THE WOW FOREVER COLLECTION · FIELD GUIDE · RESEARCH CHECKED SEPTEMBER 22, 2026
Short answer: yes, the WoW Forever beta client leaks memory, and the fix is to log out to the character select screen and back in. Frame rate starts high, slides over 30 to 90 minutes, and settles at a floor around 30 FPS. Video memory climbs the whole time, at roughly 100 MB every five to ten minutes on one measured card, until it fills and spills into system RAM. Relogging frees it without closing the game. Blizzard's official known issues list does not mention it, no developer has posted about it in any of the forum threads, and the only acknowledgement is a customer support reply on X saying the team is "aware and working on fixes and improvements already." The current beta build is 1.60.1.69913 and the leak is in it.
What it looks like
The pattern is the same in every report. You log in and the game runs the way your hardware should run it. Somewhere between 15 minutes and an hour later it is noticeably worse, and by the second hour it is bad. The slide is gradual enough that most people blame the zone they are in, then notice it does not recover when they leave.
The first thread went up on the US beta forum on the evening of September 17, a few hours after servers opened. A player with a Ryzen 7 9800X3D and an RTX 5080 was getting 25 to 30 FPS at 1440p. That first night was also the night the servers were on fire, so some of those numbers were server-side, but the replies underneath it already had the signature. One player on a laptop 4070 Ti with 8 GB of VRAM started at 55 FPS or better, slid into the 30s, relogged, and was back near 60. Another on a 2080 Super went from 32 FPS to 170 after relogging. His hardware had not changed. The client had.
By September 18 there was a dedicated thread titled "FPS Slowly Drops" whose opening line is the whole bug in one sentence: "FPS in the game slowly drops as you play - only way to fix it is to log out to character screen and log back in." Replies in that thread put numbers on it. One player went from 120 to 130 FPS down to 50, relogged, and was back above 120. Another on a 5070 Ti went from over 150 FPS to 30, where it stayed until relog. Two separate people noticed the same thing about that number: it does not keep falling. It "seems to hit a floor at 30 FPS."
The most useful detail in that thread came on September 20 from the 5070 Ti owner, who watched his CPU load while it happened. At full speed the game was using 50 to 60 percent of his CPU. At the 30 FPS floor it was using 15 percent. The game was not working harder as it slowed down. It was working less, and waiting on something.
How fast it leaks
Two players measured video memory directly, and a third-party site ran a timed session. Their numbers agree with each other closely enough to trust.
An RTX 2080 Ti owner, posting September 19, watched dedicated VRAM climb "roughly 100mb every 5-10mins." His card has 11 GB, and usage peaked near that with frame rate at about 28 FPS. After relogging the same scene used 4.9 GB and ran normally. That is about 6 GB of leaked video memory in one session.
A player on the EU forum the same morning, on a 16 GB card, reported the card completely full after two hours of play with a further 4 GB "spilled to system RAM." That spill is the important part. Once a GPU runs out of its own memory the driver starts paging textures across the PCIe bus every frame, and frame rate falls off a cliff regardless of how fast the card is.
Tier1Settings published a measured session on September 20 against builds 1.60.1.69893 and 1.60.1.69913. Committed system memory grew about 29 MB per minute during play. Dedicated VRAM grew 12 to 15 MB per minute. The process handle count grew by about 12 per minute even while standing still. Over 88 minutes the frame rate fell from around 110 to around 40. Logging out to character select and back dropped VRAM from 6.24 GB to 4.58 GB and GPU utilisation went from 22 percent to 52 percent, which is the same underutilisation the CPU-watching player saw from the other side.
Put those together and the leak is roughly:
- VRAM: 12 to 20 MB per minute, so 1 GB every 50 to 80 minutes.
- System RAM: about 30 MB per minute, so 1.7 GB per hour. One player on September 21 reported the client sitting at about 10 GB of RAM.
- Handles: growing even when idle, which means something is being allocated on a timer, not only when you move or load assets.
The rate varies with what you do. Cities make it worse, and layering appears to make it worse, both covered below.
Who it hits
Everyone we could find, on every platform. This is not a driver bug.
- Windows, Nvidia: RTX 2080 Ti, 2080 Super, 3080 Ti, 3090, 4070 Ti laptop, 5070 Ti, 5080, 5090. The 5090 reports matter because a 32 GB card should be immune to VRAM exhaustion for hours, and it still slows down, which points at the system-side leak and the CPU-side stall rather than VRAM alone.
- Windows, AMD: RX 9070 XT, RX 9060 XT, RX 7800 XT. An AMD owner on the EU forum reported on September 18 that since the minor patch the leak "builds up quickly up until there's none left and the pc hangs and crashes/reboots," with a memory dump in Event Viewer. That is the worst outcome reported: a hard system crash rather than a slowdown. It is a single report and we have not seen it repeated at that severity.
