121–129 of 129 posts
I hope this AI craze will crash soon enough. Maybe then various things normalize in price again. And consumers get cheaper products with less limitations.
Hopefully this will put pressure on the market to produce much more efficient AI models. As opposed to bigger, then bigger, and then even BIGGER models (which is the current trend). FYI: gpt-oss:120b is better at coding (in benchmarks and my own anecdotal testing) than gpt5-mini. More importantly, it's so much faster too. We need more of this kind of optimization. Note that gpt5-mini is estimated to be around ~150 bi…
Manufacturers learned a valuable lesson a few years ago: overproduction leads to lower prices. Samsung was the first to address this issue by scaling back, and other manufacturers soon followed suit (collusion, cough cough). The past couple of years have been extremely profitable for the entire industry, and they’re not about to increase production and risk hurting their profits. I suspect they would rather face shor…
And on top of that fear, you have the pure technical reality: you can't just flip a switch and start pumping out wildly complex HBM instead of mass-market DDR5. That's like trying to retool a Toyota factory to build Bugattis overnight. So you get this perfect storm: a massive, near-vertical demand spike hits an industry that's both terrified of risk and physically incapable of moving fast. So yeah, they're absolutely milking the situation for all it's worth. But it's happening less because they're master villains and more because they're both scared and incredibly slow
Earlier quoted context omitted.
None of this is rent-seeking.
Retail store literally had to pay rent to a landlord. How’s that not a rent seeking business?
Earlier quoted context omitted.
In the UK I was looking at DDR4-3200 SODIMM last week for some mini-pcs... and decided to pass after looking at the price graphs. It's spiked in the last few weeks.
What graph you used for UK-specific prices as it seems the earlier graphs referenced here are US-only?
There is a bit of annoyance as items come in and out of stock (ie. out of stock often means inaccurate price); so its often better to find a product on amazon and look here.
8GB SODIMM stick https://uk.camelcamelcamel.com/product/B0C7Z4HJ8L?context=se...
regular 2x16GB pair https://uk.camelcamelcamel.com/product/B07RW6Z692?context=se...
I have a script watching on some items on overclockers and crucial, including RAM. So for those by graph I really meant "eyeballed an email search".
Earlier quoted context omitted.
lower prices are ok if they are selling more units, the question is whether the price point * units is Pareto optimal overproduction means unsold units which is very bad, you pay a cost for every unsold unit underproduction means internal processes are strained, customers are angry, but a higher price per a unit... can you increase the price by more than you are underproducing?
If we want to engage with game theory here, I would argue that overproduction is a much safer bet than underproduction from the perspective of Samsung, et. al. Underproduction brings additional caveats that manifest as existential risks. For example, encouraging your customers to move to entirely different technologies or paradigms that completely obviate the need for your product in the first place. If you leave a b…
This shortage is the best thing that could have happened for R&D in model efficiency. The "who has more parameters" race is about to hit the physical wall of hardware availability. Now the real race for efficiency begins: quantization, distillation, Mixture-of-Experts, new architectures Hardware constraints are the single biggest driver of software innovation
There are a huge number of varients of attention schemes, like seemingly thousands. More posted on arxiv and blogs every single day. There are similarly a huge number of papers and talks on quantization, codebooks, number formats, everything. Like genuinely covering everything from analog compute to training lookup tables in FPGAs.
The AI model architects similarly are not sleeping on things. They genuinely take a bottom up approach to the design of their model and make sure that every mm of die area is being used.
And top down of course has extreme influence with people spinning up entire hardware companies like groq and cerebras to do AI as fast and efficiently as possible. Everyone wants to be the shovel seller.
The idea that anyone is sleeping on increasing inference efficiency when $$$$$$$$$ are being spent is ridiculous.
And genuinely they have actually made massive strides. Some models cost almost nothing like GPT OSS 120B which can be used to create slop at the speed of light.
Earlier quoted context omitted.
It might be subjective, but doesn't this count at least partially as a currency on the gold standard? https://en.wikipedia.org/wiki/Zimbabwean_ZiG
From wikipedia: >A gold standard is a monetary system in which the standard economic unit of account is based on a fixed quantity of gold. and >The Zimbabwe Gold (ZiG; code: ZWG)[3] is the official currency of Zimbabwe since 8 April 2024,[2] backed by US$900 million worth of hard assets: foreign currencies, gold, and other precious metals. >... >Although the rate of devaluation of the ZiG may vary,[13] the ZiG has co…
Earlier quoted context omitted.
I can add Spain price trends to PCPartPicker. Quick question though - do you want the price trends to cover just Spanish retailers, or should it trend the prices across all of the EU?
That would be incredible! Personally I only buy the stuff I can find inside the country, inside the country. But then some stuff I have to order from Germany/France/Italy when it's only available outside our borders. So I don't know the right approach here, I can see value for both price trends for multiple reasons, unfortunately :) Wish I could give a simpler answer!