There seems to be a memory leak or some performance related issue with the site which has happened within the last week or two. As soon as I load the site my CPU temp climbs, cpu fan ramps up to maximum, and eventually given enough time the page crashes due to memory usage being exceeded.
I currently use, and have always used Opera browser without any issue until now (aside from severe lag if I try to run any prompt without closing the history slideout on the site) and here are the details:
Opera One(version: 109.0.5097.68)
Update stream:Stable
System:Windows 10 64-bit
Chromium version:123.0.6312.124
I haven’t changed anything with my browser configuration that could have caused this, it simply started happening sometime after Vector Stores were released I believe after April 17, 2024
Essentially what it does is the same thing that I’ve experienced before when I visit a site that has a malicious crypto miner injected within the site, CPU heats up like mad, fan is at max and still doesn’t help much with the temp, and everything gets really laggy on my entire system.
This is the ChatGPT page? You put API in your tags.
How large is your History?
Lastly, what areyour computer specs? I just took a snapshot of random intervals and haven’t noticed anything crazy. I also don’t have any history.
Looking a bit deeper into the history issue closing it really shouldn’t do anything at all, honestly. On the surface it looks like they just literally reduce the width and then apply a visibility:hidden modifier, which keeps it in the DOM. Irrelevant
It doesn’t do anything nefarious. It just grabs all your playground history and returns the length.
(You’ll need to be on an platform.openai page, preferably the playground for this to work)
For additional context, typically localStorage only allows for a maximum of 5MB per domain. I do believe that this limit can be adjusted though. Who knows
I just did some tests, and there doesn’t seem to be a memory leak. The memory footprint is about half of a normal Wikipedia page, and goes down over time.
Was playground reached through openai.com, or by a direct platform link, one can ponder.
It does show that they need to target all browsers for testing, even the 0.1%, or links, although OpenAI has been doing better at not breaking everything.
Selecting a huge amount of text on this forum will freeze a browser, such as picking everything from topic title to the footer, BTW.
Remember the good ol’ days of “why can’t 32-bit Windows use all of my massive 4GB of RAM?” - the 2GB limit of 32-bit applications still applies.
I only gained 300MB by exercising the playground in different browser than Opera. Memory footprint not an issue to me.
(
The playground history does have a bad behavior - if clicking a history sidebar item, there’s no escape except “restore” button, or browsing up to the top of the history list to “current editor”.)
Okay I was wrong, I’ve looked into it more, it’s not a memory issue, it’s something on the site causing high CPU usage, the out of memory page crashed because of me killing that thread in task manager . The memory usage is pretty low, but the CPU usage is just maxed out. It is definitely something inside of the playground. I noticed a while ago when I was in the browser dev console that the playground was spamming a ton of requests for the available models. so maybe that’s a clue but I think that might be unrelated,
My PC specs aren’t the best but this type of thing has never happened before until now and I never had the CPU getting maxed out with the fans at 100% and my CPU temp around 80
Let’s keep in mind: Both the Playground and Docs are platform.openai.com and use React. It’s common to run into hellish infinite render loops. Considering that this is only happening to you, I think it’s fair to assume that some state is messing things up.
What happens if you view the docs & use the playground while signed in on incognito mode?
A year or two ago I switched to opera because both Chrome and Firefox were laggy for me yet Opera loaded fast, had a nice sidebar that I used to like when I used to use Maxthon browser. The Opera sidebar had Telegram, Youtube Music, and eventually my favorite addition… ChatGPT right from the side bar (which has advantages over opening it in a separate window or in a second tab since I could pin it open and have it there between tabs in the same window)…
It’s funny you mentioned this, because as of today it stopped working and just says “Error” when I try using that in the Opera Sidebar which is strange to me given that they have their own Aria AI which is built on OpenAI APIs
Yeah Opera’s got a “browser.js” file that it injects into every page that is kind of funny to read through, it’s just like a bunch of random hard coded alterations to specific sites like some developer would have as a personal userscript to fix all their annoyances they have on their frequent used websites
I personally would recommend Arc, not only does it not steal your data like Google, but it also loads super quickly and has some nice features with and without AI.
With Arc Max (not a paid subscription), you can ask ChatGPT right from your search browser (you can turn this off if you don’t like it). This opens up chat.openai.com and uses some sort of web automation tool to automatically input your search query into the prompt box. (NOT AN INJECTION)
In the screenshot above, you may also see Instant Open, which automatically opens the most relevant website to your prompt. For example, you could look up Wikipedia and click Instant Open, and wikipedia.org would open instead of having to go through a search engine. These are just a few fun addons it has, not counting its nice UI and Spaces feature.
EDIT: They have some gamification features so if you download it using this link, whether to test or to use, I get a new Arc application icon: A friend is sharing Arc with you!
The problem is caused by previous chats (displayed in the left tab) that ChatGPT has not optimized to minimize memory usage. As a result, the more chats you have stored, the more memory any new conversation will start off with.
For someone like me who uses ChatGPT daily for work and has accumulated hundreds of chats, it has become nearly impossible to continue a conversation without encountering high memory usage. Google Chrome’s maximum tab memory is around 2.2GB, and reaching this limit causes significant slowdowns and potential crashes.
So, even a brand new, empty chat session takes up around 350MB of tab memory. This indicates that the memory footprint of previous chats is substantial and affects the performance of new conversations.
OpenAI needs to optimize the memory usage of previous, unopened chats to prevent them from consuming excessive resources during active sessions. This optimization is crucial for users who rely on ChatGPT for extended and frequent use.
As a temporary workaround, when you notice the chat slowing down and sense an impending crash, you can open a new tab and continue the same conversation there. This approach resets the tab memory without losing the conversation. However, this is only a temporary fix, as the tab memory will increase rapidly with each new prompt.
In summary, optimizing the memory usage of previous chats is essential to improve the overall performance and user experience. I hope this information helps both users understand the issue and developers at OpenAI identify and resolve the problem efficiently.
I tried deleting all my previous chats. I still have a high memory and tab crashing error. I can’t send more than 2 prompts before the tab exceeds 5 GB and dies. This is making ChatGPT unusable.
