nowhere in the game’s own logs. This guide covers when NUMA actually costs you ticks, how to pin correctly, and how to prove the pinning worked. If you are sizing a dual-socket node, our dual-socket dedicated server configurations<\/a> list the core and memory layout per tier.<\/p>\n\n\n\n

Why cross-socket access is expensive<\/h2>\n\n\n\n

Each socket has its own memory controller and its own set of DIMM slots. When a core on socket zero reads an address that lives in memory attached to socket one, the request must traverse the interconnect between the sockets. That path adds latency and consumes interconnect bandwidth, and if many cores do it simultaneously the interconnect itself becomes a shared bottleneck.<\/p>\n\n\n\n

Operating systems mitigate this by trying to allocate memory on the same node as the core that first touched it \u2014 first-touch policy. That works well for a process that starts, allocates, and stays put. It works badly for a process that starts on one core, spawns worker threads, and lets the scheduler migrate them across sockets, because the memory placement was decided by whichever thread happened to fault first.<\/p>\n\n\n\n

Signs that NUMA is your problem<\/h2>\n\n\n\n

The signature is a gap between average and tail performance that does not correlate with player count. Ticks are usually fine and occasionally terrible, and the bad ticks have no obvious input that explains them. If you graph tick time as a distribution rather than an average, you see a long tail rather than a shifted body.<\/p>\n\n\n\n

A second sign is asymmetry across otherwise identical nodes. Two servers with the same map, same mods, and same player count performing differently usually means their worker threads landed in different places relative to their memory. That is a placement problem, not a capacity problem, and buying more RAM does not fix it.<\/p>\n\n\n\n

Confirming the topology before you pin anything<\/h2>\n\n\n\n

Start by reading the node layout: how many nodes, which CPUs belong to each, and what memory each node owns. Confirm that the machine reports the topology you expected \u2014 a dual-socket board that reports a single node usually means NUMA is disabled in firmware or that the interconnect is presenting as one domain, which changes the whole analysis.<\/p>\n\n

Record the distance table too. The relative cost of remote versus local access is machine-specific and is the number you will want when deciding whether pinning is worth the operational complexity. If remote access costs are small, pinning buys you little and adds fragility.<\/p>\n\n\n\n

Pinning threads and memory together<\/h2>\n\n\n\n

The correct pin has two halves: bind the process’s CPUs to a single node’s cores, and bind its memory allocation policy to the same node. Doing only the first is a common mistake and often makes things worse, because the threads now run on one socket while their memory remains wherever it was first allocated.<\/p>\n\n\n\n

Use a strict local allocation policy so that any allocation the process makes lands on the bound node. Then make sure thread counts match the core budget you reserved \u2014 if the server spawns more worker threads than the node has cores, the scheduler will spill the remainder onto the other socket and the pinning becomes decorative.<\/p>\n\n\n\n

Running one node per game server on a dual-socket box<\/h2>\n\n\n\n

The cleanest deployment on a dual-socket machine is to divide it: one game server instance bound to socket zero’s cores and memory, another bound to socket one’s. Each instance then behaves like a single-socket server with a hard partition, and the interconnect is used only for I\/O and operating system work rather than for game data.<\/p>\n\n\n\n

Two cautions. First, leave a couple of cores per socket unbound for kernel and interrupt work; a fully saturated socket handles network interrupts poorly. Second, watch disk and network paths \u2014 if both instances share a controller or a NIC queue set, you have created a new shared bottleneck while fixing the memory one. Our dedicated server tier comparison<\/a> shows which configurations include separate controller paths per socket.<\/p>\n\n\n\n

Verifying the pin held<\/h2>\n\n\n\n

Do not assume the pin survived. Check which cores each game process is actually running on, over time, and confirm the set stays within the intended node. Then check the memory placement statistics: the count of allocations served from the local node versus remote. A successful pin shows almost everything local after the process has been running steadily.<\/p>\n\n\n\n

Finally, measure the effect rather than trusting the theory. Capture the tick-time distribution before and after pinning, with the same map and player count. If the tail does not improve, the bottleneck was elsewhere and you should unpin rather than carry the operational cost.<\/p>\n\n\n\n

What pinning does not fix<\/h2>\n\n\n\n

Pinning improves memory access locality. It does not give you more single-thread performance, more cache, or a faster disk. A server that is CPU-bound on one thread will remain CPU-bound after pinning \u2014 the win is in removing a source of variance, not in adding capacity.<\/p>\n\n\n\n

It also does not survive careless change well. Any reconfiguration that restarts the service under a different unit file, or that changes the thread count, can silently break the binding. Treat the pin as part of the deployment manifest and assert it in monitoring, or it will quietly revert during the next unrelated change.<\/p>\n\n\n\n

