And, as with anything, faster means hotter
I think we need something faster that doesn't produce so much heat. Anyone got a revolutionary idea handy?
Early details of the specifications for PCIe 7.0 are out, and it's expected to deliver data rates of up to 512 GBps bi-directionally for data-intensive applications such as 800G Ethernet. The announcement from the The Peripheral Component Interconnect Special Interest Group (PCI SIG) was made to coincide with its Developers …
Photonics? Optical transistors are a thing; so you don't even have to convert to/from electrical and photonic systems.
An optical CPU doesn't have huge advantages over a silicon CPU in terms of performance or dimensions; but it might be able to win on heat output. Optical is clearly better when signals need to go around a circuit board than copper.
The physical limits of copper; as well as the cost of the raw materials are a factor too that might help justify a completely different physical architecture.
I am not even remotely qualified to understand optical transistors, but can't help but think two things..
1) Doesn't an optical signal have to be converted back to a digital signal at some point?
2) A PCB is full of corners and bends - does an optical signal *actually* improve latency and bandwidth over the same trajectory?
1) Working optical transistors means you have working Optical RAM and CPU; and optical interconnects. Permanent storage would need a photon/electric conversion. Things like a BIOS ROM might need some silicon still during boot. Driving human I/O; monitors; speakers, keyboard etc would need some electronic elements too.
2) If you can build the major components of the system optically, then the latency can theoretically be reduced; and risk of cross talk between parallel signals is avoided.
If you were really sneaky, you might even be able to run multiple wavelengths down the same pathways to cram more data in parallel.
Optical computers are of course just fancy R&D projects in present form, but they do offer an alternative beyond silicon / Gallium Arsenide with copper traces.
With the limitations on track length, complexity and quality requirements of the Mobos will increase costs big time. The fact that Graphics cards barely need PCI 4, seems to me that it is unlikely to be needed in consumer PCs by 2025, servers with dense storage and the like could fulfill a need if Nand could be driven faster or another technology comes along. Main benefit seems to me to be the bandwidth which means you need less lanes, storage devices might on need 1 lane rather then 2 or 4, even Graphics cards may only need 4 lanes rather than the 16 now, however, I can't see the mobo manufactures doing the sensible thing if they can increase prices and margins.
So long as they're not enforcing a new physical connector then you can plug a PCIe 6 or 7 device into a lower spec mobo and vice-versa. The hardware just has to negotiate as per the highest speeds achievable with both bits of kit and go from there.
If you don't need all the speed, then don't get the latest and greatest. Just wait until price reduction kicks in over time and then invest.
I could see a possibility of crypto miners having a field day with this (faster data rates when splitting the lanes etc) but I couldn't care less for them so.... Yeah.