Cromefire_

[Cromefire_ @Cromefire_]

@tekwendell Don't have a cluster, just a standalone node, but I'm still using ceph which it natively supports and can be replicated across the cluster. In my case it helps me run my K8s cluster on top with persistent volume support.

[Cromefire_ @Cromefire_]

@tekwendell Did you see the news of potentially having FSR4 on older GPUs using MESA? Might be worth a video to see if it's any good and maybe there's a way to get it working Windows as well if it is?
https://t.co/2CtY8cfTvP (only found reporting in German for now though)

[Cromefire_ @Cromefire_]

@FFmpeg @isostandards It's the same issue with the new gain map JPEG standard now before it was just Google's own standard but they just released the standard along with a OSS impl to the web, now it's slowly superseded, but with an IOS standard, so now you can't just implement it...

[Cromefire_ @Cromefire_]

@Am252230Moses @R4K1B_ @compguru910 @LupintheI Which is exactly what I said...

[Cromefire_ @Cromefire_]

@HardwareUnboxed @tekwendell But that would sound kinda correct for prediction fixes. Engines and even engine versions look very different on their critical path, some benefit a lot from better prediction, while some were already predicting well. And the second predictor shouldn't make anything worse.

[Cromefire_ @Cromefire_]

@Am252230Moses @R4K1B_ @compguru910 @LupintheI Alan Wake's results seem far more like what you should expect from AMDs inferiors cores: -30-50% not -300%.

[Cromefire_ @Cromefire_]

@MuhammadHameedo @LupintheI Yeah it's always slower, but they are usually only 30% in RT slower not like 300%. Somehow games with NVIDIAs RT tech always run far far worse than when developers or Unreal implement it (although of course all current UE games with Software RT basically). Look at Snowdrop.

[Cromefire_ @Cromefire_]

@beyond_fps @LupintheI The Ubisoft RT runs well on all GPUs though, while NV RT seems to not be one but optimized for anything other than NV. I'm ready surprised that they don't outright just crash.

[Cromefire_ @Cromefire_]

@IWaseemP I'd like to see way more Kotlin, but I know it's gonna be JS and Python, because people can't seem to get enough of shooting themselves into their own foot...

[Cromefire_ @Cromefire_]

@uranusjr @snappercayt @marcinmoskala @perplexity_ai At least something. No local is the safer fallback in most cases because it's at least predictable. You mostly only want localization in UI, C#'s localization by default has caused so many bugs, I'm considering writing an Analyzer to just disallow anything without invariant.

[Cromefire_ @Cromefire_]

@sameerlove4u2 @marcinmoskala The kotlinconf talk linked in the followup tweet

[Cromefire_ @Cromefire_]

@snappercayt @marcinmoskala @perplexity_ai The question is whether python does it correctly?

[Cromefire_ @Cromefire_]

@Linux4Everyone It's easy though to get new drivers thanks to HWE and amdgpu-install. Yes you may want to wait to wait 3-6 Weeks after launch of new GPUs, but in return at least everything else isn't constantly breaking. And AMD and NVIDIA do at least officially support those platforms.

[Cromefire_ @Cromefire_]

@nelomah @plasmicnebula @WagieCapital @RobDenBleyker Not really an evolution of the word, but something forced onto the entire field to just sell more stuff. Now it's even used to sell non-ML stuff. And things like AMD Ryzen AI 300, srsly? Yes it has an NPU, but that doesn't justify declaring it "AI". Marketing is just crazy.

[Cromefire_ @Cromefire_]

@nelomah @plasmicnebula @WagieCapital @RobDenBleyker or "ML girlfriend" or "ML driven data analysis" around as cool as the same thing with "AI" to the public? Within the more technical circles people talk about ML models, LLMs and transformers etc. but the public know AI from the Movies so they "rebranded" it basically.

[Cromefire_ @Cromefire_]

@nelomah @plasmicnebula @WagieCapital @RobDenBleyker Basically when you look back to like around 2015-ish or before, much of the real definition was still intact, while almost all recent sources have been corrupted by the marketing that really ramped up with ChatGPT back then.

[Cromefire_ @Cromefire_]

@nelomah @plasmicnebula @WagieCapital @RobDenBleyker https://t.co/47my7t0KVd
Wikipedia, before marketing got to it.

More recent, kinda still has some of the original definition there, although they also slap the "AI" label on the whole file now and fell for the whole weak/strong thing: https://t.co/h47evpBhCp

[Cromefire_ @Cromefire_]

@edgar2318 @plasmicnebula @WagieCapital @RobDenBleyker So you're out of arguments... Good talk I guess. Next time maybe pick an area in which you have expertise.

[Cromefire_ @Cromefire_]

@edgar2318 @plasmicnebula @WagieCapital @RobDenBleyker That's what marketing has dubbed AI, yes, maybe you finally got it.