Why did CPUs stop progressing?
not needed for AI
>64 bitWhat larger pointer size needed? What does AVX10 fall short of?
>>110274886Diminishing returns
>>110274886Thermodynamicswe have reached the end of silicon scaling
>>110274886They're still progressing. It's just that none of us plebs are allowed to buy, and even if we were allowed we wouldn't be able to afford to maintain, quantum CPU's.Give it around 20 years, intelligence agencies are already preparing to the leap with store now decrypt later policies for when the tech becomes more widely available.
>>110274886Physics. Scientists used to be able to make atoms and molecules smaller but now they've hit the limit
>>110274935then why is software getting slower still
You can make 10GHz CPUs but you need liquid nitrogen to cool them.
>>110274935optical computers is happening
all wrongjust wait for Zen 6
Common desktop CPUs go up to 512 bits.
single core ipc keeps improving anon
>>110275195No, not really. It depends on the complexity of the hardware, you can have a 10GHZ CPU if you simplify the design.
>>110274886they dont want us having too much compute power. they are throttling us
The human eye can't see above 32 bits anyway
>>110274886The real answer you're looking for is timing. Every CPU has an internal clock by which each cycle is coordinated. In standard signal engineering, this basically requires that the clock have a refresh rate twice its output frequency. We have solutions for most frequency groups - there are holes but the physics is done. But we haven't engineered methods of getting those hypothetical clocks into the manufacture process for cheap. This is why we're exploring other substrates, like diamond - because alternative manufacture processes might unlock new clocks in addition to their other properties (like heat resistance.)Still, there isn't a ton of interest from industry leaders, because the efficient use-case for any processor is parallelization in datacenters. 100 cheap processors are WAY better than 50 expensive ones. And the same is true on the chip itself - lots of cores are way more useful than a single threaded hyperbeast. If you could have a 32-core 4Ghz or a 1-core 120Ghz, you're still better off with the former except in very specific circumstances.
>>110275185jeets with access to claude code
>>110276487truth nuke
>>110274886Use case?
>>110276408I'm trying to do it with the RISA architecture (RISA-1).
>>110274949>quantumlol
>>110274886They didn't.https://www.intel.com/content/www/us/en/developer/articles/technical/advanced-performance-extensions-apx.html
Giga genius speaking. There is no purpose for >64 bit mathematically proved
>>110274949Quantum does not work my dude
>>110274886You can only cut silicone so fine. The transistors are already less than a hundred atoms across.
>>110274886the transition between 32 to 64 was painful there are still holdouts refusing to upgrade mainly cause they're poorimo we should go 128 just to finally drop compatibility with all the legacy shit
>>110274886legacy bloat, and hitting the physical limitations of physics, silicon and die shrinkage.We're pretty much at the limits of how small we can make things and how much power and heat they can take before electron leakage and damage occurs.Until a new architecture or way of designing is created, i think processors have peaked.
>>110277424>imo we should go 128 just to finally drop compatibility with all the legacy shitCouldn't we emulate all of that like Rosetta or whatever the Microsoft equivalent is?
>>110276503thanks chatgpt
>>110277446nothing worth preserving , a clean slate is much more prefferable
>>110275185they were always getting slower but was outpaced by hardware progress, now the bloat and slop have caught up to the hardware
>>110277424128bit won't even happen this century. I'm not joking. 64bit supports up to 16Exabytes of RAM. the RAM increase from 2000 to 2026 is TINY compared to how much there is left. with how slow RAM increase has become in the past decade. I doubt we'll even reach 1TB by 2100
Isn't AMD64 already so overkill for most tasks that the CPUs can't even address 2^64 memory? The data bus is the full 64-bits but the memory bus is only like 58-bits or something IIRC.
>>110277425you have been saying this for a decade already.
