RAMs need more voltage than before for stable overclock

[itsakjt]

Hi everyone,

I am having a problem with my system again. Since I purchased the RAMs on November 2013, I overclocked it to 1866 MHz 9-10-9-24 @ 1.6625V and it was stable. However, recently I have been having some instability issues which I was sure related to the memory. I cleaned the system entirely but it didn't improve. But increasing the memory voltage to 1.675V did the job. Now there's an important note:

My RAMs are good with other processors(FX 8350) even at 2000 MHz 10-11-10-30 and is stable at 1.65V even at that high frequency. But because of IMC limitation in my Phenom II CPU, it won't go over 1896 MHz and for some reason, the RAMs need more voltage to become stable.

My question is why do I require the extra voltage to make it stable now? Is it because of the RAMs or the CPU IMC is failing? Any help would be appreciated. The CPU still has a month of warranty left. The motherboard is fine. I confirmed RAM voltages with a friend of mine having the same board and we have got same values.

 

[itsakjt]

Anyone?

 

[natr0n]

Phenoms cant handle such high memory oc.

my ram runs 1667mhz just about maxed on my phenom.

You hit IMC limit basically on your chip.

 

[karakarga]

I am using AMD FX-8350 processor too, with Asus Sabertooth 990FX/GEN3 R2.0 mainboard. Asus has a unique specialty of allowing high ram speeds over 1866MHz. I am using G.Skill Trident X 2x8GB 2400MHz Rams at 2400MHz, 1,65Volts. SiSoft Sandra indicates 23~24GB/s overall speed! Not only Asus Sabertooth 990FX/GEN3 R2.0 but also earlier Asus Sabertooth 990FX/GEN3 can also allow this "hiper ram" specialty...

 

[_JP_]

IIRC, Denebs favor tight as hell timings, as opposed to blistering speed, for performance increases.

 

[itsakjt]

Phenoms cant handle such high memory oc.

my ram runs 1667mhz just about maxed on my phenom.

You hit IMC limit basically on your chip.

I know it does not for most Phenom II users but all these days, they ran fine with me. So why do I suddenly require the extra voltage?

I am using AMD FX-8350 processor too, with Asus Sabertooth 990FX/GEN3 R2.0 mainboard. Asus has a unique specialty of allowing high ram speeds over 1866MHz. I am using G.Skill Trident X 2x8GB 2400MHz Rams at 2400MHz, 1,65Volts. SiSoft Sandra indicates 23~24GB/s overall speed! Not only Asus Sabertooth 990FX/GEN3 R2.0 but also earlier Asus Sabertooth 990FX/GEN3 can also allow this "hiper ram" specialty...

Its basically the IMC's capability. I have got a friend who pushed his RAMs to 1872 MHz on an Athlon II X4 system while the board supports 1800 MHz max.

IIRC, Denebs favor tight as hell timings, as opposed to blistering speed, for performance increases.

Yeah heard about that but due to some reason, mine favors better frequencies. For example, 1866 MHz 9-10-9-24-1T works better than 1600 MHz 8-9-8-20-1T in my system. I did a lot of tests and got the same result every time.

 

[_JP_]

So why do I suddenly require the extra voltage?

Because of wear.
Overvolting isn't a care-free measure to take on hardware and in the long run causes mitigation.
It's that part of OC'ing where you really should start knowing about what you are doing.

 

[buildzoid]

I'm guessing that your IMC degraded. With Phenom IIs you've got to go light on the voltage as they are damaged much more easily than newer AMD CPUs.

 

[itsakjt]

Oh. I get it. So the thing is I have still got a month of warranty left for the processor. Should I RMA it?

 

[micropage7]

Oh. I get it. So the thing is I have still got a month of warranty left for the processor. Should I RMA it?

just test it, if theres no physical err you could rma it i guess

 

[itsakjt]

just test it, if theres no physical err you could rma it i guess

Yes. I have already requested RMA. And there's no physical damage apart from some scratches in the IHS. No bent/broken pins. Every information is perfectly legible.

