For a long time, the chip manufacturers had an inclination to simplify the hardware and rely on the software adapting and optimizing. But for decades this bid failed. Now we have the unrelenting AI capable of finetuning kernels relatively quickly. Maybe simpler hw will work this time? Note: not sure if TPU/NPU is not only simple but also too limited.
I have some doubts. For example, there is no software replacement for out-of-order execution: It is unknown before runtime in which cache level the required data will be or which values it will have (which changes the latency of a few instructions).
That was the PS3 SPU. It was very fast for its time, but only for the small subset of code that could work within its constraints, and viciously difficult to program for.
I think Leopold of "situational awareness" fame predicted this, tried to make it widely known on podcasts and x, and still started billions worth of hedge funding based on this.
> I think Leopold of "situational awareness" fame predicted this, tried to make it widely known on podcasts and x, and still started billions worth of hedge funding based on this.
Ironically enough it's because of its bet on RAM companies that all of Situational Awareness' publicly trading stocks got bought by the highest bidder (Citadel) when the (leveraged) shit hit the proverbial fan and Situational Awareness created a hole of tens of billions.
Situational Awareness was not forced to sell its share in Anthropic for, by contract, investors in Situational Awareness have no right to force the sale (before this month of September I think) of shares in companies not publicly trading yet.
Citadel bought those RAM stocks right at the end of July, during the dip. Amazing timing for for example MU bounced back from $746 back to above $1000 (now around $950).
Now Situational Awareness may still do exceptionally well: they've got $5 billion or so left in Anthropic AIUI.
it is probable that you don't get the deepest pockets by wasting money. anyone that pays so much for memory means they expect to get even bigger value from that.
This is a misunderstanding of bubble markets to an extreme degree.
In a bubble market where B has massive amounts of capital allocated to it in a way that will likely collapse in the future than it's very likely they will never recuperate the 2M they put in the RAM. Instead buyer A that would get at least 1M of usefulness if not more, gets far less, crippling their business. Then when B collapses a ton of RAM is dumped on the market for far less than 1M cratering the RAM providers business for years.
I am taking some bromelain and papain capsules which according to some (small) studies should help. In my experience, the floaters became more transparent, but don't disappear. This is still useful i guess.