Intel Haswell Overclocking Clubhouse.

[jaggerwild]

Get you some of DAT!

jaggerwild|4790|1.212|5402Mhz|54X100.05|1.55|x40(4000MHz)1600Mhz|Phase Change|GigZ97X-SLI-CF|stock Mem

http://valid.canardpc.com/ll55xv

http://hwbot.org/submission/2619510_philly_cheese_steak_cpu_frequency_core_i7_4790k_5402.61_mhz

 

[TheHunter]

Hey @cadaveca or any other ram timings expert

I did some voodoo magic and got read 38.8GB/s @ 2600MHz, finally after 3.5yrs lol But only half way,

this one highlighted was @ 10 for 2400MHz , now I set all 3 to 11.. and tWTR 10 was manual 8 and tRTP 10 was also 8, could be one of them for write and copy?

EDIT: although maxmem2 and cpuz benchmark showed the same score as 2400mhz,
maxmem 2400 had 34gb/s read, now at 2600 26gb/s, but overall score was the same 26.6gb/s and 0.5ns lower latency.


EDIT2: this is the best I could do and fix write, copy.. I lowered these 3rd timings close to 2400 timings; 2nd timings are somewhat mix of 2400-2600 playing it safe..

If I chose last 3 @ 10 it won't post,

and what it looks like in aida64 now




So is this the best I can do or?

 

[emissary42]

In the UEFI try

tRDRD_dr/_dd 5
tWRRD_dr/_dd 4
tWRWR(_dr/_dd) 4

Also most people just use ASRock Timing Configurator (ATC) or ASUS MemTweakIt to check on timings.

Single sided Hynix MFR @ 2800:

 

[TheHunter]

Ah I thought those ram tools were just for OC'ing and ROG specific, that's why I never used it

Ok will have a look.. So can I OC ram too or just timings?

 

[emissary42]

You can try both, success varies from kit to kit and with IC type / quality.

 

[TheHunter]

Yes these g.skill 8GB sticks are not so good, but from what I saw most TridenX 8GB 2400mhz OC crappy, some couldn't even do 2500MHz.

My max is 2600 CL11, 2666mhz already failed even if at CL12... For example My old Crucial 2133 CL9 Elite 4GB sticks did 2666MHz @ CL11


ps; Your 40GB/s looks great , what is that ram, Corsair Platinum?

 

[TheHunter]

Ok, so I used default 2400MHz 3rd timings and relaxed 2nd a bit and got this @ 2600Mhz.. Finally

I tried 2666mhz, but I wouldn't boot.. doesn't matter anymore, this looks perfect now


Thanks @emissary42

 

[emissary42]

Yes these g.skill 8GB sticks are not so good, but from what I saw most TridenX 8GB 2400mhz OC crappy, some couldn't even do 2500MHz.

Well most 2x8GB DDR3-2400 CL10 use 4Gbit Samsung ICs and those were nowhere near their 2Gbit D-die in terms of overclocking potential. Your final result does look pretty good, though,.

ps; Your 40GB/s looks great , what is that ram, Corsair Platinum?

These were ADATA XPG V3. Had a user review planned with them, but the ADATA Rep never got back to me on a few questions, so i just did a few quick tests and sold them off.

 

[mouacyk]

Ok, so I used default 2400MHz 3rd timings and relaxed 2nd a bit and got this @ 2600Mhz.. Finally


View attachment 81265


View attachment 81266

I tried 2666mhz, but I wouldn't boot.. doesn't matter anymore, this looks perfect now


Thanks @emissary42

Being that these were 2400MHz default, what final voltage did you have to use to get the 2600MHz?

 

[TheHunter]

Being that these were 2400MHz default, what final voltage did you have to use to get the 2600MHz?

Hi,

I didnt change voltage, always 1.65v.

Only digi+ dram mobo power current from 100% to 120% or mv by GB/Msi and a bit higher cpu sys agent, but I didn't go over 0.060v+ yet, no idea how it acts @ 0.100 - 0.150mv+ offset.. It was the same with my old 4x4GB CL9 2133mhz Crucial blx elite when I ran them at 2400 or 2667 for testing.


Although its not completely stable, can crash more demanding games or even hard freeze the system.. Still need to adjust few timings, Im thinking all 3 tRDWR to 12 instead of 11. One guy @ guru3d tested the same ram as mine but 4GBsticks CL9 2400 and he said 2666 CL10 had a hard time too even at 1.70v, he has haswell too, but Gigabyte mobo..

This was his attempt.

 

[mouacyk]

@TheHunter - Your 2x8GB might be fully stable if you bump up voltage a few notches and/or System Agent a few notches too. I needed 1.7v and +0.05v to SA to be prime 24-hour stable at 2666MHz 11-13-13-35-2T with 4x8GB. I have mixed a 2666 and 2933 kit. These double-density chips will probably need that extra voltage at higher speeds and higher capacities.

 

[TheHunter]

Ok so I tested it yesterday and speeds are great, but stability not so much, tRDWR 12 or 13 no change, at auto it sets 10 and somehow boots, no way it does it if manual..

I also set cpu system agent 0.100v+ offset and it didnt make any difference either, but I didnt try 1.70v, feel a bit scared. :O


Usually BSOD memory management, I think 0x1A., happened 3x when I tested simple thing like Aida64 memory benchmark, by memory latency (app error at first, then bsod ). Also seems to be more picky now @ XMP2 profile.. I had it at XMP1 at first, but I noticed default 2400mhz is faster @ XMP2 - auto tighter timings. Maybe I'll re-check XMP1 again.


