Recent content by neckthrough

OK I love everything about BG3. Shadowheart and Astarion are *superb* characters. But iconic? Hells to the no. To me "iconic" means the characters transcend the boundaries of their own games, even genres, and become mascots of gaming in general. Shadowheart and Astarion have not yet earned that...

This perceived risk of "evil intelligence" is just feeding our egotistic desire to play god (perhaps because of how powerless we feel otherwise). Let's not distract ourselves from the real challenges. Explainability. Software complexity. A loss of confidence in any media as evidence of truth...

Could you ELI5 why? Is it a fundamental algorithmic issue? Or an artifact of the current HW/SW interface?

Based on what you described it seems the bisection bandwidth is 2.4TB/s (which is how you should measure aggregate network bandwidth instead of just adding up all the link capacities), unless I'm misunderstanding the topology.

Oh hell no. I'd pay $499. Maybe $549. Anything above that I'll pass and move my gaming PC to the living room. At least I'll get proper RT and upscaling.

Indeed. I've been meaning to buy a PS5 for a while and almost bit the bullet when the slim came out. But I think waiting a few months for the Pro makes sense.

It's not a paradox, it's a well-understood tradeoff. Specialization is more cost-efficient than general purpose if the specialized units are well-utilized, but turns into dark silicon if they are not being properly utilized. I think we all agree on that. Where we're disagreeing is, you insist...

What does the ownership model of the NN weights have to do with fixed-function vs. general-purpose hardware? I'll concede that I was overly generous to optical flow, it's fairly niche. Matrix acceleration is absolutely generic today. This is widely accepted by the broad computer architecture...

We are arguing about fixed-function vs. programmable hardware. You are conflating hardware fixed function blocks with proprietary software algorithms that make use of those boxes. The hardware boxes that ML upscaling uses are generic matrix crunchers. In fact this is one of the best examples of...

No, there is no doubt that fixed function hardware is more efficient. Efficiency isn't the problem. Their "fixed" nature is. So you want to be judicious about what idiom the fixed hardware is accelerating. Correct, we aren't certain. Some things stick. Others don't. Choosing the idiom carefully...

You could drive along the Pacific Coast highway (CA 1) from SF to San Diego. Suggested stops: Visit the aquarium in Monterey, drive to Carmel via 17-mile Drive Watch elephant seals in San Simeon Stroll around Solvang (a Danish settlement/tourist trap) and try the amazing pea soup at Pea Soup...

Not sure what you mean here in terms of complexity. Variable length ISAs are ugly, no question. But the dataflow backend of a modern superscalar CPU works on decoded microinstructions, which are clean. Making it wider is problematic for the wakeup/issue logic because it scales quadratically, but...

At the end of the day it's a tool. It can yield insanely amazing results but it needs an artisan in charge. As an analogy, I do a lot of photo editing in Lightroom and Photoshop and their new AI tools are *amazing*. But they are not 1-click do-magic buttons (well, maybe the NR is). It takes...

Continental DWS are great all-rounders, and will be workable in light snow, especially when new. But nothing beats a set of dedicated snows. Keep them mounted on a cheap set of rims and swapping them is easy peasy.

I don't know where Nvidia's headspace is, but in general corporations *do* understand that consumers are more sentimental than businesses. This is especially true with today's influencer culture that (for better or worse) enjoys serious sway over consumer sentiment. Therefore, even though math...