- Linux: Nobara, Bazzite and Linux Mint players in the September 21 thread all describe the same slide and the same relog fix. The Nobara player added that it "worsens with layering," meaning the more layer switches you go through, the faster it degrades. A Mint player on an RX 6700 XT documented a related soft lock, ERROR #109 "A thread has become unresponsive," which several others confirmed after the September 18 update.
- Mac: The September 21 thread opens with a Windows player saying a leak was "confirmed for Mac." We looked for that confirmation and could not find it. The Mac Technical Support thread and the 151-reply Mac crash thread are about a different bug, a whole-machine freeze on entering the world tied to graphics quality above setting 1, and neither contains a Blizzard post. Apple Silicon Macs share memory between CPU and GPU, so a leak of this shape would hurt them faster than a discrete card, but the reports we have from Mac players are about the freeze, not the slide.
Cities and layers make it worse
The clearest trigger is entering a capital. A community thread on Barrens Chat traces the pattern: fine in the Skyborne starting zone, then a collapse to about 25 FPS in Ironforge, and no recovery on leaving. AccountShark's week-one summary describes the same thing with Stormwind. A city loads hundreds of player models, each with their own gear textures, and the HD character models in Forever carry far more texture data than the 2004 originals. If the client is failing to release those textures when the players walk out of view, a capital is exactly where you would expect the leak to run fastest.
Layering is the second trigger. Forever has no server list; every player in a region shares one world split into layers by the Familiar Faces system. Each layer switch loads a new set of nearby players and unloads the old set. If unloading is where the leak is, more switches mean more leak, which is what the Linux player reported.
Neither of those is proven. They are the two mechanisms that fit every report, and they are what to test if you want to help Blizzard find it.
Is it actually a memory leak?
Probably, in the plain sense: the client allocates memory during play and does not give it back until the world is torn down. Three things point that way.
First, the growth is monotonic. Nobody reports memory going up and then coming back down on its own. It only comes down on relog, which destroys the world state and frees whatever was attached to it, and on client restart.
Second, relogging is a full fix, not a partial one. Frame rate returns to the starting number, not to somewhere in between. If this were thermal throttling or a driver problem, logging out to a menu would not restore 100 FPS.
Third, both CPU and GPU utilisation fall as the game slows. A game that is genuinely GPU-bound runs the card at 95 percent or more. A game that is CPU-bound pins a core. This client does neither when degraded. It sits at 15 to 22 percent and produces 30 frames a second, which is what a process looks like when every frame is waiting on memory traffic: paging textures over PCIe once VRAM is full, or walking allocation structures that have grown too large.
What we cannot tell from outside is whether it is one leak or two. The VRAM growth and the system RAM growth could be the same objects (a texture leaked on the GPU side has a matching header on the CPU side) or two unrelated bugs that shipped together. The handle growth while idle suggests at least part of it is on a timer rather than tied to loading, which is a different class of bug from a texture that never gets freed.
One comparison worth noting: the AMD player who reported the hard crash said it resembled what happened to Retail after the Midnight patch, and an EU player said the Forever version is "less severe than Retail." Modern WoW has had its own memory growth complaints through 2026. Forever shares that engine. If the underlying bug is inherited, Blizzard may already have a fix in the Retail pipeline. That is speculation and we are labelling it as such.
What Blizzard has said
Very little, and none of it in the places you would look.
The official known issues list, posted by community manager Kaivax on September 17 and updated on September 18, has no entry for memory, VRAM, frame rate, performance, or crashes of any kind. It covers Waylaid Crates, pet character-sheet display bugs, Glancing Blow math, Barber Chair Lua errors and the minimap sitting off screen. Eight of its items are tagged as fixed in the next build. The leak is not one of them.
We read seven forum threads on the subject across the US and EU boards, including the 89-reply FPS thread and the two threads with "memory leak" in the title. None contains a post from Blizzard staff. Barrens Chat's summary of the situation, "Blizzard has yet to respond," was accurate as of September 21 and is still accurate as of this morning.
The one acknowledgement is a reply from the Blizzard customer support account for EU on X. A player reported a "major memory leak on the current beta" and the account replied that the team is "aware and working on fixes and improvements already." We could not load that post directly, so we are relying on the search-indexed text of it, and a customer support reply is a support agent confirming a ticket exists, not an engineer confirming a diagnosis. Take it as evidence the reports have been seen and nothing more.
Several sites are reporting a fix as "on the roadmap" or "patch in development." None of them cites a source for that, and we could not find one. The next beta build will exist because the level cap rises to 30 in the first week of October and that needs a build. Whether the leak is fixed in it is unknown.
Do not confuse this with the server-side memory problem. Senior game producer Tom Ellis posted a breakdown on September 18 of why the servers fell over on opening night, and it mentions that the world simulation pools were "running hot on CPU and almost out of memory." That was Blizzard's hardware running out of memory under a beta-sized environment taking a launch-sized population, and it was fixed the same day. The client leak on your PC is a different bug and has not been addressed.
