that's jiggawatts, much different!

"We aren't the same as you."

LOL, share you aren't. But at least you draw a paycheck from Putin.

"But yes it took me a month to accept that MW or GW is ok to use to denote data center capacity. Now every other number from investment in compute or number of people or amount of water is also thumbruled to per MW of IDC capacity."

It took you a month to realize that the metrics billionaires use exist to quantify how rapidly they can gain wealth, not how much technical capacity exists (which they don't even care about)? This is right in the metric itself, you should understand that immediately.

"Just scaling everything pro-rata from from some GW/MW thumbrules is how we have reached this bubble stage TBH. It's like every real estate developer denoting how much MSF Million square feet they have under development regardless of whether it's in NYC or Timbuktu. And people investing based on that."

Yes, and it's even worse than that. Billionaires think that every AI problem gets solved merely through scale, that's why they only care about what limits scale. That's where the GW metric comes from, and they fact that they use is tells us just how incompetent they are.

"1) While the FLOPs per megawatt ratio is going to change over time, the limiting factor right now isn't compute per se, it's the ability to get power the computers. "

Citation please. Obviously, billionaires want free power and are making a point of this, but so what? Furthermore, while watts is a measure of power, it is NOT a measure of inference.

"But the fact that memory bandwidth is a problem doesn't make FLOPS any less obsolete."
FLOPS are not obsolete, nor have FLOPS ever been the sole measure of computing capability. It is one measure, and it's a chosen measure here by people like you for the sole reason of arguing against it. The topic is megawatts, the dumbest possible way of expressing inferencing capability.

"What matters for AI inference is being able to do 50 TFLOPS across tens of gigs of RAM (or hundreds?)."
Which is why FLOPs are not absolute, despite your ignorant claim.

"But we don't have a good unit-of-measure for that. But we do have a perfectly good unit-of-measure of today's limiting factor: gigawatts."
Sure we do, even if you don't, and gigawatts is NOT a measure of today's limiting power, just like horsepower is not a measure of the amount of freeways we have.

"Because it's the primary bottleneck for datacenter capacity."

Citation please. Power is A bottleneck, it's not the PRIMARY bottleneck.

"GPUs are cheap."
They are not.

"Hauling mobile megawatt generators to datacenters because utilities can't provide the power is not."
And this is a false choice. There are more constraints that these two.

"It isn't. But it's more useful than FLOPS."
No it's not, and no matter how many times you repeat it it remains about the most idiotic thing you could say.

"You are so fucking illiterate, lol"
Look who's talking!

"What measurement would you prefer, since FLOPS is literally irrelevant to them?"

Ignoring that FLOPS are extremely relevant, despite your moronic claims to the contrary, we certainly don't want WATTS which are completely meaningless.

Let's not forget that as stupid as your comments are regarding FLOPs, you make them in defense of something even more stupid.

A FLOP has just as much meaning as a watt, yet is far more relevant to computation. You can say what you want, but all you're telling us is how stupid you are.

"AI inference has zero dependence on FLOPS. None."

Completely false, and yet you want to claim one is meaningless while the other isn't? How do watts related to inferences?

"FLOPs are a better metric than Watts, though."

Because they are, but where is a single example of where I've used that metric with respect to inferences?

Do you even speak English, you idiot?

Let me guess ... the wells were in "schools" and "hospitals" ... right under the rocket launchers, lol

Well gee; that just totally obviates my point, lol

On a decent web host, that's a single click to install a free Let's Encrypt certificate, lol.

"Redesigns" often make things worse. Designers are given free reign, users come in a distant last.

Incremental improvements are more likely to make things better.

Because Amazon! Because capitalism! Heck, probably because Trump somehow!

Sheesh, get with the program! ;)

Well, that's good - I hate a brash, rude, in your face drone ...

Current approaches to AI may need this, but what that says is they are doing it wrong. An "intelligence" is more than a neural network, no matter how big.

A human figures out what do to, an AI fakes it based on ample precedent it has been trained on. The two are not the same.

Correct, and no one should ever forget that.