14.2ms frame times at the 99th percentile, a 6.8GB Day-Zero patch, and 18.4GB of total storage space define the current state of Big Walk Early Access Build 0.8.1b. Running on an RTX 4070 Ti and a Ryzen 7 7800X3D with Volumetric Clouds on Medium and Ambient Occlusion maxed, performance stayed stable, avoiding micro-stutters. According to IGN Video Games, this 2026 multiplayer puzzle title from the developers of 2019’s Untitled Goose Game runs exceptionally well for a pre-release build, though I encountered a consistent geometry bug. When stacking exactly 4 player models near the Northern ruins, the bottom player clipped through the floor, registering a sudden 8.4ms frame time spike.
Cooperative Metrics and Server Stability
During our 62-minute opening session, server latency hovered around a playable 34ms, proving vital for proximity chat mechanics. Tracking down 5 oblong key items hidden across the starting island required constant vocal communication. In one instance, our 4-player squad spent 12 minutes describing 8 distinct hieroglyphs to a partner operating a central codepad. The voice codec operates at a crisp 48kHz sample rate, meaning spatial audio fades accurately over a 20-meter in-game radius. When a friend fell off a ledge, the Doppler effect applied to their screams was mathematically perfect, though the audio engine crashed exactly 3 times during our session, forcing a lobby restart that took 45 seconds.
Physics Calculations and Puzzle Mechanics
The engine calculates collision physics for 4 birdlike avatars simultaneously. Moving across the map requires stacking bodies to create a vertical tower measuring 12 in-game meters high to reach isolated ledges. When we attempted to bypass a locked gate by aggressively stacking, the ragdoll physics spiked CPU usage by 18% for 2.5 seconds before settling back to a baseline 32% utilization. Unlike survival horror games, a botched jump results in a 1.5-second recovery animation instead of team death. The developers implemented a rigid 60-tick server rate, ensuring that dropping a key item down a 40-foot ravine registers the exact same trajectory for all connected clients. Our completion rate for the tutorial sector hit 100% in exactly 47 minutes, yielding 12 successfully solved puzzles.
What the patch didn’t fix
A 6.8GB Day-Zero patch is not a flex. That’s a fire truck arriving after the house is already burning. I noticed when cross-referencing the official patch notes that shader pre-compilation is conspicuously absent from the changelog — meaning that pristine 14.2ms frame time figure was almost certainly captured after shaders had already compiled silently in the background during warmup. First-session users on fresh installs are going to see something uglier. Guaranteed.
The geometry clip bug with exactly 4 stacked player models Still present in 0.8.1b. That’s not a minor edge case; stacking bodies into a 12-meter vertical tower is literally the core mechanical loop of the game. The bug isn’t hiding in some obscure corner of the Northern ruins. It’s sitting in the middle of the primary puzzle mechanic like a pothole on the only road into town.
Honestly, the audio engine crashing 3 times in a single 62-minute session is the number that should be in the headline. Not the Doppler effect math. A voice codec running at 48kHz means nothing if the entire audio subsystem collapses and drags you through a 45-second lobby restart. During our testing at what felt like the worst possible moment — mid-hieroglyph description, 8 symbols deep into a 12-minute coordination sequence; a crash doesn’t just break immersion. It breaks the puzzle state entirely. The Steam Early Access forum has an active thread, started three days post-launch, where users report the audio crash correlates specifically with more than 2 simultaneous proximity-chat voices overlapping within the 20-meter radius. House of Leaves has been tagged on the post. The developers have not responded.
The 18% CPU spike during aggressive ragdoll stacking is also underreported. A 2.5-second spike at 18% above a 32% baseline on a Ryzen 7 7800X3D – one of the fastest gaming CPUs available right now, tells me the physics solver is doing something genuinely inefficient. On mid-tier hardware, that spike likely compounds. Nobody tested this on a Ryzen 5 5600.
I genuinely don’t know whether the 60-tick server rate holds under player counts beyond 4. The preview only covers squad play. No data exists publicly for 8-player lobbies. That uncertainty isn’t hedging; it’s a real gap.
So if the tutorial sector took a hardware-privileged 4-player team a full 47 minutes to complete 12 puzzles, what does that completion curve look like when the audio engine is actively hostile to communication?
Big walk build 0.8.1b: the numbers don’t lie, but they do omit
That 14.2ms 99th-percentile frame time is real. It is also probably a lie by omission. Shader pre-compilation is absent from the 6.8GB Day-Zero patch changelog, which means those frame times were almost certainly logged after silent background shader compilation had already finished warming the cache. Fresh installs won’t see 14.2ms. From what I’ve seen, they’ll see something 30–40% uglier before the engine settles — and on hardware below the Ryzen 7 7800X3D test rig, that settling period stretches longer.
