Some Core i7 4790K "Devil's Canyon" Overclocking On OpenBenchmarking.org
Phoronix: Some Core i7 4790K "Devil's Canyon" Overclocking On OpenBenchmarking.org
Earlier this week I published the first Linux review of the Intel Core i7 4790K "Devil's Canyon" processor that's a refreshed Haswell CPU with 4GHz base frequency and 4.4GHz Turbo...
Did you use the stock cooler to overclock or an aftermarket one?
Looks like both Intel and AMD have hit a wall on clock speed again
If this was with a good cooler-or just with low temps due to moderate total dissipation, this means Intel has not been able to increase clock speeds since they first went to 33nm. The 33NM chips from both Intel and AMD, and all smaller fab chips since then it seems max out between 4.4 and 5GHZ, and rarely over 4.6 GHZ. Yes, AMD managed to create a 5GHZ special Piledriver chip by binning them for clock speed, but Intel can surely do that too.
Originally Posted by mastermikeywwt
Intel has shrunk their dies a couple times since 33nm but it seems cannot get any further clock speed increases. All the way back to Ivy Bridge they've been able to reduce dissipation a bit but not increase clock speed with die shrinks. The only way AMD is losing anything is not by being on larger fab sizes, but rather by using a core design that is a rough match for hyperthreading but requires far more transistors.
Taking this back even further, extreme overclockers using LNG barely got AMD bulldozer or its descendants higher than a Pentium 4 Prescott nearly a decade ago, and those were often sold in the 3GHZ range. Thus the focus ever since on more cores-but then you run into the fact that only certain jobs like video encoding or a few paid games are able to generate that many threads, and fewer yet scale well.
Smartest move Intel made in recent years might have been that internal ring bus that got them 10% more throughput at same clocks and number of threads/cores.
yhe, Intel is hitting a wall there, maybe that is why they decided to integrate a video processor into the package to give more value for the processor and the cores (X99 is going to have 8 cores with hyperthreading), but I wonder how fast a single-core haswell/ivy/sandy could be if there was an unlocked model of those
My motherboard has a feature called "Processor Downcore." I can set the CPU to only use 3/2/1 cores (or 6-the board was meant for Phenom I, not Thuban). Maybe there's a similar one on an Intel board somewhere?
The practical wall for the end user is what happens when you feed it enough voltage to hit the higher clocks: past a certain point, you have electromigration fun, and your CPU's not good for very long. A 5+ GHz part that puts off 300W and is only good for a few months isn't a very good product for most end users. That, and you have to check for single-bit-error gremlins once in a while.
It's like why most GPUs max out at very similar frequencies: you aren't allowed to push the voltage past a certain point because doing so would make them fail much more quickly.