Chicho said:
JEMC said:
The GT 1030 case is well known, but it's an isolated case, not a norm. Usually all GPUs from a generation use the same kind of VRAM with the biggest difference being the speed of the modules. That's why I wondered if it was still possible to use different kinds of memories without designing a new memory controller, or if it's something of the past. |
Amd did the same they took a 5500 and gave it GDDR5 instead of GDDR6 but they called it 5300XT so there was no complains. |
Which is incidentally what inspired me of the idea, as it would be pretty similar. Though I didn't remember how big the gap was in MT/s between GDDR5 and GDDR6; using GDDR5 would cut the bandwidth to about a third, which is way too slow even for an entry-level GPU.
Five-year old GPUs alludes to the professional GPUs like Hopper, but on desktop we got the RTX 3000, which is so memory-starved that the generation, which looked great at first glance, aged pretty badly.
| JEMC said: MODS, EMULATORS & FAN PROJECTS Popular PlayStation emulator clamps down on AI submissions: 'Leave behind something useful to humanity when you're gone, instead of peddling slop'
https://www.pcgamer.com/games/popular-playstation-emulator-clamps-down-on-ai-submissions-leave-behind-something-useful-to-humanity-when-youre-gone-instead-of-peddling-slop/
Open-source dev teams have been fighting a rising tide of AI slop contributions ever since LLM tech was loosed upon the Earth, as we documented earlier this year with news that free and open-source game engine Godot was buckling under the weight of AI-generated pull requests (or PRs—that is, code contributions asking to be 'pulled' into the main project).
Well, here's another one. As spotted by GamingOnLinux, popular PS3 emulator RPCS3 has had to come out swinging against the number of LLM-generated pull requests currently being submitted to the project. In a post on X, the devs put it plain: "Please stop submitting AI slop code pull requests to RPCS3. We will start banning those who do without disclosing.
"There are plenty of resources online to learn how to debug and code instead of generating slop that you don't understand and that doesn't work."
Expanding in the replies, the team wrote that RPCS3 has been receiving greater and greater numbers of AI-generated PRs for its macOS build in particular, which most of the project's team do not use save one member, who also handles most of the testing for the emulator on Apple systems. "We have had to revert a few slop PRs that caused big regressions a few times, enough is enough," wrote the team.
In response to questions (and more than a little AI evangelist bellyaching), the devs pointed out that it is not the use of AI code in PRs that is the issue for them, but that it is undisclosed. "We won't ban if disclosed, except for abuse cases, e.g. throwing a lot of random slop at us to see what passes reviews. Hint: programmers that can understand the problem, the solution, and the implementation can write the same code without AI, and tend to use LLMs to automate repetitive code refactoring instead. It is not the case with the AI slop PRs we have seen." |
This will definitely not help making emulators better.
This here, however...
https://www.youtube.com/watch?v=oYjYmSniQyM
Wonder if similar tests can be made for other, newer consoles, too.
