Broadwell Won’t be Available to Consumers Until 2H 2014

[SessouXFX]

It all goes back to Moore's Law, folks. You can only shrink the die so much before **POOF!**

22nm is good enough for desktops, to be honest, going lower for the sake of it is more than likely a hassle at this point. I'd leave the 14nm for mobile devices, if they're able to sort it all out.

 

[Weak1ings]

What if instead of going smaller... the Processors got Bigger? (not the architecture but the total size of the CPU with more cores, more power more everything.) Like the GPUs have done by getting bigger, more monstrous, power hungry but more powerful.

 

[InvalidError]

What if instead of going smaller... the Processors got Bigger?

CPU manufacturers want to make devices smaller so they can cram more of them per wafer to spread the costs of wafer processing over more sales and reduce the number of unusable devices due to defects and process variations across the wafer.

If you make dies twice as large, you can make less than half as many per wafer and are twice as likely to get unusable dies due to defects so the costs per viable die more than doubles. On GPUs where a large chunk of surface area is shaders - thousands of them - defects are likely to affect only a subset of them that can probably be shut off so the chip can be sold as a slightly lower-end model. On CPUs, losing a slice of L3 cache or 1-4 cores due to a defect in L2 cache or even just a core accounts for a much larger chunk of total available resources.

Another problem with making CPUs more parallel is that there is almost no mainstream software that makes decent use of more than two cores. A massively parallel CPU would have more complex arbitration logic between cores and get in the way of individual cores' IPC that could translate into worse performance for most software out there. So, making mainstream CPUs so much more parallel than what mainstream software can use is largely pointless and counter-productive.

 

[Weak1ings]

What if instead of going smaller... the Processors got Bigger?

CPU manufacturers want to make devices smaller so they can cram more of them per wafer to spread the costs of wafer processing over more sales and reduce the number of unusable devices due to defects and process variations across the wafer.

What if they returned the idea of multi-processor Motherboards? so I could pack 2 i7s (though they hardly lack as is so far)

If you make dies twice as large, you can make less than half as many per wafer and are twice as likely to get unusable dies due to defects so the costs per viable die more than doubles. On GPUs where a large chunk of surface area is shaders - thousands of them - defects are likely to affect only a subset of them that can probably be shut off so the chip can be sold as a slightly lower-end model. On CPUs, losing a slice of L3 cache or 1-4 cores due to a defect in L2 cache or even just a core accounts for a much larger chunk of total available resources.

Another problem with making CPUs more parallel is that there is almost no mainstream software that makes decent use of more than two cores. A massively parallel CPU would have more complex arbitration logic between cores and get in the way of individual cores' IPC that could translate into worse performance for most software out there. So, making mainstream CPUs so much more parallel than what mainstream software can use is largely pointless and counter-productive.

 

[Draven35]

What if they returned the idea of multi-processor Motherboards? so I could pack 2 i7s (though they hardly lack as is so far)

*points to the Dell T5600 review, et al.

You can, just multiprocessor boards is a feature they reserve for Xeons

 

[SessouXFX]

Haswell-E engineering chip spotted in the wild....

Can't wait. This may very well be one of the last builds I ever do. Waiting til Haswell-E comes out might be wiser than jumping on Broadwell immediately...