 

[buildzoid]

If the CPU can still run 1600mhz ram as specced your RMA won't get accepted. Seriously you can't RMA just because something doesn't OC the way you expect it to or because it degrades when you OCed it. If the chip was dead I think they would accept it but as long as it run 1600mhz at 1.5V they won't.
Also the next Phenom is likely to exhibit the same behavior if you run the ram at 1.65V + high frequency for an extended period of time.

 

[Vario]

just lower the speed and tighten the timings

 

[itsakjt]

If the CPU can still run 1600mhz ram as specced your RMA won't get accepted. Seriously you can't RMA just because something doesn't OC the way you expect it to or because it degrades when you OCed it. If the chip was dead I think they would accept it but as long as it run 1600mhz at 1.5V they won't.
Also the next Phenom is likely to exhibit the same behavior if you run the ram at 1.65V + high frequency for an extended period of time.

Accepted this. Would cancel my RMA if I find my CPU stable at 1600 MHz.
Nevertheless, an upgrade might be coming for me and this time, it will be an Intel. A Z97 and a Haswell-refresh I think.

just lower the speed and tighten the timings

Tried but I loose some bandwidth. Nevertheless I have found a workaround. Bumped up the CS/ODT drive strength from 1.5x to 2x in the DRAM driving configuration. Now I need 1.6625V. Default RAM voltage is 1.65V. RAMs are good and goes 2000 MHz + from 1600 MHz stock when mated with a FX 8350. Also I think overheating might be the culprit. The ambient temp here is 37 degree C. Just found that the CPU even goes above 63C at full load and full stock frequencies. Will be buying a cooler but fingers crossed till June 4th because of semester exams starting from 23rd this month.

 

[buildzoid]

Accepted this. Would cancel my RMA if I find my CPU stable at 1600 MHz.
Nevertheless, an upgrade might be coming for me and this time, it will be an Intel. A Z97 and a Haswell-refresh I think.



Tried but I loose some bandwidth. Nevertheless I have found a workaround. Bumped up the CS/ODT drive strength from 1.5x to 2x in the DRAM driving configuration. Now I need 1.6625V. Default RAM voltage is 1.65V. RAMs are good and goes 2000 MHz + from 1600 MHz stock when mated with a FX 8350. Also I think overheating might be the culprit. The ambient temp here is 37 degree C. Just found that the CPU even goes above 63C at full load and full stock frequencies. Will be buying a cooler but fingers crossed till June 4th because of semester exams starting from 23rd this month.

Yeah Phenoms really like low temps. Loosing bandwidth won't matter that much Phenoms prefer low latency so going from 1870 9-10-9-24-1T to 1600 7-8-7-22-1T should yield better benchmark scores especially if you tweak the secondaries and tertiaries.

 

[itsakjt]

Yeah Phenoms really like low temps. Loosing bandwidth won't matter that much Phenoms prefer low latency so going from 1870 9-10-9-24-1T to 1600 7-8-7-22-1T should yield better benchmark scores especially if you tweak the secondaries and tertiaries.

Tried in all these 3 years but in vain. I can push 1600 MHz 8-9-8-24-1T. That's it. I tried 8-8-8-24-1T and Windows does not even boot(BSOD while starting windows screen). I have forgotten what 7-9-7-24 does. I tried 1600 MHz 8-9-8-20-1T but still bandwidth is less. And tweaked the sub timings as well i.e. Write CAS latency from 8 to 7,6,5; Write recovery time from 12 to 10, 8 etc. The Write recovery does some improvements(setting it from 12 to 10 increases AIDA 64 memory copy speed by 150 MB/s).
Till date, no RAMs were ever stable in my system at 1600 MHz 8-8-8-20. The other timings can be tightened but the RAS to CAS delay had to be set to 9 for 1600 MHz and above. The best combination I ever got was 1744 MHz 8-9-8-24-1T with my old 2*2 GB Kingston Value RAMs. But for some reason, my present 2*4 GB HyperX Blu RAMs are unstable(or maybe the IMC cannot handle 8 GB) at those settings.

These are the scores I get!