But I found something more interesting with cpu sys agent, saw this debate once, I used 0.080v and set cpuv from 1.288v to 1.274v and its stable - e.g. mmo RIFT can fail quick in heavy raid, its now 64bit and very multi core.
It passed other stuff too, even today as I type this - new cold boot and its a little colder compared 1.288v (up to 2-3C),.

I remember it benchmark'd 4.7ghz ~1.26v, early bios 3yrs ago, although not 100% stable but it passed some demanding apps. Maybe that bios auto adjusted higher cpu sys agent, shame I didn't monitor it.

Hmm might try 1.265v later again, lol I was already thinking it degraded a little, well it use to be ok at 1.284v now 1.288v, but I also used a little lower cpu sys agent since its native 2400mhz ram..


Also did a quick 4.8ghz test with cpu sys agent 0.100v+ 2400mhz ram and I stopped by 1.335v, got a bsod watch dog 0x101 and this one is tough to solve.. I already tested it once and it can be anything cpuv, cachev, input voltage, digi+ cpu/ram current %, cpu sys agentV :S


EDIT: multi edits, added stuff..

 

[mouacyk]

@TheHunter You'll be fine at 1.7v on the DDR3, even up to about 1.8v for 24/7. However, above 1.65v and especially in 4 dimm configurations, make sure you have airflow -- or the heat will build up and de-stabilize your clocks causing random BSOD's and crashes. As for SA, I would not go above 1.25v (air) or 1.3v (water) on it.

I remember being able to run SuperPi and HyperPi 32m at 1.65v, but it would stop in longer runs of prime 95 or realbench or LinX with >90% of memory. Only by slowly incrementing the DDR3 voltage was I able to reduce the memory errors eventually and achieve stability. When you see that incrementing DDR3 voltage doesn't help anymore, then it's time to up SA by a notch or two and see if that helps.

 

[Toothless]

Anyone know if leaving XMP off for a 4x8 ram kit is better than having XMP on for overclocking?

 

[cadaveca]

Anyone know if leaving XMP off for a 4x8 ram kit is better than having XMP on for overclocking?

Depends. But yes, mem OC can affect CPU limits at the high-end, but really its more about the required VCCSA and VCCIO voltages increasing heat that is really the problem. So, test CPU OC first, find max, then do memory.

 

[Kursah]

Anyone know if leaving XMP off for a 4x8 ram kit is better than having XMP on for overclocking?

Depends on if timings or voltage from that XMP profile are stable or not with the remaining system configuration. I usually run XMP set and then tune from there. With 4 Dimms you might bump up the voltage a touch, I had to. Especially I'd you're overclocking your RAM.

I run 4x4 1866 cl9 @ 2133 cl10, and I gotta run 1.55v to be stable on all 4 DIMMS, 2 DIMMS will do 2133 @ 1.5v. Not a huge OC but easy enough to achieve for what is likely more noticeable power consumption than performance gains. I was still able to achieve my 4.8 OC at 1.26v and 1.65v on CPU input. But my OC isn't very extreme on the memory, if you push beyond where I'm at VCCSA and IO become more of an necessity for tuning in some cases.

If I ran 1.50v on the RAM my system would not be stable. I tried messing with the System Agent IO voltages in increments up to .150v, didn't help me but really I wasn't far out of adjustment.

But for more extreme OC's, and for initial testing, exactly what Dave suggests. Start with CPU, get it and the cache sorted out if you do cache tuning. I do... though I leave the max multi at 40 (on my 4790, was 39 on my 4770) and set min to 8X. Noticed a drop in idle wattage consumed from the wall of around 8-10W as measured with UPS and software. The RAM OC made up for most of that.

I also suggest undervolting at stock, and using adaptive voltage. I have found I can run my undervolted settings and as-soon as my CPU hits turbo, applies my OC voltage (with some compensation to actually achieve that voltage past the negative offset from undervolting) and works great. Keeps my system cool, quiet and a little more power efficient plus it added a little to the challenge of system tuning. I was able to shave off -.065v off of my stock VID for stock speeds (up to 4GHz, turbo is beyond that). You probably know how the adaptive works, I found it to be worth using and tuning with once I knew my undervolt @ stock and overvolt @ OC stable spots. I left most settings on auto, monitor via software, nothing really changes except what I change or what XMP changed that is detected. And I lucked out and can hit 4.8 on air, but I've seen quite a few 4790's accomplish that...though the one in my server most absolutely cannot...that's why its in the server!

Good luck on your OC tuning!

 

[Toothless]

I should've included overclocking the 4790k, not the ram.

Oops.

 

[TheHunter]

In my case I gave up ram Oc 2600mhz for now and focused more on lower cpuv AND I've managed to lower it further..


Vccsa 0.010v = cpuv 0.003v or so..

That said
cpuv 1.278v - vccsa 0.110v offset
Cpuv 1.272v - vccsa 0.130v

Now atm testing cpuv 1.266v - vccsa 0.150v

Use to have cpuv ~1.288v - vccsa 0.050v

Vccin ~ 1.78v with llc auto aka max llc8. Cpu vrm current auto aka extreme.