Processors

[erdalb]

Hello,

This is an idea from an ex computer engineer who is not at all experienced with latest processor design technologies. I still wanted to share this idea that came up during a quiet moment and found here as the place to share it.

So we have processors with cores right? And each core generally runs at relatively slower frequencies than a single core processor of older times. I see cores with 2.something Ghz in online usable virtual machines but we had greater frequency processors many years back. So lets say I need the higher frequency for a single threaded app.. but these slower cores can't give it to my app.

The idea which needs an expert to flesh out the details is to put in the practically fastest clock capable chip with the sole purpose of just orchestrating the assigning of start times of instructions for many much slower cores. So lets say I am capable of creating an 8 Ghz clock chip. I use this to use 4 2Ghz cores designed to mimic the workings of a single processor by working together and processing instructions in a staggered way. The 8Ghz chip only makes possible the action of creating the accurate start timing of each instruction given to a core. Since it has 8Ghz frequency capability, if we write software for cores that can utilize them to behave as if they are a single processor even though they are separately working together, then we will be able to use 4 cores and this 8Ghz clock chip to have an equivalent of 8 Ghz processor at hand. This can enable many single threaded applications to run with much greater performance. Since the only task for the 8Ghz clock chip will be to give the staggered start signals to the slower cores to work overlapping eachother, as if like a pipeline.. but since they are individually slower, they need the faster chip that has greater resolution to provide just the starting time for each granular instruction, as if this is happening in a single processor, it won't  heat that much I presume with this small task load. So we can cram in a very fast orchestrator clock chip to combine the working power of slow cores to appear as a single fast processor. 

Anyway.. this idea might be even nonsensical since I am foreign to the inner workings of modern cpus. Still.. wanted to share it and send Much Love. Maybe I can play Dwarf Fortress in a much faster environment one day  

[elstaci]

I believe this might be better suited at AMD Forum's Developer from here: https://community.amd.com/t5/newcomers-start-here/bd-p/newcomer-forum

[gnif]

If I am understanding you correctly this would not provide a perceivable benefit. Each core would still be clocked at 4GHz and process their operands at 4GHz. Since a single threaded process has to run in a linear fashion (ignoring speculative execution), the other CPU core would just be idling while waiting for the first CPU core to finish.

Please note that while CPU clock speeds have not gone up substantially as they used to, the IPC (instructions per clock has gone up substantially). Combine that with other techniques such pipe-lining and new instruction sets like AVX512, we are getting the performance of a higher clocked CPU core without needing to increase the clock rate.

Increasing the clock rate is only yielding diminishing gains while substantially increasing the power consumption and temperatures due to gate leakage and other factors. It's better to be smart about execution then just be at a higher clock rate.