 

[buildzoid]

Ram bandwidth is important for gpus for cpus you want low latency. For example when I first set a cinebench r11.5 score greater than 15 I did it on 16gb of 1333 7-7-7-21 ram because while bandwidth dropped I got a nice decrease in latency compared to 1600 9-9-9-24 resulting in the small score increase I needed to break the 15point barrier.
AIDA 64 gives raw values but the best way to measure ram performance is super pi32m and cinbench r11.5. As for timings I would try the following:
1600 7-9-9-24-1t
1600 7-9-8-24-1t
1600 7-9-7-23-1t
If you can't go below 8-9-8-24-1t try increasing FSB until the ram is unstable.

 

[itsakjt]

Ram bandwidth is important for gpus for cpus you want low latency. For example when I first set a cinebench r11.5 score greater than 15 I did it on 16gb of 1333 7-7-7-21 ram because while bandwidth dropped I got a nice decrease in latency compared to 1600 9-9-9-24 resulting in the small score increase I needed to break the 15point barrier.
AIDA 64 gives raw values but the best way to measure ram performance is super pi32m and cinbench r11.5. As for timings I would try the following:
1600 7-9-9-24-1t
1600 7-9-8-24-1t
1600 7-9-7-23-1t
If you can't go below 8-9-8-24-1t try increasing FSB until the ram is unstable.

Thanks I will try.

 

[itsakjt]

So I tried some of the settings and 7-9-7-24 is impossible. 8-9-8-24 is stable and good but I loose some bandwidth. I am still wondering how this change suddenly took place. My chip had always been a good overclocker and I did not abuse it much too. For 2 years, I ran it undervolted for the core clock(3.7 GHz @1.375V stock voltage 1.4V and the NB at 2.8 GHz 1.25V). I had 2*2 GB Kingston Value RAMs then rated at 1.5V and personally I liked those RAMs more than my present HyperX Blu. Ran them for 2.5 years at 1600 MHz 8-9-8-24-1T and 1744 MHz 8-9-8-24-1T. I used to get more bandwidth too with my old RAMs at lower frequencies than this one My present RAMs are single sided. What are your views?

 

[Ferrum Master]

I had once simply oxidate, blackish markings on some pins... but using socket F LGA CPU's. But... they were really old, only acceptable reason for me, as the bus uses paired mirror signals usually those errors cancel out, but, who knows how it works using overclocking.

The second thing... transistors may deteriorate with time due to heat or leakage... It is not an excluded option.

Clean your gun barrels soldiers

 

[itsakjt]

I had once simply oxidate, blackish markings on some pins... but using socket F LGA CPU's. But... they were really old, only acceptable reason for me, as the bus uses paired mirror signals usually those errors cancel out, but, who knows how it works using overclocking.

The second thing... transistors may deteriorate with time due to heat or leakage... It is not an excluded option.

Clean your gun barrels soldiers

Yeah everything is clean and good as new. I even went to the extent of cleaning all the excess solder flux and residue but still no or minor improvement.
I also think the chip has degraded but still can't accept it actually. Because the only difference I did in.the last 6 months is moving to a 2*4 GB 1.65V Kingston HyperX Blu pair RAMs from 2*2 GB 1.5V Kingston Value RAMs. I didn't even.had to increase the IMC voltage.

 

[de.das.dude]

since i know the torture that CPU has been through, i think the cpu is aging.
even without overclocking cpus will age and their limits go down. happened to mine. also one thing that adds to the issue is these boards are am3+ optimised.

 

[itsakjt]

since i know the torture that CPU has been through, i think the cpu is aging.
even without overclocking cpus will age and their limits go down. happened to mine. also one thing that adds to the issue is these boards are am3+ optimised.

I think its because of overheating. The temperature goes up to 68 degree C on load. Anyways, after exams I am getting a new system. A Haswell refresh + Z97. As you know, now I am a reviewer and I want no/minimal bottleneck.

 

[de.das.dude]

nice. are you reviewing for ocfreaks?

 

[itsakjt]

nice. are you reviewing for ocfreaks?

Kinda yes. Though I have posted the review in thinkdigit and here as well(in a thread).
Check the review of the Asus R9 270X Direct CU II TOP I got from Asus India and leave your feedback and suggestions.

http://www.ocfreaks.com/forums/f110/asus-r9-270x-dc2t-2gd5-graphics-card-review-3487/