F317066CL9T-6GBPS REPORT - problems:insane voltages needed to run way lower than spec

[GSKILL TECH]

We simply use all tests as they all test slightly differently. Many times we even run them together, just to completely overload the system. For IBT, 32 threads 10-20 passes on xtreme is sufficient. Since we are the second line for quality control (Taiwan office being the first), we typically do not test any more than 10-20 passes. So far, this has not been an issue. I don't see how it would either since most people may never reach anything close to 4 threads of load during the lifespan of their computer.

Thank you
GSKILL SUPPORT

[moebius]

Personally i had hard lockups/errors/autoreboots after 11 and 15 passes of IBT 32 threads xtreme.10 is definitely not enough.
The reason for the use of 32 threads being advised is that it leads to results that are reliable ,compared to the total unreliability of IBT 8 threads (even with xtreme stress mode), in a really shorter amount of time(sometimes it leads to stability guarantees results that cannot be achieved with 8 threads).
I have had my system in a configuration that passed 19 passes of IBT 8 threads xtreme,but then hard-lockupped at the first of IBT 32 threads xtreme.
ANyway,if you make use of high threaded math programs(for example),that's for sure that you don't want your matrix to be filled with wrong results,so i think that the very least to test stability is 20 passes..Now i don't know what is the stability testing methodology in taiwan,but it would be surely something to point to like an example in the industry if they tested with more than 2 hours (on a 4 ghz system) test each stick(or kit) of memory.
I just wanted to know what were your parameters of stability,given that as i found out,for example, Corsair(on their gt 2000 cas7) put in their xmp a vtt volt of 1.6.This could be the reason why i found so weird the voltage required on my system to run these sticks.But it was far from being higher than usual if Corsair consider it the default voltage.
I think G.Skill should advertise such high requirements,so that it is something more clear that to run such speeds on X58 systems,nothing less than 1.6 vtt is USUALLY needed to run with REAL stability.

Anyway using IBT with only 8 threads,as well as using LinX (whose amount of threads cannot be customized),cannot be called true stability tests,given aforementioned considerations(and tests).
Btw ,compared to other stability tests IBT is the only one that can saturate COMPLETELY system resources(cpu and memory),in such a way that nothing can be done during its use(even right clicking on the desktop while it's running leads to no result if not after minutes sometimes).prime 95,which is considered to be the best together with IBT,doesn't come as close to test as heavily as ibt does cpu or memory,given that you can easily browse the web or do other stuff while running it.
So,even after 12 hours of testing with prime 95 ,it is LIKELY that your system is stable,but you don't have any warranty that all of your cpu registers have been tested,not SURE (or almost) like with IBT that tests all of them at once.
Add to this the fact that Prime 95 needs heaps of time more than IBT and that IBT simulates the variability of loads so that the changes in voltage (when the load decrease and increase) stress furtherly the cpu(more warranties of stability)and i really can't see the use of other longer and less reliable stability tests.Anyway i thank you for your answer and explainations.

Kind regards

[GSKILL TECH]

I agree with you on the testing as a consumer. But the reason why we only run 10-20 loops is because most of the time the memory we test is new anyway, so very likely it is good if it passes 10 loops. Now for a end user, if they want to make sure their memory is working flawlessly, then of course they can run it for as long as they want. Otherwise if we had to test each set for hours at a time, mutliple times, all of our memory would be limited edition since we can't even produce them that quickly.

Noting the high requirements would be ideal since not all users may be aware, but it is expressly implied when you purchase memory that is the highest rated for your motherboard and it also includes the "(OC)" on the side. In my opinion if you are unaware of these requirements, you shouldn't be putting an overclocking system together. People should stick to the JEDEC standard if they are looking to plug and play. Unfortunately people purchase all the top of the line parts, then contact us to help them set everything up since no other company really wants to. We have no problem with this, but it does become one when a customer blames us for bad memory or say we're a horrible company simply because they plugged memory in and it doesn't work at DDR3-2200. Or they say we're a fraud/scam company because we're selling DDR3-1333 at DDR3-2000+ or whatever. Just a little insight from a tech, but in the end it's no big deal, the satisfaction of helping others that do appreciate the support outweigh the rest. Paying for a computer technician can get quite pricey, so I'm just doing my part to help people save time and money in these tough economic times.

Thank you
GSKILL SUPPORT

[moebius]

I agree with you on the testing as a consumer. But the reason why we only run 10-20 loops is because most of the time the memory we test is new anyway, so very likely it is good if it passes 10 loops.