The workarounds, ranked
1. Log out to character select and back in. This is the fix. It takes 20 seconds, works on every platform, and in the measured cases returns between 1.5 GB and 6 GB of video memory. Do it whenever frame rate has clearly dropped, and do it preemptively before a dungeon. You do not need to close the game; the character select screen is far enough. A full client restart does the same thing and takes longer.
2. Relog after leaving a city. Since capitals appear to be the fastest leak, treat a trip to Stormwind, Ironforge or Orgrimmar as something that costs a relog on the way out. The slowdown does not reverse when you leave the city, so leaving is not enough on its own.
3. Set Secondary Lighting to Fair. Open the System menu, then Graphics, and change Secondary Lighting from its default to Fair. This is the setting that fixes the Mac enter-world freeze and the Linux ERROR #109 thread lock, and a player in the FPS thread reported it "resolved the issue" for the slide as well. It is not clear whether it slows the leak or just removes enough GPU load that the leak takes longer to matter. It costs you the new lighting, which is one of the things Forever added, so try it only if relogging every hour is not enough.
4. Lower texture resolution if you have 8 GB of VRAM or less. The leak eats headroom at a fixed rate, so a card with more free VRAM at login lasts longer before it spills. Texture Resolution is the setting that moves the baseline most. On the laptop 4070 Ti report the player started with roughly 3 GB of headroom and lasted under an hour. Halving texture memory roughly doubles that.
5. Cap your frame rate. Several players noticed their GPU running hot and loud even at modest frame rates. Setting Max Foreground FPS to your refresh rate, or disabling Target Framerate if you had turned it on, reduces heat and power. It does nothing for the leak. It just makes the hour before you relog quieter.
6. Drivers, cache folder, background apps. These are on every performance guide because they are always worth doing, and a couple of sites report large gains from them. We have no evidence they change the leak rate. Update your GPU driver because the beta launched on a brand-new client and driver vendors will be shipping profiles for it. Deleting the Cache folder inside the Forever install directory is harmless and fixes unrelated stutter. Neither is a fix for this.
Two things not to do. Do not buy RAM; the leak will fill whatever you have, it will just take longer. Do not reinstall the client; the leak is in the build, and you will download 60 GB to get the same build back.
How to check whether it is happening to you
You do not need any tools. On Windows open Task Manager, click Performance, then your GPU, and read Dedicated GPU memory. Note it 30 seconds after entering the world, then again after an hour, then again after a relog. On a Mac use Activity Monitor's Memory tab and watch the World of Warcraft process. On Linux, nvidia-smi or radeontop in a terminal. If the number climbs steadily during play and drops back at relog, you have it.
If you want to report it, the Bug Report forum is the right place, and the report Blizzard's engineers can use has: the build number from the bottom-left of the login screen, your GPU and driver version, your operating system, the zone sequence you played (starting zone, then a city, then out again is the most useful), and the memory number at each step. Reports that just say "memory leak" are already in the pile. Reports with the rate are what narrow it down.
What this means for launch
The beta runs to October 21 and launch is November 4. A leak of this size is not something Blizzard can ship on November 4, because the launch-day population will be sitting in Stormwind for hours and every one of them would be at 30 FPS by dinner. It will be fixed. What the beta cannot tell you yet is whether it gets fixed in the next build or the last one.
The next build is the moment to retest. When it arrives, the level cap moves to 30, and the first thing worth doing is the Task Manager check above in a capital for an hour. We will run it and update this page with the result. If the numbers are flat, the leak is gone. If they are lower but still climbing, it was two bugs and one of them is fixed.
Until then: relog every hour, relog after every city, and keep your eye on the number rather than the frame rate. The frame rate is a lagging indicator. The memory is the bug.
What we will update
The build number this is confirmed in, Blizzard's first developer post on the subject if one appears, a measured session on the next build, and whether the Mac client has the same slide or only the enter-world freeze. If you have a timed measurement on hardware not listed above, particularly an Apple Silicon Mac or a 5090, the live streams page shows who is currently playing and the forum threads below are where the engineers are most likely to read it.
If you are still stuck at the launcher rather than in the game, that is a different problem: see the beta access guide. If you are deciding whether your hardware can run the beta at all, the system requirements guide covers what Blizzard has actually published. And if you are relogging anyway, the talent calculator has every tree for planning the level 30 build.
Sources: US beta forum threads FPS in Forever (September 17), FPS Slowly Drops (September 18), VRAM memory leak (September 19), Memory leak? (September 21) and the Linux/AMD ERROR #109 thread (September 18); EU beta forum threads Memory leak (September 19) and Game crashes PC/GPU since minor patch (September 18); Kaivax's WoW Forever Beta Known Issues (September 17, updated September 18); the Mac Technical Support thread and Mac client crashing thread; Tier1Settings' measured session (September 20); Barrens Chat's FPS thread; AccountShark's week-one status (September 21); the Blizzard CS EU reply on X, which we could not load directly; and GamesRadar's report on Tom Ellis's server-side breakdown.