KeplerBoy

The DGX also contains the 200 GbE networking and linux support.

Not at all. This is a more like what Apple has been doing the past few years. A bunch of decent arm cores paired with a beefy integrated GPU.

It's not that different in Europe depending on the route.

Checked bags are a hassle and cost money.

It is completely different. FPGA tooling is not the same as a driver for a consumer product.

A lot of the serious CUDA compute stuff is also not supported on all platforms (it's linux only, because why would you do such stuff on windows).

And that's probably necessary if you want a competent model. There simply isn't much norwegian literature on let's say banana farming.

How true is this statement: "He asserted that any country with its own language that did not have a sovereign LLM trained in that language was at a disadvantage as a globally trained, English-speaking LLM would not know about that country’s history, news and culture that was described in the local language."

I thought all big players already train on basically everything remotely available to them no matter the language or quality, so his take sounds like an opinion formed in the early days of generally available LLMs.

Interesting fact from the page: "The lowest drag coefficient in Ferrari history, achieved through aero-styling convergence, active air shutters, and ride-height logic that lowers the front by 10 mm even while cruising"

I guess not having large air intakes and generally a slightly larger frontal area helps with that (the coefficient of drag is always multiplied by the area, so this might not be the most aero Ferrari ever, that's a different claim).

Interesting idea, but ultimately not going to happen (or matter). I doubt the latency in that DSP Pipeline is below a millisecond, heck given the state of non-critical automotive Software it might a second.

This style might have worked as an apple car. It sure as hell doesn't work as a Ferrari.

There are trillions to be made. That moat won't be as insurmountable in hindsight.

They just renamed it to antigravity ide. I don't think that product is deprecated.

Antigravity is just a vs code (more correctly: codeium) skin with Google telemetry and agent Integration. You can switch back to Microsoft's or cursor's flavor in minutes.

Isn't that much better than PayPal? Why would i want my money to end up in some intermediate PayPal account?

They seem to be doing let's say case studies on how AI based simulations can help industry.

They did injection molding for example, and I'm sure they're testing similar approach to everything which can be modeled by PDEs, which is well pretty much everything ever of engineering interest (I'm assuming this is somehow connected to a research project funded by Engel, one of the leading injection molding machine companies, located in the same Region): https://www.emmi.ai/models/neuralmould

Feels like the opposite.

Stuff is still happening and you need to be part of a big lab to see it. NanoGPT is fun but at some point you need that datacenter.

Can you elaborate why you chose slang-splat over let's say Lichtfeld studio? How does it compare to the other splat-training tools?

If you don't focus stack and try to train on partially unfocused images, the optimizer will try to match the rendered view to be also partially unfocused.

You would have to mask out the blurry areas for each image. I guess one could just implement a feature where the optimizer only optimizes gaussians within the sharp distances relative to the camera.

I'm not sure i agree. The blobs are exactly where the surface appear to be because they are constrained by multiple viewing angles.

Otherwise the splat would fall apart as soon as the viewing angle is changed slightly (Which it absolutely does in many examples on supersplat, you cannot really create an out of distribution view with 3GS, it's not magic)

Thanks I appreciate the answer!

Probably link training.

The very fact that it is a tens-gigabit-per-second connection in an environment full of RF noise (we all love our high powered switching power supplies) makes it hard to get a reliable signal instantly. Still shouldn't take longer than a few hundred ms though (worst case).

I wish there were any good literature on GPU Direct RDMA and GPU NetIO. Got any tips?

That's hardly a guide. It's the defacto documentation, you have to read this either way.

It's just a skin. Of course it uses more resources than windows 7.

Nothing changed. You can still code the old way. All those 100x productivity gains are probably closer to 10% productivity gains after you account for all the added debugging and steering.

It's just higher nyquist zones.

That ship has sailed on the web a long time ago.

You can't have good performance without good DevX. There's a reason why we get a new python dsl for nvidia GPUs every week.

lol, are they doing stochastic invoicing?

Well the cheapest crap fans are almost free, Noctua fans are certainly not free. So the added cost is the entire price of a Noctua fan.

Because someone might argue blender is taking funding from an entity that wants to make proficient blender users obsolete.