Overclock.net › Forums › Intel › Intel CPUs › *Official* Intel DDR4 24/7 Memory Stability Thread
New Posts  All Forums:Forum Nav:

*Official* Intel DDR4 24/7 Memory Stability Thread - Page 182

post #1811 of 5482
Quote:
Originally Posted by cekim View Post

I think you are misinterpreting my statement and I was admittedly vague...

We have to consider frequency, cycle counts and voltage not independently or even linearly related but as a matrix solution to the problem.

Using the absolute time of ((1/freq) * #cycles) as a "guideline" for what the DIMM/Kit can do, but not an absolute. The lousy cooling DIMMs get alone adds another huge fudge factor as you push the envelope here. The manufacturer bins according to meeting some threshold they believe they can price at any given time relative to overall order book needs. They only care that it can do what they say it can do. If it could do more or more with slightly more than 1.35v, then you _might_ get that for free or you might get nothing more than promised.

I was speaking both about my current kit(s) and their scaling as well future kits that might provide improved ratio of clock:CAS. There are now 3200CAS14 kits where 3000CAS14 was as good as I could do when I bought my 128G kit. Even worse is the 2800CAS15 32G kit I have in another machine or a 3000/15 kit in still another. Bigger kits empirically and logically put more strain on the IMC and thus a given kit might have better characteristics as individual or fewer than included dimms than they do as a kit (I've seen this repeatedly).

Then there is the matter that the CPU/IMC were designed with various assumptions and sweet-spots of frequency of each component (CPU, Cache, DDR) which affect their fifo depths, settings and pipe-line depth assumptions. These are ratios to which we have little to no information other than empirical. BIOS writers have some and Intel even more about this, but to us lowly users its a black box.

Of course they are not independent. When I raise frequency, I should expect any given timing to be more at risk of violating my sample's requirements than they would at a lower frequency because I am asking the DRAM's internal components to do the work of address, refresh, charge, discharge, etc... in less absolute time ((1/freq) * #cycles)).

For those playing the home game:
14cycles @ 3000MHz = 0.00000000466662 or 4.67nS
12cycles @ 2800MHz = 0.00000000428568 or 4.29nS

So, I sped up memory frequency by 7%, but now the Column Address Strobe sequence takes 8% LONGER in absolute time.

Let's go even more ridiculous and look at a 4266/19 kit on the `egg:
19cycles @ 4266MHz = 0.00000000445379 or 4.45nS

Which one will produce the best overall system performance? Dr WaitState decides... ;-) (well he and the IMC's ability to gang together page reads and writes and some large number of other factors that are chip architecture, OS architecture and even application dependent).

So, put all that together and the "best" performance of your system that is possible may very well not be the highest clock rate your IMC/DIMM will tolerate, so I will reserve my hunger for 4GHz memory until or unless I see that they can actually improve my system performance or prices make it a moot point.
Quote:
Originally Posted by cekim View Post



Let's go even more ridiculous and look at a 4266/19 kit on the `egg:
19cycles @ 4266MHz = 0.00000000445379 or 4.45nS

The actual latency is 2x your figures.prolly a good idea to point that out instead of leading people down a wishy washy path.
post #1812 of 5482
Quote:
Originally Posted by Silent Scone View Post

I got a great shredder last month, not let me down once. tongue.gif
Thanks again GRABibus smile.gif. You're scrutinous with your coverage

Scone, did you see my 8x4GB G.Skill DDR4 3000, Asus X99-A II HCI 400% run? A few days back I posted it. smile.gif

Edit: http://www.overclock.net/t/1569364/official-skylake-haswell-e-broadwell-e-24-7-ddr4-memory-stability-thread/1740#post_25309657
Edited by KedarWolf - 7/4/16 at 4:01pm
    
CPUMotherboardGraphicsRAM
Intel Core i7 5960X Asus Sabertooth X99 Gigabyte 1080 Ti Founders Edition Corsair LPX 
Hard DriveHard DriveCoolingOS
1.2TB Intel 750 PCI-E SSD 2TB Western Digital Black EKWB Predator 360 With One Titan X In Loop. Windows 10 Anniversary Edition 
MonitorKeyboardPowerCase
Acer XB280HK 4K Gsync QNIX QX2710 LED 27" WQHD 2560x1440 Corsair AX1500i Thermaltake Core X9 
MouseMouse Pad
MadKatz R.A.T. 8 Zowie 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel Core i7 5960X Asus Sabertooth X99 Gigabyte 1080 Ti Founders Edition Corsair LPX 
Hard DriveHard DriveCoolingOS
1.2TB Intel 750 PCI-E SSD 2TB Western Digital Black EKWB Predator 360 With One Titan X In Loop. Windows 10 Anniversary Edition 
MonitorKeyboardPowerCase
Acer XB280HK 4K Gsync QNIX QX2710 LED 27" WQHD 2560x1440 Corsair AX1500i Thermaltake Core X9 
MouseMouse Pad
MadKatz R.A.T. 8 Zowie 
  hide details  
Reply
post #1813 of 5482
Quote:
Originally Posted by Raja@ASUS View Post


The actual latency is 2x your figures.prolly a good idea to point that out instead of leading people down a wishy washy path.

