Is Skylake worth the investment?

MageTank · December 12, 2015

I was referring to the link you posted, not the one in your sig.

Oh, ffs, 2nd and 3rd timings? Really??? Sure, if you are going to go THAT deep on tweaking, go ahead and get the most ballz to the wallz hardware you can.

But for the other 99% of the people, that is NOT a representative test. Just like benchmarks aren't accurate reproductions of real world games, playing with anything over primary timings is NOT a good representation of the auto settings.

Do you want a consistent methodology or a real world one (with stuff like antivirus, music, skype, and other things that would be what would actually happen for everyone else, but completely screw up with results)?

Personally, I vote for the second, but I'm pretty sure I'm a minority here.

I take it you've never overclocked DDR4 yet, have you? Finding the stable RTL timings takes minutes. Determining if they are stable, can be done within an hour. Want tighter RTL's? Make IO-L offsets looser. Want tighter IO-L? Make RFR Delay looser. Keep loosening until you fail to post, then go back one notch to the value that it last posted at. Then, run memtest86 and do a ton of Block Moves. You can skip the other tests for now, as you are only testing IMC timings, and Block Move is the quickest way to determine if it will fail or not. Ta-Da, free bandwidth in under an hour. Once you get these baselines, doing it for each clock speed variation becomes progressively easier, and you can even start guessing how your board will train the RTL's and IO-L's automatically, so i don't quite understand what you are complaining about. My 3200mhz memory > Any 3200mhz memory you can ever buy from a store. All because i took the time to get the extra performance out of it. Just like most other people take the time out to overclock their GPU's and CPU's for extra performance. Also, aside from tRFC, most of your secondary timings are pretty meaningless for performance.

As for the last bit, consistency > all else. We are doing this for the sake of science, right? Well, to do that, we remove all variables. Is this not how accurate tests are done? Either way, at this point, its your word (And the testing methodology of those sites) against my word, Digital Foundry's word, and several people on OCN. It doesn't matter anyways. I will tell people it can help, and the price difference is tiny, you can keep saying it won't help. It will be up to the purchaser to choose whichever they want, but at this point, I could care less. I did my job, and i know first hand what faster memory can do. If you want more evidence from me, feel free to request whatever test you want. I don't mind torturing this Pentium, or my $45 kit of ram.

Imakuni · December 12, 2015

I take it you've never overclocked DDR4 yet, have you? Finding the stable RTL timings takes minutes. Determining if they are stable, can be done within an hour. Want tighter RTL's? Make IO-L offsets looser. Want tighter IO-L? Make RFR Delay looser. Keep loosening until you fail to post, then go back one notch to the value that it last posted at. Then, run memtest86 and do a ton of Block Moves. You can skip the other tests for now, as you are only testing IMC timings, and Block Move is the quickest way to determine if it will fail or not. Ta-Da, free bandwidth in under an hour. Once you get these baselines, doing it for each clock speed variation becomes progressively easier, and you can even start guessing how your board will train the RTL's and IO-L's automatically, so i don't quite understand what you are complaining about. My 3200mhz memory > Any 3200mhz memory you can ever buy from a store. All because i took the time to get the extra performance out of it. Just like most other people take the time out to overclock their GPU's and CPU's for extra performance. Also, aside from tRFC, most of your secondary timings are pretty meaningless for performance.

As for the last bit, consistency > all else. We are doing this for the sake of science, right? Well, to do that, we remove all variables. Is this not how accurate tests are done? Either way, at this point, its your word (And the testing methodology of those sites) against my word, Digital Foundry's word, and several people on OCN. It doesn't matter anyways. I will tell people it can help, and the price difference is tiny, you can keep saying it won't help. It will be up to the purchaser to choose whichever they want, but at this point, I could care less. I did my job, and i know first hand what faster memory can do. If you want more evidence from me, feel free to request whatever test you want. I don't mind torturing this Pentium, or my $45 kit of ram.

Bro, I'm running apps in which are so RAM intensive that increasing bandwidth has more effect than adding a new separate instance on another physical CPU core; trust me, I'll be squeezing every single drop of performance out of my RAM and playing with 3rd timings, just as you do.

But would the average... hell, would even the "pseudo-expert" computer user do that? No. And that's what I'm trying to argue here.

MageTank · December 12, 2015

Bro, I'm running apps in which are so RAM intensive that increasing bandwidth has more effect than adding a new separate instance on another physical CPU core; trust me, I'll be squeezing every single drop of performance out of my RAM and playing with 3rd timings, just as you do.

But would the average... hell, would even the "pseudo-expert" computer user do that? No. And that's what I'm trying to argue here.

Oh, i wouldn't wish this torture on the average user, especially if they are as OCD as i am, and refuse to settle for certain number combinations. Not to mention, once you get done, it takes several hours of all 13 memtest86 tests (I just run it for 12 hours, and if i get no errors, i'm good to go) and another 12 hours of 512-4096 (6GB committed if you only have 8gb of ram like me) Prime95 just to be 100% certain that my ram is stable. Nobody will have that kind of time to overclock their ram, and to that point, i agree with you.

My point is, the cost difference between a Z and H series board, is on average, $20 (Not including the super mega 1337 gaming led pro series godlike maximus rampagetooth mark 5S boards), and the difference between DDR4 2133mhz CL15 and DDR4 3000mhz CL15 memory is $10. Not only will the Z series boards offer more performance when factoring in CPU overclocking, you also can get performance from overclocking the memory. Even one of your sources mentioned Fifa or F1 or whatever it was, getting a pretty substantial boost in FPS from memory speeds. From a Price:Performance standpoint, that extra $10-$30 on average just seems like a worthy investment when you compare it to how much more people will spend on a CPU for an extra 10% FPS. Seriously, on average, the difference between an unlocked i5 and unlocked i7 clocked exactly the same, will more than likely be 1-10% in 90% of the games one would play (very few exceptions to this rule) and people will spend a $100 difference in price for that performance.

Then again, worth is subjective, so i won't dwell on what something should be worth to someone.

TL:DR for all of my posts? Memory is cheap, and can make a difference in SOME titles, in CERTAIN situations with CPU overhead. If you have a board that supports higher clocked memory, consider chipping in that extra $10 and get a pre-overclocked kit and rest easy. If you have an H series board, just buy whichever ram is the cheapest, regardless of its speed.