Game: Hyrule Field slowdown and severe rendering corruption in Castle Town on RTX 3070 #1215

Open
opened 2026-11-16 19:30:28 +00:00 by BeanIssues · 2 comments
Collaborator

Reported on Discord

Longshot version and channel

Not provided.

What happened

The reporter says performance was smooth in Kokiri Forest and the dungeon, then began lagging upon reaching Hyrule Field. A supplied screenshot also shows severe scene/geometry corruption in Castle Town, with fragmented and floating buildings and platforms, and bright green-white corruption obscuring the lower scene; the image displays 25 FPS.

Where in the game

Hyrule Field (slowdown); Castle Town, labeled “INNKEEPER CASTLE TOWN” in the screenshot (rendering corruption).

Graphics card and driver version

NVIDIA RTX 3070, 8 GB. Driver version not provided.

System

Ryzen 7 5800X, 16 GB RAM, 1080p. OS not provided.

Steps and frequency

Reporter reached Hyrule Field after playing the first part of the game, including Kokiri Forest and the dungeon, and then noticed lag. Exact reproduction steps and frequency are not known. The screenshot documents the Castle Town corruption.

Not known yet

Longshot version/channel, OS, NVIDIA driver version, exact reproduction steps and frequency, graphics settings, and a performance profile/log. Please collect a profile as described at https://longshot.magicbean.dev/help/performance/: enable measurements, reproduce the slowdown for 60–90 seconds, then provide the Bug report folder and longshot_log.txt.

Longshot #348 tracks Hyrule Field slowdown on a different RTX 3070 system, while riding Epona as adult Link; this report does not establish that it is the same condition. #348

Reported on [Discord](https://discord.com/channels/1569849200047427645/1569849200617726065/1571965212662439937) ### Longshot version and channel Not provided. ### What happened The reporter says performance was smooth in Kokiri Forest and the dungeon, then began lagging upon reaching Hyrule Field. A supplied screenshot also shows severe scene/geometry corruption in Castle Town, with fragmented and floating buildings and platforms, and bright green-white corruption obscuring the lower scene; the image displays 25 FPS. ### Where in the game Hyrule Field (slowdown); Castle Town, labeled “INNKEEPER CASTLE TOWN” in the screenshot (rendering corruption). ### Graphics card and driver version NVIDIA RTX 3070, 8 GB. Driver version not provided. ### System Ryzen 7 5800X, 16 GB RAM, 1080p. OS not provided. ### Steps and frequency Reporter reached Hyrule Field after playing the first part of the game, including Kokiri Forest and the dungeon, and then noticed lag. Exact reproduction steps and frequency are not known. The screenshot documents the Castle Town corruption. ### Not known yet Longshot version/channel, OS, NVIDIA driver version, exact reproduction steps and frequency, graphics settings, and a performance profile/log. Please collect a profile as described at https://longshot.magicbean.dev/help/performance/: enable measurements, reproduce the slowdown for 60–90 seconds, then provide the Bug report folder and `longshot_log.txt`. ### Related reports Longshot #348 tracks Hyrule Field slowdown on a different RTX 3070 system, while riding Epona as adult Link; this report does not establish that it is the same condition. https://git.magicbean.dev/MagicBeanDev/Longshot/issues/348
Author
Collaborator

Files from the Discord report:

Files from the Discord report: - [image-74.png](https://git.magicbean.dev/attachments/2e326c74-6303-4747-b9f7-a3c1a3fe4f38)
Member

Automated run for this issue: a walk through Castle Town, not reproduced on this hardware (partial coverage)

This is an automated scenario run from the test player described in #1198 (comment), written from this issue's text. The build was a build of #68 at commit 63ddacdc9, played on an NVIDIA GeForce RTX 4090 under Linux, headless at 1280x720. It is one run on one machine; it says what happened there and is not a check that anything is fixed.

What was tried. The report is severe rendering corruption in Castle Town (floating, fragmented buildings, green and white patches) at about 25 frames a second, on an RTX 3070 under Windows. The report gives no time of day, settings, place or steps, so the run put Link at the Market as a child at noon and walked 30 generated points through the town for about 4 minutes. Local picture checks looked at 348 pictures taken during the walk, and an image model also looked at 8 of them, for corruption in the picture.

Result: not reproduced on this hardware. No picture was flagged as corrupt, the game did not crash, and 3 frames in the whole walk took longer than 40 milliseconds. The limit is the coverage: Link got stuck against stalls and people on 32 of the 53 legs and reached 20 of them (it walked 249 metres in all), so parts of the town were not seen. A clean run on an RTX 4090 under Linux does not answer a report from an RTX 3070 under Windows.

Two pictures from the walk at noon: a lane with flowers and a barrel, and the market square with people and a "Talk" prompt. Nothing looks broken in either.

What would help.

  • The reporter's screenshot's place: which street or square, and from where they looked at it (the in-game bug report gives only the map name today; coordinates would let the scenario go straight there).
  • The time of day and the graphics settings (resolution, any upscaler or frame generation) of the report.
  • A run on hardware like the report's (an RTX 3070 or similar under Windows), or a save in the town from the report.
**Automated run for this issue: a walk through Castle Town, not reproduced on this hardware (partial coverage)** This is an automated scenario run from the test player described in https://git.magicbean.dev/MagicBeanDev/Longshot/issues/1198#issuecomment-39088, written from this issue's text. The build was a build of https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/68 at commit 63ddacdc9, played on an NVIDIA GeForce RTX 4090 under Linux, headless at 1280x720. It is one run on one machine; it says what happened there and is not a check that anything is fixed. **What was tried.** The report is severe rendering corruption in Castle Town (floating, fragmented buildings, green and white patches) at about 25 frames a second, on an RTX 3070 under Windows. The report gives no time of day, settings, place or steps, so the run put Link at the Market as a child at noon and walked 30 generated points through the town for about 4 minutes. Local picture checks looked at 348 pictures taken during the walk, and an image model also looked at 8 of them, for corruption in the picture. **Result: not reproduced on this hardware.** No picture was flagged as corrupt, the game did not crash, and 3 frames in the whole walk took longer than 40 milliseconds. The limit is the coverage: Link got stuck against stalls and people on 32 of the 53 legs and reached 20 of them (it walked 249 metres in all), so parts of the town were not seen. A clean run on an RTX 4090 under Linux does not answer a report from an RTX 3070 under Windows. ![Two pictures from the walk at noon: a lane with flowers and a barrel, and the market square with people and a "Talk" prompt. Nothing looks broken in either.](https://git.magicbean.dev/attachments/749cd646-1a56-427e-89ef-6de6b39fb3b0) **What would help.** - The reporter's screenshot's place: which street or square, and from where they looked at it (the in-game bug report gives only the map name today; coordinates would let the scenario go straight there). - The time of day and the graphics settings (resolution, any upscaler or frame generation) of the report. - A run on hardware like the report's (an RTX 3070 or similar under Windows), or a save in the town from the report.
Sign in to join this conversation.
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot#1215
No description provided.