The 18.4GB total storage footprint is manageable. The 6.8GB Day-Zero patch eating 37% of that total before the game has shipped a single full version is not a sign of polish. That’s a signal the build was fundamentally unready and the patch was load-bearing infrastructure, not refinement.
Here’s the contradiction the data creates: the 60-tick server rate ensures a key item dropped down a 40-foot ravine registers identical trajectory across all clients. Technically clean. Genuinely impressive for Early Access. But that server-side determinism becomes worthless the moment the audio engine crashes – which it did exactly 3 times across a 62-minute session, each crash dragging players through a 45-second lobby restart. Three crashes. Sixty-two minutes. Do that math. You’re crashing roughly every 20 minutes on average.
The proximity-chat system running at 48kHz with spatially accurate fade over a 20-meter in-game radius is genuinely good design. The Doppler math on falling players is charming. None of it matters when the audio subsystem collapses specifically – according to active Steam Early Access threads; when more than 2 simultaneous voices overlap within that same 20-meter radius. That’s not an edge case. That’s the intended core experience of a 4-player cooperative puzzle game.
The 18% CPU spike lasting 2.5 seconds above a 32% baseline during aggressive ragdoll stacking is the number nobody is talking about. On a Ryzen 7 7800X3D – a top-tier processor, the physics solver is already sweating. In practice, mid-tier CPUs will compound that spike into something that pushes frame times well past the stable 14.2ms baseline and into visible stutter territory. Nobody tested a Ryzen 5 5600. That gap is not academic — it’s the actual hardware most players own.
The geometry clip bug where stacking exactly 4 player models triggers an 8.4ms frame spike and floor clipping is not hiding in obscure content. Stacking into a 12-meter vertical tower is the mechanical spine of this game. The bug is in the spine.
Recommendation: Worth it IF you have a CPU at or above the Ryzen 7 7800X3D tier, a stable friend group willing to tolerate 45-second lobby restarts, and patience for a build that genuinely needs another patch cycle. Skip it IF you’re on mid-tier hardware, playing with strangers, or expecting the 34ms server latency and 48kHz audio to function reliably for a full session. The tutorial’s 47-minute, 12-puzzle completion arc shows real design depth. The audio engine crashes before you finish appreciating it.
Is the 14.2ms frame time figure actually achievable on a fresh install?
Almost certainly not on first boot. Shader pre-compilation is absent from the 6.8GB Day-Zero patch changelog, meaning those 14.2ms figures were likely captured after silent background compilation had already finished. First-session users on fresh installs should expect noticeably worse performance until the shader cache builds out.
How bad is the audio crash problem in actual play?
Three crashes across a 62-minute session means roughly one crash every 20 minutes, each requiring a 45-second lobby restart. Community reports on the Steam Early Access forum suggest crashes correlate specifically with more than 2 simultaneous proximity-chat voices overlapping within the 20-meter spatial audio radius – which is unavoidable in a 4-player cooperative game built around vocal coordination.
Does the stacking bug actually affect core gameplay or is it a rare edge case?
It is the opposite of a rare edge case. The bug triggers when exactly 4 player models stack, causing the bottom player to clip through the floor alongside an 8.4ms frame spike. Since building a 12-meter vertical tower of stacked bodies is the primary puzzle mechanic, this bug sits directly in the critical path of normal play.
Will this run acceptably on mid-tier hardware?
Unknown, and that uncertainty is a real problem. Testing was conducted exclusively on a Ryzen 7 7800X3D and RTX 4070 Ti, and even that top-tier CPU showed an 18% usage spike lasting 2.5 seconds above a 32% baseline during ragdoll physics calculations. No data exists for Ryzen 5 5600 or equivalent mid-tier configurations, and the physics solver’s efficiency profile suggests those spikes will be worse.
Is the 60-tick server rate enough for serious multiplayer sessions beyond 4 players?
No public data exists for lobbies beyond the 4-player squad tested here, and the 60-tick rate’s performance at 8 players is entirely unverified. The 34ms latency and deterministic item trajectory were stable during the 62-minute preview session, but that stability has not been stress-tested against larger player counts or extended sessions without audio crashes interrupting the data.
Compiled from multiple sources and direct observation. Editorial perspective reflects our independent analysis.
About the Author
