This is an odd issue on a older gaming desktop. The sliced graphics display is not displaying correctly. I also have a Prusa and its graphics displays okay on the same computer. Guidance needed to resolve this issue.
Nvidia graphics card I suppose, try downgrading its firmware (or upgrading if one is available). Playing with Nvidia graphics settings first on your current firmware might help
I am familiar with this issue and it would take way too long to walk you through the dozens of options in a forum post. So instead of giving you a fish, I am going to teach you how to fish. 
Please note that I have over 40 years of Desktop admin experience and this method has been a godsend for even someone with my skillset in terms of cutting down on troubleshooting and remedying exactly this sort of issue. In the end, you may need to resign yourself that this simply won’t work as your hardware is just too old but it is worth a shot if you can find the motivation to work through this process.
First step before anything else. Go into the Slicer’s preferences and if something like “render using OpenGL” is available, turn it off or on. If I recall correctly, it was removed after v1.9 so it may not be in your version of the slicer. Try that first.
Apparently, this is a known defect in older AMD graphics drivers. Now before you go reaching for the ■■■■■■■■answer that tech support will give you “Update your driver”, we all know that never works. Try this troubleshooting process.
Go into ChatGPT, Gemini or whatever AI you prefer. Paste this prompt below into the AI and have it walk you through the steps.
Note that part of the diagnostics will be to drop down to the command prompt and/or PowerShell in Administrator mode. If you are not familiar with operating from the command shell or terminal, ChatGPT can walk you through it but know that you will be making changes to your graphics driver settings at the OS level that may impact other programs. Before you issue any of the commands, issue a secondary prompt that states “Remember, I may want to reverse all of these steps at the end”.
What CPT may do is offer to write a PowerShell script which you must copy and paste into a text file using Notepad and rename it with a “.ps1” extension. Then from PowerShell, you simply navigate to the folder where you saved that file and issue the command from the prompt in this format (note: you do not want spaces in your file name):
./command_name.ps1
Note: any commands it asks you to perform should be done in Admin mode(command prompt or PowerShell) to ensure that the system security will not block you.
Paste the following prompt into the AI verbatim:
I am using OrcaSlicer / Bambu Studio on Windows with an older AMD discrete GPU (AMD Radeon R5 240, GCN 1.0, DDR3 VRAM).
Symptom:
- The 3D preview renders outer walls as partially transparent
- Internal infill is visible through solid perimeters
- Geometry and slicing are correct and G-code prints normally
- The same model renders correctly on other GPUs
- This appears to be a GPU / OpenGL rendering issue, not a slicing issue
Hardware details:
- GPU: AMD Radeon R5 240
- Driver: Legacy AMD Windows driver
- OS: Windows
- Slicer: OrcaSlicer or Bambu Studio (Qt + OpenGL based)
Task:
Act as a diagnostic assistant and walk me through troubleshooting ONE STEP AT A TIME.
Process rules:
- Provide only a single troubleshooting step per response
- Each step must include:
a) Exact action to perform
b) What output or visual result to expect
c) What conclusion can be drawn from each possible outcome
- After each step, explicitly instruct me to copy and paste:
- Console output, error messages, or
- A clear description of what changed or did not change in the 3D preview
- Do NOT advance to the next step until I provide the requested output
Include steps that cover:
1) Controlled tests that prove the issue is GPU rendering and not slicing or geometry
2) Verification of OpenGL behavior on legacy AMD drivers
3) Environment-variable methods to force software OpenGL or alternate Qt rendering paths on Windows (QT_OPENGL options)
4) Confirmation criteria for success or failure of each rendering path
5) Final remediation options once the root cause is confirmed
Constraints:
- Do not assume any GUI toggle exists for hardware acceleration
- Do not suggest STL repair, re-slicing, or filament/profile changes
- Keep the process verification-first and reproducible
- Focus on isolating and proving the rendering failure before suggesting fixes
Here’s what that interaction might look like in ChatGPT
Here is what Google’s Gemini
___________________________________________________
Disclaimer: You cannot let the AI think for you. You must read and follow its instructions but if you don’t understand something then use this prompt after it asks you to perform a task: “Explain this to me in layman’s” terms.
