Technical question about clock speed increases

[sid]

It seems like about 5 years now as we haven't really progressed past 3ghz signigicantly.

Thinking about it I realised that 3ghz equates to a cycle duration of about 333ps. During such a short time, the distance that light travels is about 10cms.

To put this into comparison the ATX board is about 30 cms long, so my point is

Are we we likely to see significant increases in clock speeds in the future? as shorter cycles mean more time wasted waiting for data to return?

Does anyone know how many cycles of current cpus are actually used to do useful work? I.e. whats the percentage idle time for the current cpus?

I'm interested in a technical response here.

Regards,
sid

 

[Zenduri]

Clock speeds I beleive are capped currently because the temperatures start to get high without good cooling over 3ghz. Thats always been my understanding to why we are at around 3ghz barrier. havent a clue what your going on about though

Take for example anything over 4.5ghz and you need to start looking at phase or LN2 cooling

 

[LegendaryDread]

Agreed, CPUs are moving towards more and more cores rather than higher clocks.

 

[sid]

Clock speeds I beleive are capped currently because the temperatures start to get high without good cooling over 3ghz. Thats always been my understanding to why we are at around 3ghz barrier. havent a clue what your going on about though

Take for example anything over 4.5ghz and you need to start looking at phase or LN2 cooling

No you've got the wrong end of the stick here I think. Its just I think its diminishing returns for the reasons I've said and also that you can probs squeeze better performance by designing a better cpu with more cores rather than something crap but running it high frequency.

Although perhaps you are right and I'm overthinking it a bit....

Hypothetically, an old cpu design like p4 could run stupidly high speeds on a 45nm tech base but you couldn't efficiently use all the extra clock cycles. You'd just be waiting for data yeah because signals simply can't travel fast enough?

sid

 

[Owenb]

No you've got the wrong end of the stick here I think. Its just I think its diminishing returns for the reasons I've said and also that you can probs squeeze better performance by designing a better cpu with more cores rather than something crap but running it high frequency.

Although perhaps you are right and I'm overthinking it a bit....

Hypothetically, an old cpu design like p4 could run stupidly high speeds on a 45nm tech base but you couldn't efficiently use all the extra clock cycles. You'd just be waiting for data yeah because signals simply can't travel fast enough?

sid

No. The problem with putting a P4 on 45nm and clocking it to ~10ghz is that you now still have a very power hungry core but it has much less surface area to dissipate the heat that it generates. That is basically the problem that Intel hit with P4. Their solution was simply to spread the heat out over a greater area, and the easiest way to do that was multi-core CPU's.
That a very simple view of it, but it's about as much as you will get out of me at 00:50!

edit: The amount of heat that a CPU core has to dissipate is very similar to the amount of heat that a nuclear reactor rod generates per square cm

 

[sid]

Well a reduction in transistor size will significantly reduce power consumption, dont forget, so 10ghz may well be doable within reasonable thermal limits.

Also consider low power cpus like atom which could run significantly higher , couldn't they.

Anyways, the heat issue wasn't something I'm interested in. ITs the performance impact due to speed of communication

sid

 

[johnerz]

Heat and electronic migration are the issues for the current technology, we will move to a non silicon based cpu to get past the current blocks.
However regardless of 3GHZ the I7/6/5 cpus are more efficient and out perform 3GHZ P4's with no effort at all

 

[sid]

Heat and electronic migration are the issues for the current technology, we will move to a non silicon based cpu to get past the current blocks.
However regardless of 3GHZ the I7/6/5 cpus are more efficient and out perform 3GHZ P4's with no effort at all

Hmmm I suppose its impossible to run at high frequency without high voltage to reduce the electrons tunnelling which then makes a lot of heat...

 

[Devrij]

I see what you're saying dude. The signal can only travel so fast, so unless we reduce the distance data has to travel (which creates heat problems) we can't use the ridiculous clock speeds we want because the data can't travel back and forth fast enough to make each clock cycle useful. So until we can eliminate the heat problem, and make things even smaller, we're stuck. Bring on quantum processors!

 

[sid]

Exactly, this problem must exist today as shown in the OP so I wonder how it really works

sid

 

[Corasik]

Motherboards are designed with this in mind though, the memory is mounted right next to the CPU socket, not at the opposite end of the board, so in practice its just a few cm long (even once you consider the path the signal tracks are layed out). Further more the CPU works with a large amount of Cache so a lot of the time programs run with considerable portions in cache, which is onboard the cpu just millimeters from the cpu cores.

Finally for bulk data transfer its true that distance and speed of light (well speed of electricity in a circuit) do affect latency, CPU's dont use serial transfer they use parallel. In recent years we've gone from single channel 64bit wide busses, through dual channel, and now with i7 tripple channel, with each step increasing the amount of data that can be transfered, and as most of the time the CPU will prefetch data before its requred a minimal amount of latency between ram and cache is not a major consideration.

At the moment the dual channel i7 is very close in performance to the tripple channel i7 which indicates that there is still considerable room for CPU improvement before memory bandwidth really limits performance. The P4's shockingly long pipeline caused intel to spend a lot of research into prefetching data to the cpu's cache to avoid having the cpu sitting around idle.

IMHO a 10Ghz i7 would not have a memory starvation issue, but it would have serious heat issues within the cpu core itself and use massive amounts of power.
The biggest problems with ramping up clock speeds is how much energy you have to pump into a cpu to get high clock rates. Material technologies are the best hope to solve this.