Corsair Vengeance 32GB not stable with Asus Crosshair V Formula-Z @ 2400Mhz

[del_piero3]

Hello all,

as stated into another thread, i do have an issue with my current setup based on a Crosshair V Formula-Z.

Basically, i tried quite a lot of different settings and i was never able to get my 32GB Corsair Vengeance 2400Mhz CAS 11 kit stable on my AMD FX 9590. If i only put 16GB, it seems stable, but 32GB isn't and it always fail on memtest block move test.
I tried to increase the timings, increase a bit the voltage of the sticks (1.65V normally), of the NB, the VDDR, nothing went good. At 2133Mhz, lowering the timings a bit and everything is running fine. But i want to know why the kit isn't working at the expected speed without tweaking so much the system ?
My system is that one :
Asus Crosshair V Formula-Z bios 2201
AMD FX 9590
Corsair Vengeance 2400Mhz CAS 11 CMY32GX3M4A2400C11R
Corsair RM1000 PSU
AMD R9 290X Crossfire
SSD Sandisk Extreme pro 1TB
Western Digital Blue 2TB
Watercooling Corsair H110
Coolermaster CM Storm case

My settings are the following, i am a bit confuse why isn't possible to get this thing stable ?

[haihane]

del_piero3 wrote:
Hello all,

....

But i want to know why the kit isn't working at the expected speed without tweaking so much the system ?
....
My settings are the following, i am a bit confuse why isn't possible to get this thing stable ?

well, i promised and so here i'll try to deliver. i'm not the best out there, i'm afraid most pro(s) have left the building and gone for ryzen or threadripper so you have to make do with me.

i remember it was said quite often here back then, that running memory speeds greater than DDR3 1600 MHz on an AMD FX series processor depends very much on your silicon luck.

What is the silicon lottery?

When it comes to overclocking processors, every CPU is different. Imperfections during fabrication cause each CPU to have different limits in terms of clock speed.

AMD FX processors have their IMC (integrated memory controller) embedded inside the chip. you might have gotten a really strong overclockable FX processor (not to mention, the popular word of mouth is that FX 9590 is an overclocked 8350 binned by AMD themselves), or a weak one with little room left to push.

take my story for example, linked here and here , these posts date back to 2013 and outlined my difficulty to get 2400 MHz stable on a standard 8350 (v1400-ish.

many bios revisions later (and at one point i temporarily gave up trying to run these 16 GB sticks on 2400 MHz), fast forward to bios uh, what's the current bios now? oh right, v2201. your RAM sticks are @ 32 GB, double that of mine. the IMC had to work harder than mine for it to go stable. i remembered running these sticks of mine on 2133 for quite a while just to retain normal Resume from S3 (sleep).

also, running your sticks on D.O.C.P is the lazy method, i'd highly recommend you to set these up manually. i can't say much for your ram sticks, but take my ram sticks for example; these settings are worth a try (though not guaranteed to work):
- Bios 2201. i entered each settings manually.
- mine are on slot #2, slot #4 (slot #1 and #3 wasn't stable for mine). yours is a 32 GB kit, i assume 4 sticks @ 8 GB each?
- i had to push LLC and what's that other setting(?) to extreme preset to get it stable. i'll get you screenshot as followup post.
- save and exit

cross your fingers and hope it runs well. verify stability with stresstest and memtest AND sleep/resume from sleep test.
if all of the above fails, i guess you should manage your expectations and just run them at @2133, since the performance gain is negligible (to be honest).

[haihane]

Got some screenshots for you:










i forgot stuff: D.O.C.P is AMD's way of making XMP profiles work. my sticks were originally made for X87 chipsets. i find manually entering the ram settings help with stability. your experiences may differ.

[haihane]

delpiero, i'm curious about this:


any reason to enter 1.2v as NB voltage manually? (did the ram sticks specify it had to be that voltage? or was there some other reason?)
i checked mine, it's on auto (which defaulted to 1.09v something)

[del_piero3]

haihane wrote:
delpiero, i'm curious about this:

any reason to enter 1.2v as NB voltage manually? (did the ram sticks specify it had to be that voltage? or was there some other reason?)
i checked mine, it's on auto (which defaulted to 1.09v something)

Hi haihane,

thanks for trying to help me out and sorry for the late feedback.
Basically, i have putted 1.2V there because if you take a look to the 2nd scrrenshot, you can see in the XMP profile that my RAM needs 1.65V - 1.2V which for me was standing for the memory controller, so that's why i have putted 1.2V there.
I am maybe wrong, i also noticed in the previous thread where i was participating that some of them increased that voltage to gain stability.
I will give a try by reducing it.

Regarding the other settings, i really tried a lot of combination as you can see in the screenshot regarding the LLC and the power phase control.
My temps are really good, my CPU while burning doesn't exceed 68-69°C, but i am still not able to get my sticks 100% stable @ 2400Mhz. @2133Mhz, they are 100% stable with timings 10-12-12-30 and 1.65V.
Last thing, i am using the XMP only to pre-load all the settings in the BIOS, then i of course modify them manually like you did ;-). It is just a lazy shortcut to start and also it was intended to show you my XMP profile for the sticks.

[haihane]

hmm. okay.
super curious. me wonders what might happen if it's left on auto (NB Voltage).


you know, you can try this experiment yourself: get aida64, and run the memory benchmark test on: (read write speeds)
a. your known 100% stable @2133 MHz settings
b. your 16 GB settings @2400 MHz

see if the effort is worth the reward.


(maybe, a little part of you is like: this is supposed to work!. i was like that. as time goes on, that became more of a: meeeh, i'd rather this thing stable 24/7. even more so that ryzen and threadripper kicks this FX out of the water in performance.)

[del_piero3]

haihane wrote:
hmm. okay.
super curious. me wonders what might happen if it's left on auto (NB Voltage).


you know, you can try this experiment yourself: get aida64, and run the memory benchmark test on: (read write speeds)
a. your known 100% stable @2133 MHz settings
b. your 16 GB settings @2400 MHz

see if the effort is worth the reward.


(maybe, a little part of you is like: this is supposed to work!. i was like that. as time goes on, that became more of a: meeeh, i'd rather this thing stable 24/7. even more so that ryzen and threadripper kicks this FX out of the water in performance.)

Ok, so, last feedback is the following :
- reduced the NB voltage to 1.125V
- raise the VDDR voltage to 1.219V
- raise a bit the NB HT voltage to 1.206V
- put the NB frequency to 2400Mhz
- put the HT link to 2400Mhz
- put the CPU and CPU/NB power response control to ultra fast
- put the DRAM power switching frequency to 500Khz

and the rest untouched compare to my previous settings i was able to achieve running my DRAM @ 2400Mhz for 10pass in memtest without errors ... and then some errors occured in the 11th and 15st pass :(.






I understand your answer about stability versus performance, so i tried to benchmark the difference in between my RAM @ 2133Mhz with settings on 10-12-12-29 vs RAM @ 2400Mhz 11-13-13-31 and that is the result :




Pretty much the same results except for the copy where i see more than 1GB/s difference, but L2 and L3 seems slower in 2400Mhz which is weird in my opinion as the CPU is still running at the same speed ...

Ps: i know my FSB is overclocked to 300Mhz, but this isn't the issue since it run 100% stable with memory settings under 2400Mhz and my stability problem is also present @ FSB stock.