KeplerBoy

I think that is what paseo is doing and I don't risk trying it.

They have an auto approve mode. At least in the vs code plugin.

Entirely possible and poor visibility is not an issue if you have radar.

different billing quotas. Although that may of course change at any moment.

That's just part of the output along the SVG/image.

Isn't Omarchy very, very similar to cachyos?

Lei Mao's blog is such an amazing resource for GPU performance engineering. I am stunned by the sheer amount of insight he puts out on his blog.

So the answer is yes? Putting in a few of those racks for special tasks shouldn't break the power assumptions of a data center.

Like every time

That's where the alfalfa ends up.

Meh, there are plenty of reasons to use one. Might be injury, old age or just keeping up with ones spouse.

That is in the training data. Confidently and correctly answering in-distribution questions (possiibly with a tool call) is expected by now.

Nobody who rides a road bike would be seen dead with something like this. There are proper road ebikes for that.

melious.ai comes to mind.

Which is not unreasonable. Just hosting it in the EU and promising not to retain / sell the data let's you charge a healthy extra and compete in many areas other players can't.

It's probably a version of this: https://dl.acm.org/doi/abs/10.1145/3306346.3323024

Subtle movement between shots can be used to create an image sharper than any of the inputs.

I wouldn't call one proper CUDA and the other one some dumbed down version. Nvidia really seems to be pushing for these DSLs to be first class within the ecosystem. In some cases probably even more cutting edge than the nvcc frontend, since it's easier to do some experimenting on a new niche package than on the tool everyone relies on.

I believe more and more production code is running kernels which didn't originate from the traditional cuda cpp route.

To be fair the entry barrier got a lot lower over the past ~5 years. Now you can write very good CUDA kernels with just a few lines of python DSL code. Zero cpp boilerplate and zero explicit compiler calls.

Stuff like Triton, nvidia warp (the language), numba, cupy jax/pallas and so many others really paved the way. You can start out really high-level, run a profiler and then dive deep into the bottlenecks.

TL,DR: Keep going, it's a great time to have fun with GPUs.

That's shifting a bit with electric cars I feel, because people are worried about used batteries. But yes, generally new cars are often just a bad deal.

You are not European, are you? I know plenty of working class people working ordinary skilled labor jobs who own their homes and drive decent cars. Some people choose to keep renting because we have strong laws guaranteeing renter's rights. Contracts are often unlimited duration with reasonable prices.

Also retiring early is not about the money in your bank account around here. You retire early by using certain government schemes where you might qualify for early retirement by working a physical job, or by reducing your hours to 10 hours are week while getting a full subsidized salary.

That doesn't clarify it and AMD says as much "Because these estimates rely on published results, internal measurements and projection-based scaling factors, they are intended to provide directional comparison rather than direct measured rack benchmarks."

It's marketing after all and nobody should make buying decisions based on that.

AMDs marketing is just as sketchy. Their new Zen6 server page claims 3.3x performance per watt over vera on "agentic workloads" for a 100 kW rack. Maybe they are comparing a CPU heavy rack to an nvidia vera rack with 50 kW of GPUs sitting idle, who knows?

https://www.amd.com/en/products/processors/server/epyc/9006-...

sounds like plain old prompt injection. These days chatgpt might look at 50 webpages when i ask it to research a topic. Seems quite possible that the final answer is influenced by such ads.

You don't an FPGA if you're taping out your own chips. But that is just a MMA accelerator with decent memory bandwidth. No secret sauce here.

Who cares about a few ms more latency on an LLM API? Maybe for voice, but most other use-cases are quite latency insensitive.

Open weight models tell a different story. Inference is not that much more expensive than what a 100$ plan would allow and will only get cheaper (for current capability models of course, frontier not so much).

you are seeing single realizations of non-deterministic processes. Confidently stating one model is better than the other is not really possible this way; that's just like stating one dice is better than the other because you rolled a six with that one on the first try.

Is that really used professionally?

Sure, the point is that carbon credits are issued for a certain area of land to be and stay forested.

Then the area should have to be replanted without granting new credits. It's perfectly natural that trees die and are replaced by new ones in a forest. The unusual thing is just the entire forest dying at the same time.