Hacker News Reader

Comment by eesmith | Hacker News Reader

You've changed the baseline.

Altman said that people use all of the computer's 32G of memory at most (pulling a number perhaps out of his ass) 5% of the time.

You referred to using a home at least 5% of the time. If I live in only two rooms totally 150 sq ft of my 3,000 sq ft house for 19 days of the year than I use 5% of the house for at least 5% of the year.

I trust you see those two 5%s are very different things.

Consider if I had a 32 GB desktop w 2GB for OS, set up to donate CPU cycles to a distributed computing project. I 100% of the time use the RAM on the desktop, even when I use just over 5% of the RAM.

Note also that Altman said nothing about how often people use less than their entire RAM. If I have 32 GB because I regularly use 28 GB and want some in reserve, then knowing that I only need the extra 4 GB about 5% of the time doesn't mean it's worthwhile to take away 4GB and have my code crash 5% of the time.

View on HN

> You've changed the baseline. Altman said that people use all of the computer's 32G of memory at most

1. No, I think you changed it - I just re-listened to that section, and I don't hear any mention of "at most", nor any mention of "all of the computer's" DRAM, nor any mention of "32G" - what are the timestamps you are referring to? The only other piece he mentioned was that he was speaking in terms of "a global allocation perspective".

> Consider if I had a 32 GB desktop w 2GB for OS, set up to donate CPU cycles to a distributed computing project

2. The scenario you laid out seems incredibly uncommon - are you actually using a 32GB PC w/ 2 GB for the desktop and donating the rest to a distributed computing project, or are even you not doing the scenario you laid out? Are you commenting on HN using a ≤2 GB RAM OS w/ the rest donated? Regardless, the 2GB desktop out of 32GB + donated scenario seems like an incredibly uncommon one that is not at all representative of the norm. I'd guess a far more common scenario than that is a device that was bought but only got a little use before being abandoned, etc.

I said 32GB because that's what's on my desktop, and because my previous comment, in a parallel thread, was https://news.ycombinator.com/item?id=50324468 where onion2k proposed that Altman was right because "Having 16GB or 32GB as standard on a PC when Apple's 8GB M1 demonstrated you can do a huge amount of work with much less proves it."

I chose 32 GB as the high-end of that range, so I could have something concrete.

We are indeed talking the same quote: “But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used 5% of the time,”

I'm fine with having misinterpreted it, but, what does that 5% number mean for global DRAM allocation, how is it determined, and how do you apply the same measure to extra homes?

I use my bicycle far less than 5% of the time, so what does that say about bicycles from a global allocation perspective, and why should that change anything?

> The scenario you laid out seems incredibly uncommon

I was trying to demonstrate how there are at least two very different ways to define what '5% of the time' means. The inspiration for that was a friend who donated processing time to GIMPS, but I don't know how much memory GIMPS needs these days, and the details were beside the point.

I also don't have a 3,000 sq. ft. house.

> I use my bicycle far less than 5% of the time, so what does that say about bicycles from a global allocation perspective, and why should that change anything?

If you're looking for a transportation analogy, I think a far better analogy would be if there was an incredible shortage of rubber for vehicle tires, with possible allocations of each tire including either a bus that would be used ~24x7 for a large number of users simultaneously, or for a car that would be used by 1 person a relatively small percent of the time. And I understand that people would disagree about desired outcomes there as well. Also, what do you mean by 'why should that change anything?' - is your question about why someone might say allocating limited resources to a use that is used by large numbers of users simultaneously 24x7 is different than allocating the resource to something to be used 5% of the time by a single user?

(Also, how would the bike analogy work - what's the shortage of and what's the alternate use of the constrained item would serve a large number of users simultaneously ~24x7 if it wasn't instead being used for your bike for 1 person ~5% of the time in the bike example?)

Since I don't know what the 5% in the RAM statement is supposed to mean, all I can do is use analogies to show why I am confused. Asking me to clarify my confusion doesn't help me understand what Altman is trying to say.

> is your question about why someone might say allocating limited resources to a use that is used by large numbers of users simultaneously 24x7 is different than allocating the resource to something to be used 5% of the time by a single user?

No. The whole point of gruturo's top-level comment which started this thread was that Altman clearly does not apply the same logic to housing, any generic "limited resources" logic cannot be what he is referring to.

Even if it were, I don't know what those numbers are. How many users? Where does the 5% number come from? How does it compare to other forms of allocation (eg, housing, or personal car ownership vs. mass transit) where we've long since decided that an unequal level of access is well within reasonable cultural norms?

As far as I can tell, this is a made up scenario to justify his use of centralized data centers, and it implies a strong centralized planning state should be preferred over a (well-regulated) free market.

In war time we know the government decides where rubber goes, how food is rationed, and can enforce travel restrictions. But this ain't war time, and Altman isn't the War Production Board to decide these matters.

If the question really concerns the ~24x7 use for a large number of users, vs. the limited benefit of a few, then use that centralized planning state to cut CO2 emissions so the next generations of humans will not suffer so much. That's another clue how we know this point isn't really about generic concepts of resource allocations.

Reply on news.ycombinator.com