When to skip it entirely<\/h2>\n\n

If your node is single-socket, or if the measured remote-access penalty on your hardware is small relative to your tick budget, pinning adds complexity for little return. Measure first. The decision should be driven by a distribution that got narrower after the change, not by the fact that the machine has two sockets.<\/p>\n\n\n\n

The general rule: pin when tail latency is the problem and the topology shows a real gap. Do not pin because it sounds advanced, because a wrongly pinned server is harder to diagnose than an unpinned one that was simply left alone.<\/p>\n\n","protected":false},"excerpt":{"rendered":"

On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket dedicated server, that assumption breaks. Each socket…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"iawp_total_views":1,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-1521","post","type-post","status-publish","format-standard","hentry","category-server-guides-tutorials"],"yoast_head":"\nNUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers - Best Dedicated Web Hosting Server Blog<\/title>\n<meta name=\"description\" content=\"On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers\" \/>\n<meta property=\"og:description\" content=\"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers\" \/>\n<meta property=\"og:url\" content=\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/\" \/>\n<meta property=\"og:site_name\" content=\"Best Dedicated Web Hosting Server Blog\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-29T22:41:23+00:00\" \/>\n<meta name=\"author\" content=\"bestdedicatedwebhostingserver\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@bestdedicatedwebhostingserver\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"bestdedicatedwebhostingserver\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"5 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/\",\"url\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/\",\"name\":\"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers - Best Dedicated Web Hosting Server Blog\",\"isPartOf\":{\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#website\"},\"datePublished\":\"2026-10-29T22:41:23+00:00\",\"author\":{\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/6eb5f9ea35033fe8e67df44397e089b1\"},\"description\":\"On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket\",\"breadcrumb\":{\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#website\",\"url\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/\",\"name\":\"Best Dedicated Web Hosting Server Blog\",\"description\":\"\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/6eb5f9ea35033fe8e67df44397e089b1\",\"name\":\"bestdedicatedwebhostingserver\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/462a07aa37399bdb149e5f91dc3cd8906656bc4c7ed391a3b6f0199c5d2ab964?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/462a07aa37399bdb149e5f91dc3cd8906656bc4c7ed391a3b6f0199c5d2ab964?s=96&d=mm&r=g\",\"caption\":\"bestdedicatedwebhostingserver\"},\"sameAs\":[\"https:\/\/bestdedicatedwebhostingserver.com\/blog\",\"https:\/\/x.com\/bestdedicatedwebhostingserver\"],\"url\":\"https:\/\/bestdedicatedwebhostingserver.com\/blog\/author\/bestdedicatedwebhostingserver\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers - Best Dedicated Web Hosting Server Blog","description":"On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/","og_locale":"en_US","og_type":"article","og_title":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers","og_description":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers","og_url":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/","og_site_name":"Best Dedicated Web Hosting Server Blog","article_published_time":"2026-10-29T22:41:23+00:00","author":"bestdedicatedwebhostingserver","twitter_card":"summary_large_image","twitter_creator":"@bestdedicatedwebhostingserver","twitter_misc":{"Written by":"bestdedicatedwebhostingserver","Est. reading time":"5 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/","url":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/","name":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers - Best Dedicated Web Hosting Server Blog","isPartOf":{"@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#website"},"datePublished":"2026-10-29T22:41:23+00:00","author":{"@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/6eb5f9ea35033fe8e67df44397e089b1"},"description":"On a single-socket machine, memory locality is a non-issue: every core reaches every memory channel at roughly the same latency. On a dual-socket","breadcrumb":{"@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/dedicated-server-numa-node-binding-game-workloads\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/"},{"@type":"ListItem","position":2,"name":"NUMA Node Binding for Game Workloads on Dual-Socket Dedicated Servers"}]},{"@type":"WebSite","@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#website","url":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/","name":"Best Dedicated Web Hosting Server Blog","description":"","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Person","@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/6eb5f9ea35033fe8e67df44397e089b1","name":"bestdedicatedwebhostingserver","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/462a07aa37399bdb149e5f91dc3cd8906656bc4c7ed391a3b6f0199c5d2ab964?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/462a07aa37399bdb149e5f91dc3cd8906656bc4c7ed391a3b6f0199c5d2ab964?s=96&d=mm&r=g","caption":"bestdedicatedwebhostingserver"},"sameAs":["https:\/\/bestdedicatedwebhostingserver.com\/blog","https:\/\/x.com\/bestdedicatedwebhostingserver"],"url":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/author\/bestdedicatedwebhostingserver\/"}]}},"_links":{"self":[{"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/posts\/1521","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/comments?post=1521"}],"version-history":[{"count":1,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/posts\/1521\/revisions"}],"predecessor-version":[{"id":1522,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/posts\/1521\/revisions\/1522"}],"wp:attachment":[{"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/media?parent=1521"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/categories?post=1521"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bestdedicatedwebhostingserver.com\/blog\/wp-json\/wp\/v2\/tags?post=1521"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}