Making Wukong comparision. Even in youtube quality we can see difference, especially right-btm corner
Nah, Turing raster was worse especially with the increase in area.Turing was actually better than this.
Or at least I think it was.
Thats a lot of smearing! Looks like we should stick to 2X FG. Are you sure they're using release quality drivers or dev builds or beta?Making Wukong comparision. Even in youtube quality we can see difference, especially right-btm corner
He's enthusiastically misleading and comes across more and more as a once smart engineer talking marketing fluff and taking people for duds. He has been blowing the trumpet of how AI will replace programming and as if everything will be robotic or done by bots, and that people will gleefully engage with "agentic" robots as if they're real people.I think people need to stop giving Jensen the benefit of the doubt just because he's the CEO of the most valuable company at time when the AI bubble is hyped like crazy. Just because he's successful doesn't mean he's automatically right. I'd argue it's more likely he's wrong because the people who are going to be the most knowledgeable in a particular subject matter are his engineers who developed it. CEOs tend to have a broad, general understanding of stuff. This error in his statement about how DLSS 4 MFG works is likely a result of that.
Actually, it's started already. Google said that almost 30% of their recent code was AI generated. It was kinda wake up call for me. So, when I tried it out and I was actually able to plug in AI generated code into a project and do stuff in like less than 1/4th the usual time. It's just amazing. And all this with currently available AI.He's enthusiastically misleading and comes across more and more as a once smart engineer talking marketing fluff and taking people for duds. He has been blowing the trumpet of how AI will replace programming and as if everything will be robotic or done by bots, and that people will gleefully engage with "agentic" robots as if they're real people.
Ofc this AI AI AI from Lisa too has become sour to the ear, and I hope this AI bubble dies down soon enough.
Yep no body should be upgrading from RTX 40 series.
This is Turing all over again.
He's enthusiastically misleading and comes across more and more as a once smart engineer talking marketing fluff and taking people for duds. He has been blowing the trumpet of how AI will replace programming and as if everything will be robotic or done by bots, and that people will gleefully engage with "agentic" robots as if they're real people.
Ofc this AI AI AI from Lisa too has become sour to the ear, and I hope this AI bubble dies down soon enough.
That's...not good. No wonder the prices were lower than people were expecting, Blackwell is underwhelming.
I thought so too. Transformers are highly efficient. So, there is a possibility that the extra CUDA & Tensor cores should compensate any extra delays, if any. Just a guess.There might be some performance loss with the DLSS 4 transformer model running on the 5080 vs the CNN model on the 4080
I do use it too, but it also depends on the language I'm using and is mainly an accelerator of getting to the juicy stuff and getting boilerplate code out of the way along with parts of the code where the algorithmic complexity is quite straightforward. It has to be my way of doing things in the end.Actually, it's started already. Google said that almost 30% of their recent code was AI generated. It was kinda wake up call for me. So, when I tried it out and I was actually able to plug in AI generated code into a project and do stuff in like less than 1/4th the usual time. It's just amazing. And all this with currently available AI.
Imagine what the future holds. I think in the future we should be able to generate entire projects with good prompting.
Blender and 3D apps that use RT cores will have bigger gains than raster game performance definitely.RT cores doubled ray intersection...so perhaps the perf increase in octane render will be bigger than 20 percent compared to 4090 after all (as the difference in core number would suggest).
They do for sure, since they are generally not bottlenecked by other things. 4090 had 1,6x more cores and performed twice as fast as 3090. Then again it significantly sped up the clocks, which is not gonna happen now. So it can very well be, that with 30 percent more cores and same-ish clocks we will see only 30 percent more performance, even in those apps. Unless those faster RT cores improve the situation, or possibly faster VRAM.Blender and 3D apps that use RT cores will have bigger gains than raster game performance definitely.
Bandwidth helps a lot in BlenderThey do for sure, since they are generally not bottlenecked by other things. 4090 had 1,6x more cores and performed twice as fast as 3090. Then again it significantly sped up the clocks, which is not gonna happen now. So it can very well be, that with 30 percent more cores and same-ish clocks we will see only 30 percent more performance, even in those apps. Unless those faster RT cores improve the situation, or possibly faster VRAM.
Is this RT "IPC" or already taking into account increase RT cores in 5090, like 50% I think?RT cores doubled ray intersection
This is pretty terrible. Pretty much the right half shows blur, and it's a screenshot of a youtube render.Seems so, on nvidia slides 4k DLSS perf mode
Not a good excuse, since they are using that as a marketing comparison, basically "Look at this amazing tech we have!".Thats a lot of smearing! Looks like we should stick to 2X FG. Are you sure they're using release quality drivers or dev builds or beta?
Jensen:... Now you combine that with Moore's Law really slowing down and how deceptive marketing practices are becoming, and frame generation is basically the accepted standard until there's enough pushback against it. ...
Best Buy has some laptops... the cheapest comes with the 8940HX (Dragon Range Refresh) and 5070 Ti and is $1899. The 5090 models start at $3999.
Probably just a matter of time.Can you adjust and disable DLSS sharpening as well?