will we ever get 10ghz cpu's?
>>108344403Who cares.The problem now is the software.
they have 30GHz processors in space where it is like 1 Kelvin?
>>108344447It's not that cold in space. Temps swing wildly when exposed to the sun or radiating bodies. Manned aircraft need AC systems to stay cool when in sun light.
>>108344447nihger what? how did they get them up there
>>108344447Thats not how space works lol, harder to cool in a vaccum than on the ground
I already have a 62ghz cpu 5ghz * 6 + 4ghz * 8 = 62ghz
Because of the infinite parallelism of the transformer architecture it doesn't really matter anymore as GPUs just keep getting more "cores" every die shrink. I shared this concern up until 2022.I genuinely think the CPU paradigm is going to end in general. Computers won't have CPUs in 2050. Instead it will use a substrate that completely bypasses the von neumann bottleneck. Using analog processes as much as possible with photonic data lines.This assumes we don't develop reversible computing in time.Once we have transformer based compilers it's essentially over for CPUs. Imagine Claude code but it just directly writes binary code instead of human readable code. And that binary is written as parallel as possible. Not to directly do what you asked for but to do what you asked for as close as possible while still being 100% parallel.Amdahls law assumes a serialization bottleneck but that doesn't have to be the case.Oh and by the way it's possible to have terrahertz CPUs if we were to make it from graphene, but because of the high band gap we won't be using transistors but another quality of superconductors which is spin direction to signify bits. But I don't think we will go this direction because of transformer architecture. CPUs are just not as relevant anymore and will probably disappear as more and more of it's purpose is delegated to specific analog circuitry.
>>108344501NEEEEEERRRRRRRRRDDD
>>108344501>CPUs are just not as relevant anymore and will probably disappear as more and more of it's purpose is delegated to specific analog circuitrya-anon... are you smoking crack?
>>108344578His entire post reeks of slop, I don't think he's high, I think he's retarded.
>>108344403Any kind of singlethreaded workload would see benefit
>>108344585>>108344578Not slop and also not retarded. Computer science degree with computer engineering masters. I'm just writing from mobile phone and thus don't want to elaborate too much.Dennard scaling stopped being a thing for silicon based transistors. The same wouldn't be true for superconductor based computing such as graphene where you would use spintronics to signify bits.I spoke about different concepts so if you have a problem with any of it formalize your argument or at the very least point out the specifics so I can elaborate.CPUs will stop existing with that I mean the architecture of registers feeding into ALUs will stop existing. Mostly because we want to bypass the von neumann bottleneck of separating data from calculation which is the biggest bottleneck in computing right now. We're memory bound and the gap is widening as every computer scientist knows. Analog computing inherently is embedding the data in the calculation itself.Essentially what I'm saying is that CPUs as we know them now are computing dead ends.But all of what I'm saying goes into the trash the moment reversible computing is viable because the architecture then would be truly different on a fundamental level.
>>108344403>AnandtechNo surprise that a site named after a jeet pushed some of the worst disguised ads as news
i already do
>>108344403yes but not electron based light pcs go up to THz magnonics achieve THz easily as well and are limited by laser tech atm.
>>108344403Transistor couldn't shrink smaller than 180nm so all TSMC does is doing tetris autism with them.
>>108344501Even if making a more analog-leaning circuit that could replace current CPUs was theoretically possible, designing it will be unfeasible for a long time unless you hire an ammount of analog circuit designers multiple times larger than the entire field currently has. The yields would be absolute shit as well. The whole reason the abstraction to digital electronics was made is that you can design and manufacture huge circuits easily that way.
>>108344688We're currently seeing a massive move to more analog circuits. Primarily because it would make AI inference 5 OOMs cheaper.
>>108344634>which is the biggest bottleneck in computing right nowLol no. The biggest bottleneck are the kernels.
>>108344634>Computer science degree with computer engineering masterswe can tellcoming to 4chan to talk like a white paper and use all the big words you know doesnt make you look smart it just makes you look like a pretentious fuckwit that doesnt actually do anything besides read academic slop
>>10834440310ghz cpu's what?money? snacks?
>>108344403as cliche as it sounds, can't exactly say, never. but with current understandings and technology, with our known physics, extremely unlikely.but, back in 2005 after they gave up hitting 10ghz, they didn't even think 5 or 6ghz was possible. but, here we are today, with mainstream processors driving around in the 5ghz zone and hitting 6ghz with boosts.good video: https://www.youtube.com/watch?v=5JWcI_xutuI
>>108348202Yeah, and it's completely fucking asinine, because most software and workloads are memory-bound, not compute-bound.
>>108347018Man I'm just autistic give me a break. I speak like this in real life as well, always have. Nothing to do with looking smart. I miss when this place was for fellow autists and not normalfags that think discussing a subject indepth is "trying to look smart".
>>108350093Sure, let's talk about indepth subjects, like how kernel interfaces are still completely fucking stuck in 1970 because that's what the NT/Linux autists copied in 1989/1991 when they designed their kernels, and how we're effectively running hybrid kernels with microkernel interfaces larping as monoliths.Like, who gives a fuck about 5 or 10 GHz machines when the kernels are still clusterfucks, and any and all attempts to alleviate the situation leads to axwounds like io_uring that don't even support batched mmaps.>despite every fucking strace revealing how much of a boon that would be
>>108350221Or, specifically for the autists with CS degrees, how they seem to be unhealthily married to the concept of microkernels despite them making no sense whatsoever. They will come up with the most ridiculous benchmarks (of course synthetic and all warmed up) and excuses to tell people why adding mode switches, increased TLB pressure, and stunted OoO execution isn't that big of a deal for their ideal (autistic) model of how kernels *should* behave, dammit!
