--- (---.delete@this.redheron.com) on January 28, 2023 10:37 am wrote:
[snip]
> I think that seeing these things through a lens of "ethics" (which usually boils down
> to "I don't want to pay more for better stuff" is not useful or informational.

I do not know about "usually", but I can declare something of my own motivations with, I believe, only modest relevant self-deception. While I am inclined to be cheap and not drawn to ostentation (so I often do not want to pay more and often enough waste money buying cheaper but less durable goods — though this also probably has a risk avoidance aspect as a more expensive good may be more likely to be more durable but there is more uncertainty), I think dislike of waste (in this case, intentionally disabling functionality).

I remember in my twenties (I think) being disturbed that a continuous extrusion machine wasted a visibly large amount of food at start-up. While the fractional waste was probably extremely small and cost-wise the waste was justified, the emotional discomfort was noticeable as was the feeling that we should be able to do better. (Doubtless it actually would be physically possible to do better, but the cost of the effort would not be justified by the reduction in waste. I know I was a larger fraction of tape when making minor book "repairs" — not deciding on an optimal length, not extremely carefully cutting, etc. I might be happier if such was done perfectly — or at least disappointed in some other thing — but falling short of perfection in such does not really bother me.)

I suspect people with "engineering mindsets" are more likely to have a perfectionist bias where they would be bothered by intentionally denting a product so that there would be enough production of irregulars to fit market demand.

Yet I would also think that licensing "extra" functionality that is present would be more attractive to those types than fusing off the functionality.

I do not know if Intel plans to charge more for software upgradable parts. A price premium would make sense (there is at least some small value in being able to upgrade) and such would probably encourage activation of features (sunk cost fallacy: one has already paid for the ability, one might as well use it).

(An interesting — to me at least — variant on this would be feature gambling. One buys a SKU with a chance that features might be enabled. With artificial mark-up on guaranteed-enabled features, this could conceivably increase profit. I am skeptical that such could be made to work for Intel, but gambling has a broad psychological appeal (similar to "free [unspecified] gift included").)

> A more useful lens is whether particular behavior grows the ecosystem.
> *That* is my concern with the details of how Intel is handling this.
> Every CPU that does not make functionality available to a developer is a CPU that is no longer growing
> the Intel ecosystem.

I suspect there are functionalities that are not broadly popular but are of significant value to some and may become valuable under certain circumstances. Buying a software-upgradable part might be similar to buying insurance.

This does discourage tinkering which discovers and potentially popularizes uses for functionality, but most CPU users are not tinkerers of that kind.

[snip]
> If you want to engage in market segmentation, I think the trick is to do it by *capacity*, not by *capability*.

Even in that one has to be careful. AVX512 that is slower than AVX2 would presumably still hinder software adoption. While a single software base has value, tuning complexity would be increased and the average benefit might not justify software development effort.

> Give everyone AVX512 (and BNNI, and the accelerators, and the rest of it) but either
> - at the low end make them work, but not as fast as at the high end. (AVX512 double- or even quad-pumped) OR
> - insert some sort of silly counter so that the user can essentially get fast
> "home use" of AVX512/accelerator for up to X billion instructions per day after
> which they are throttled, plus an On Demand key to end the throttling.
> Both of these grow the ecosystem in a way that simply not having the functionality does not.

Throttling would seem to work well with a pay-per-use model. If one is temporarily doing work that finds the feature useful, one can discover the feature's utility and make a decision to license a time period for the feature. This is kind of a free sample and might be positively marketed as such — bonus trial features. Casual tinkering might not often run into the limits of such a free trial (until either the tinkering becomes interesting enough that paying for the feature feels reasonable or the result is known to make the feature worth the extra cost).

I think giving users the choice to "upgrade" hardware might have some benefits, but I rather suspect that Intel would use such to drive revenue/profit growth rather than increase good will.