After ASUS, here comes Gigabyte. Another Kaby Lake overclocking guide:
https://www.joomag.com/magazine/gig...series-overclocking-guide/0767815001483933769
https://www.joomag.com/magazine/gig...series-overclocking-guide/0767815001483933769
It is hard to have exact idea, cache overhauls are generally pretty messy. Though on napkin math, it should perform very slightly better than Skylake-S, but actually save on cache die space.Does anyone have a vague idea on how much the Skylake X with his L1 1MB will improve performance?
Does anyone have a vague idea on how much the Skylake X with his L1 1MB will improve performance?
Well, SKL-X is not actually listed on that Wiki page.You seem to have some incorrect information on SKL-X. L1 cache will be 64 MiB/core, see: https://en.wikipedia.org/wiki/Skylake_(CPU)
My bad. 1MB of L1 - holy smokes, that's nuts! Dog slow cache unless Intel really came up with something special.Well, SKL-X is not actually listed on that Wiki page.
I believe the grapevine has it as 1Mb of L2 cache per core for SKL-X, not L1. Previous chips had 256Kb.
Bit I don't think it is confirmed.
Does anyone have a vague idea on how much the Skylake X with his L1 1MB will improve performance?
I want to upgrade from current i7 2600k running at 4.8ghz to a 6/12 threads...Ryzen seems to have IB/Haswell IPC and that doesnt excite me to much...so my only choice would be Skylake X or wait another year...
104 threads, sheesh. They really are trying to bury POWER9.
It was obvious Power 9 had no long term future once they started pushing Open POWER.
Marginal at best. These tests were done using 2c2t on same generation of Intel but with different L3 side. While L2 is a bit faster, overall gain will be similar as these cases add more amount.
Marginal at best. These tests were done using 2c2t on same generation of Intel but with different L3 side. While L2 is a bit faster, overall gain will be similar as these cases add more amount.
???L2 a bit faster? Did you test it or how can you say this? Intel changed the cache structure, you have no clue how it performs based on a current available Intel CPU. Also when you say L2 cache is a bit faster, you are way off even based on an available CPU.
https://forums.aida64.com/topic/2864-i7-5775c-l4-cache-performance/
L2 Read a doubling, Copy 66% faster, latency 4x faster.
???
Lol. L2 may be faster, but the moment you increase the size of L2, its latency increases more drastically, so yes, gain will be marginal at best. go read the reviews of Athlon64 CPUs with different cache sizes. its "old" but the principle stays the same.
yes i agree, that if you lower system wide clock, you can indeed have 1MB L2 cache as fast. it will save some power, i think as well. But will the increased cache size make up for the loss of raw clock, especially for application that donot overly depend on cache? I cannot say.Intel said that Server core will be different from Desktop/Mobile, maybe they target 4Ghz multi core instead of 5ghz quad and can pump the savings into 1MB L2 as fast as their 256KB caches.
???
Lol. L2 may be faster, but the moment you increase the size of L2, its latency increases more drastically, so yes, gain will be marginal at best.
Intel is testing out Skylake-X, and beats out current 6950X with 8C because it can hit higher clocks.
We need to remember that many-core server parts tend to benefit from process improvements just as much as mobile parts, and SKL + KBL brought quite a performance uplift over BDW in the mobile space. Looking at corresponding parts - BDW 5600U, SKL 6600U and KBL 7600U - turbo for MT loads went from 3.1 Ghz with BDW to 3.9 Ghz for KBL. That's a 25% frequency advantage from power savings alone (assuming they kept same power usage criteria for turbo). 10 core vs. 8 core is also a 25% advantageInteresting rumor about Skylake-X here. Apparently the 8C/16T SKU (or one of them) is able to beat 10C/20T Core i7-6950X. Better IPC thanks to Skylake core with new cache structure + higher clocks.
That's a 25% frequency advantage from power savings alone (assuming they kept same power usage criteria for turbo). 10 core vs. 8 core is also a 25% advantage