Hello

Good catch. I skipped over most of that as it was starting to read as more of a recital than one's actual findings.
post #1814 of 5482
Quote:
Originally Posted by Praz View Post

Hello

Good catch. I skipped over most of that as it was starting to read as more of a recital than one's actual findings.
I wasn't trying to measure or predict latency, just showing an explanation of why my specific observation is backed up by the relative behavior of two different configurations and further extends to other possible configurations.

You guys sure are grumpy today.
Edited by cekim - 7/4/16 at 4:34pm
post #1815 of 5482
Quote:
Originally Posted by cekim View Post

I wasn't trying to measure or predict latency, just showing an explanation of why my specific observation is backed up by the relative behavior of two different configurations and further extends to other possible configurations.

You guys sure are grumpy today.

Hello

Seriously? When you are attempting to show the latency difference between modules why would you use a known faulty formula. It would probably be best to correct the formula in your post and the resultant values and just move on from here.
post #1816 of 5482
Quote:
Originally Posted by Praz View Post

Hello

Seriously? When you are attempting to show the latency difference between modules why would you use a known faulty formula. It would probably be best to correct the formula in your post and the resultant values and just move on from here.
Oh, now I see why you guys got your undergarments creased. I screwed up my math. My apologies. The hazards of typing on vacation.
post #1817 of 5482
Quote:
Originally Posted by cekim View Post

Oh, now I see why you guys got your undergarments creased. I screwed up my math. My apologies. The hazards of typing on vacation.

Well you are trying to teach people that already know this stuff. Not sure what the point is. Recall you jumped into a comment between JPM and myself, without knowing the full details.
post #1818 of 5482
Thread Starter 
Quote:
Originally Posted by KedarWolf View Post

Scone, did you see my 8x4GB G.Skill DDR4 3000, Asus X99-A II HCI 400% run? A few days back I posted it. smile.gif

Edit: http://www.overclock.net/t/1569364/official-skylake-haswell-e-broadwell-e-24-7-ddr4-memory-stability-thread/1740#post_25309657

Yep, will add recent results shortly. smile.gif
post #1819 of 5482
Quote:
Originally Posted by Silent Scone View Post

Yep, will add recent results shortly. smile.gif

Thank you. smile.gif
    
CPUMotherboardGraphicsRAM
Intel Core i7 5960X Asus Sabertooth X99 Gigabyte 1080 Ti Founders Edition Corsair LPX 
Hard DriveHard DriveCoolingOS
1.2TB Intel 750 PCI-E SSD 2TB Western Digital Black EKWB Predator 360 With One Titan X In Loop. Windows 10 Anniversary Edition 
MonitorKeyboardPowerCase
Acer XB280HK 4K Gsync QNIX QX2710 LED 27" WQHD 2560x1440 Corsair AX1500i Thermaltake Core X9 
MouseMouse Pad
MadKatz R.A.T. 8 Zowie 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Intel Core i7 5960X Asus Sabertooth X99 Gigabyte 1080 Ti Founders Edition Corsair LPX 
Hard DriveHard DriveCoolingOS
1.2TB Intel 750 PCI-E SSD 2TB Western Digital Black EKWB Predator 360 With One Titan X In Loop. Windows 10 Anniversary Edition 
MonitorKeyboardPowerCase
Acer XB280HK 4K Gsync QNIX QX2710 LED 27" WQHD 2560x1440 Corsair AX1500i Thermaltake Core X9 
MouseMouse Pad
MadKatz R.A.T. 8 Zowie 
  hide details  
Reply
post #1820 of 5482
Hey! Scone!!! I just realized you never added my 5820K nor 6900K results! I'm crushed ... cryingsmiley.gif




...

OK, I realize I never posted the 6900K here, and the 5820K posting was not a request to register ... ahh, to be young again rolleyes.gifbiggrin.gif
Edited by djgar - 7/5/16 at 8:46am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › *Official* Intel DDR4 24/7 Memory Stability Thread