6700K OC vs 5820K OC vs 5960x OC

[moonbogg]

I'd still rather have a slightly slower 6 or 8 core chip than the newest mainstream quad. I'm done with the toy platform. Sticking with the big guns. Seriously, an overclocked 8 core Skylake-E will last such a long time and stay strong for pretty much as long as you want it to. I'm headed straight for one when they land. Despite faster quads and faster chips being available, my 3930K is still beast enough to wreck any game or app and look how old it is. The Skylake-E will last even longer.

 

[know of fence]

I would say that it's not the memory bandwidth alone, it's more that Skylakes architecture needs the higher bandwidth to get the best out of it. I read somewhere that it has more registers per core for operations which is how they manage to do more at lower clockspeeds/power consumption so I would speculate that the additional bandwidth lets Skylake breathe.

Do you have a source for that, because if anything the effect is small and incremental. As was discussed in the Does Memory Matter? topic, both HW and SL improve given faster RAM.
Curiosly when it comes to relative gains: slowest architectures improve most in games because of bottleneck reasons. But fast architectures also require fast memory. Going DDR3 1600 to 2133 really gets Sandy Bridge going the most in current games.

 

[Deders]

Do you have a source for that, because if anything the effect is small and incremental. As was discussed in the Does Memory Matter? topic, both HW and SL improve given faster RAM.
Curiosly when it comes to relative gains: slowest architectures improve most in games because of bottleneck reasons. But fast architectures also require fast memory. Going DDR3 1600 to 2133 really gets Sandy Bridge going the most in current games.

That's an interesting chart I hadn't seen before on that page. I think initially when people ran Skylake benchmarks with 2133MHz ram, there wasn't much of an improvement over Haswell, until they were then tested with 3000MHz+ ram where were significant gains to be had in some circumstances. GTAV was a notable example, especially being a game that people would want to invest in hardware for if they want the 60+fps experience. The overall premise of that thread was whether anything much above 3200MHz would give an advantage and I've yet to see any proof that it will, but I digress. Maybe demands for ram bandwidth actually have risen.

I wish I could find where I read about the extra registers. The things that I can reference that have cemented the idea into my mind are things like the much bigger increase in i3 performance this generation, how Skylake laptops are able to deliver great performance at lower clockspeeds/power consumption than previous gens.

I did some testing myself with Sandra and you can clearly see that Intel has worked wonders with the power/performance ratio. One way this could be possible would be extra registers so more data can be handles at once (per core), and more data being handled would benefit from wider memory bandwidth.

X= Load applied over time, Y= Clockspeed needed to perform specified load.


It's the i3 performance increase (vs Haswell) that keeps me coming back to this. It's proportionally a much higher increase than with the top tier processors with current games. It leads me to think that there must be something allowing more to happen per core that brings these i3's closer to previous i5 performance, and that the top tier processors have yet to encounter the kinds of gaming loads that would max them out and bring a similar performance increase to the 6700k. GTAV is the closest we've come afaik, and that clearly does show that increased memory bandwidth really helps Skylake pull ahead.