The fact that the memory is new doesn't mean that it cannot have faulty chips.Your statement is kind of pointless.A memory is not good if it passes 10 passes because it can easily be faulty even after having passed 10 passes.
I must agree that making ten passes ensures a lower percentage of returns...

Noting the high requirements would be ideal since not all users may be aware, but it is expressly implied when you purchase memory that is the highest rated for your motherboard and it also includes the "(OC)" on the side. In my opinion if you are unaware of these requirements, you shouldn't be putting an overclocking system together. People should stick to the JEDEC standard if they are looking to plug and play.

I would gladly note the requirements if they were written somewhere. The fact is that an important requirement is missing(on your site as well as on the memory package as well as in the memory profiles).This requirement is the high vtt voltage.This requirement is well specified in other brands memory and it seems that neither you were aware of it.
In your posts infact you told that my voltages were not normal at all even if i wrote you that i needed voltages in the order of 1.6 vtt,so in line with what is needed with this speed and this chipset(Corsair puts 1.6 volt in their xmp profile for ddr3 2000).

Unfortunately people purchase all the top of the line parts, then contact us to help them set everything up since no other company really wants to.

Every serious company has a forum where it has a direct contact with users that experience problems,so i would not say "no other company really wants to".
G.Skill sells products meant to be used in overclocked configurations.It is normal that a serious company,as G.SKill is, provides tech support for overclocked configs.

We have no problem with this, but it does become one when a customer blames us for bad memory or say we're a horrible company simply because they plugged memory in and it doesn't work at DDR3-2200. Or they say we're a fraud/scam company because we're selling DDR3-1333 at DDR3-2000+ or whatever.

I never made such statements about the quality of the company.Please don't point on the fact that i did.
Of course from a 400$ kit you expect smth more than from a 200 one.If the speed attainable with this memory kit depends upon the IMC,the same cannot be said for timings.These depends upon the memory chips.I just tested again with IBT the timings of 7-9-6-20 at ddr3 2000 with this kit and they were not at all stable.It would have been the least to attain them once i managed to stabilize 2000 speed with this kit,but i did not.So what i can say is that i am not really satisfied from this product,and not because i did not manage to attain 2133 speed(that depends upon the cpu's integrated memory controller will to go high with uncore clock) but because timings attainable at a considerably lower clock were not those expectable.

so I'm just doing my part to help people save time and money in these tough economic times.

You are doing your job,and all those who enjoy it,and i am one of these, can appreciate it and the quality of G.Skill support

Thank you

[GSKILL TECH]

No No, I am not accusing you, complaining, or anything of that sort, simply sharing a little of my experience. Regarding testing, our Taiwan office does a much more rigorous job testing memory, and that allows us (USA) to save a little time. In return, it may only lower the percentage by running 10 passes, but chances are almost 99% that they are not defective considering our returns to sales per month is only 1%.

Thank you
GSKILL SUPPORT

[moebius]

The reason for you finding it is enough 10 passes is that in taiwan they make heavier tests than 10 passes so when you test the memory ,ten passes always give good result and it would give the same result if you made 20 or 30 passes(not considering few isolated cases).
What i can tell basing on my experience is that there is a high percentage that JUST a 10 passes stability test gives few warranties of stability.
(this test not only tests the ram but the whole system stability:cpu mb and ram, so errors in it can mean the instability of any of these components.)
To give you an idea:
i am testing this kit with these settings

BCLK =202
multi=21
VTT= 1.725
IOH =1.44
VCORE=1.39375
DRAM=1.72
DDR3 SPEED=2025
QPI SPEED=7290
ALL TIMINGS =AUTO (9 9 9 24 tRFC=98) except CR set manually to =1N

(The system was previously tested and found stable with the same settings at CR=2N and 1.67 dramV).
With the settig CR=1N,though,it will be stable for 11 passes IBT 32 threads.At 12th it gives a error.Now it seems that it is not possible to stabilize it further.So you can see that ONLY a ten passes test would have not been enough to tell that my system was stable.I have tens of such examples ...
I wrote all this ,first of all to share my doubts,secondly to report the overclockability of this kit and least but not last because i see around everyone testing the stability of a system with a 10 passes 8-threaded LinX.This is simply ridiculous,since it can easily happen that asystem stable 19 passes 8 threaded IBT xtreme mode(xtreme mode makes it a heavier test than LinX ) fails,with a system freeze,at 1st pass of IBT maxed out.
To have more consistent results of the REAL overclockability of a system it is pointless to use such light stab test(it is not pointless only if you are into the bragging game of posting screenshots,but that's another story).
So i wanted to know which methodology you used to test your systems.
Thanks for your answers