My ChatGPT has been trained by me to troubleshoot desktop issues and I’ve trained it to write code with my coding preferences. You’re not likely to get the identical troubleshooting steps someone else might experience but it will still guide you through the important steps. In my case, CPT will remember that I prefer scripts so it will always offer a PowerShell script. Your experience may be just a series of prompts which is a longer more tedious process. The prompt I gave you should have the AI ask you to paste a screengrab of the output of each step so it can help you validate what has happened. If it doesn’t ask you for a screenshot confirmation, then just copy and paste a screen grab anyway and ask it “verify output”.
2 Likes
Thanks, I’ll start this process and see how it goes. I thought it might be something like modern code working with old, deprecated GPU hardware/ drivers. I’m familiar with CMD Prompt and scripts. Enough to fumble around.
I too use an AI to assist in various engineering ventures. It has been a learning experience for both AI and me. I find it most useful parsing device specifications that are poorly written or incomplete.
Do you recall in the end what the fix was? In general terms like, editing a driver file or setting some global variable Bambu Studio pays attention to, or ?
Yes, I do. On AMD-only systems, the fix was to use Windows system environment variables to force Qt (and therefore Bambu Studio) off the AMD OpenGL hardware path and onto software rendering. This effectively disables hardware OpenGL for Qt-based apps. Here’s an example screenshot which will show you a snippet of the results CPT provides, the actual remedy is pages long.
Two important clarifications that I should point out:
-
This is not a Bambu-specific setting. These are global Qt/OpenGL variables. That means other Qt applications may also be affected, which is why you’ll want to instruct the AI to provide a solution which include a clear reversal path. But you have to prompt it to do that, it won’t do it by itself. The prompt I provided above should have included that but doublecheck.
-
Because of that scope, a PowerShell wrapper that sets and clears the variables only for the Bambu Studio session is the safer approach. An AI will generate such a script if explicitly asked, but as I just stated, it will not propose that mitigation on its own unless prompted.
Although I used an AI to work through this, it still required supervision and trial-and-error. The value of the AI here is synthesis of a wide range of technical information from the net. It pulls together scattered forum posts, driver notes, and partial documentation that no single person can possibly read end-to-end. The downside is that it will confidently repeat incorrect material if not challenged.
________________________________________
Click here for comments on the perils of relying on AI.
There’s an old computer science adage most of us learned in first-semester coursework: “Garbage in, garbage out.” In the context of the internet, the volume of low-quality information available to train and inform AI systems makes this a real problem, so caution is necessary.
I routinely catch GPT making factual errors, and when I ask it to cite sources, the trail often leads back to low-signal Reddit posts built around “I heard that…” speculation, which the model often incorrectly treats as authoritative. That is one of the reasons Reddit is my last source of information. I even run a userscript that blocks Reddit from appearing in my Google search results because the noise-to-signal ratio has become that poor.
It is also worth noting that this forum is frequently cited - and frequently incorrect - in 3D printing troubleshooting contexts. I have seen GPT suggest links to threads here where the cited answers were written by me, even though my prompt had nothing to do with that topic and the post I put up was helping someone on an unrelated problem. 
That moment is amusing, but it also illustrates the core risk: if the system is demonstrably wrong about subjects you personally know well, blind trust is unjustified.
AI can be an effective tool but only if you supervise it closely and continuously verify its claims.
A workflow that helps me keep the AI honest:
- I Keep one chat window open during active troubleshooting.
- I Keep a second chat window open purely for verification.
If something fails a basic sanity check in the verification chat window, I challenge it in the troubleshooting chat and instruct the AI to prove its own claim. Then you get the inevitable “You’re right to call me out” from CPT and it corrects itself.
Also worth noting: since GPT 5.2, error rates and just plane ■■■■■■■■answers (AI hallucinations) have skyrocketed compare to earlier versions (e.g., 4.1). IMHO, CPT has reverted back to that argumentative teenager who simply only wants to do have the job and punch out for the day. So the same rule as always applies as to real people - “trust but verify”.