News for hackers. Stuff that matters.

A comment on Intel Kaby Lake Potentially an Overclocker's Dream, Seen Breaking 7GHz with LN2.

Re:Intel Kaby Lake Potentially an Overclocker's Dream, Seen ... by lmm on Wednesday December 28, 2016 @12:05PM (#13271823)
So the single-thread, integer-only, no-branch performance (admittedly a niche, but an interesting one) is still a fair way behind what was possible with the Pentium 4 (which could clock up to 11GHz)? I get that these days performance/watt is what matters and losing 100-odd cycles when you mispredict a branch isn't great, but it's still sad to think we might never match the best-case performance of the older line.
[ Parent | on HN ]
Re:Intel Kaby Lake Potentially an Overclocker's Dream, Seen ... by richardboegli on Wednesday December 28, 2016 @02:22PM (#13272881)
> Pentium 4 (which could clock up to 11GHz)? Linkage?

Only thing I found with Google, was Intel predicting 10Ghz CPU by 2011.

https://encrypted.google.com/search?hl=en&q=pentium%204%2011...

http://www.geek.com/chips/intel-predicts-10ghz-chips-by-2011...

[ Parent | on HN ]
Re:Intel Kaby Lake Potentially an Overclocker's Dream, Seen ... by nickpsecurity on Wednesday December 28, 2016 @12:29PM (#13272019)
"but it's still sad to think we might never match the best-case performance of the older line."

More a positive note to those of us sticking with older hardware in various use-cases for simplicity and fewer backdoors. Good to know the integer performance still rocks. ;)

[ Parent | on HN ]
Re:Intel Kaby Lake Potentially an Overclocker's Dream, Seen ... by tedunangst on Wednesday December 28, 2016 @12:15PM (#13271913)
No improvement in IPC?
[ Parent | on HN ]
Re:Intel Kaby Lake Potentially an Overclocker's Dream, Seen ... by lmm on Friday December 30, 2016 @05:25AM (#13283675)
Not for the single-threaded integer-arithmetic case, AIUI - ADD is 1 IPC (at least when there are data dependencies) and pretty much always will be. In cases where branching happens, modern processes can maybe do more speculative execution, and obviously going from non-SIMD to SIMD is a huge improvement in the cases where a new processor lets you do that, so there are plenty of real-world improvements, but I don't think to the core case. I'm not an expert by any means though.
[ Parent | on HN ]