After not caring so much i decided to check up on my RAM as they are unstable on the XMP profile,
i think the MB is to blame for it (as i guess the RAM itself is supposed to be stable at it).
Anyhow,
here is the normal speeds supported:
Here is what i currently have it at (HCI Memtest 2700% passed):
Not sure if that overclock is on par with the XMP though, got no clue about the timings so just tried something.
If its stable with timings lowered that much, the chips are probably pretty good.
Id set them to the standard CL9 settings, and see if you cant hit 1700 or so MHz with 1.6 volts DDR
that old i7 will probably like at least 1600MHz better.
Perhaps try to do the XMP timings with 1600MHz. that would be worth while.
BUT your i7 is old enough not much past 160MHz will make much difference.
a mix of speed and timings will be better than timings alone. either way, 1333MHz, even at CL5, just doesnt offer the bandwidth that todays programs want, even if the platform itself was perfectly okay even with 1066MHz ram
crucial or corsair has a latency calculator on their website, you should be able to plug your numbers in and see what works best. CAS Latency and Command Rate are the two most important numbers outside of overall speed.
The question was, what timings do i need to have equal performance as the XMP profile (1600mhz) at 1333mhz.
Cause my MB seems to be picky if i use that, but 1333mhz is fine (not sure where the limit goes or how picky it is, kinda hard as you need to isolate it from the CPU which is hard as i need to change the BCLK).
I don't have an i7, i got an i5 760
I thought the latency was very obscure, tried searching for it but all i could find was (a ton of factors differ, and it goes from chipset to chipset and whatnot) so there was no formula.
But if there is, that's awesome, then i will use that to find what settings i need to be on par with the XMP 1600mhz
It probably is running pretty close the 1600 XMP profile performance wise @ 1333mhz 6-7-6-20.
You could probably try keying in the 1600mhz profile manually instead of using the XMP profile. So basically just up CL, tRCD, and tRP and set 1600mhz and see if its ok. Might be some of the other timings in the xmp profile that is throwing things out of whack.
You most likely wont be able to notice the difference between 1600 / 7-8-7-20 and 1333 / 6-7-6-20, but benchmarks should reveal the difference.
maybe se if you can drop tRAS down, maybe to 17 (from 20) @1333mhz
It probably is running pretty close the 1600 XMP profile performance wise @ 1333mhz 6-7-6-20.
You could probably try keying in the 1600mhz profile manually instead of using the XMP profile. So basically just up CL, tRCD, and tRP and set 1600mhz and see if its ok. Might be some of the other timings in the xmp profile that is throwing things out of whack.
You most likely wont be able to notice the difference between 1600 / 7-8-7-20 and 1333 / 6-7-6-20, but benchmarks should reveal the difference.
maybe se if you can drop tRAS down, maybe to 17 (from 20) @1333mhz
Actually have tried settings stuff manually, but seems to result in the same problem,
it might be that the CPU gets overclocked (2800 - ca 2880) (about 80mhz), but that seems weird, especially since it clocks itself to around 2.9ghz normally with the boost.
I will hopefully be able to overclock the CPU (again, have had it at 4ghz for long, but always had very obscure issues, and finally might have located it to the ram).
So if i get it stable at 1333mhz, i hope to be able to overclock the CPU and keep the RAM somewhere near 1333mhz, as i assume that would work fine?
Or does the RAM divider thing have an impact on stuff?
Will try play with the tRAS, but first going to see if it's stable at 1T (as i just used 2T to begin with) and i know it has some impact at least.
No clue what tRAS or the others have for impact though (except the CAS).
Trying to check my maximum BCLK, so i set the lowest divider on RAM and CPU to try to eliminate those factors,
but i am not sure how to stress test the BCLK (VTT/IMC?) part well.
I tried Memtest64 and it worked fine for 10 hours till i closed it, but i know it's not stable as it crashed when i played The Witcher 3 within maybe 10-20 minutes.
Done some short runs on Prime95 Large and Blend(Custom for RAM) with no success there (might detect it on longer runs).
But would like to know what to stress test to push the IMC instead of guessing around
EDIT:
Worth noting might be that i got the Red Screen (Solid) when it crashed during The Witcher 3.
But i haven't overclocked my GPU currently, and i doubt it was something with the temps
(i have played the game A LOT longer than the time it took for that crash).
EDIT 2:
Okay at least HCI Memtest seems stable as well (ca 580%, 1700mb x4).
Did some changes, and not sure if it was similar to what i had before, but it seems stable,
but if i crash in The Witcher 3 again with these, i will take a shot of it, though now it's likely to do with BCLK as i am trying to find the limit of that one,
so i got loose timings and clock.
Okay, even though it seemed to be stable, i got a crash in the Witcher 3.
Quite certain it's the BCLK, it seems to be possible to spot the instability when it's at 192 and above, but at 190 i can't seem to find it (easily at least).
So i guess it's borderline-stable, and The Witcher 3 just happens to push the strings correctly.
This doesn't have much to do with the RAM itself though, i had them really loose so i could remove that factor.
That being said, it 190 is almost stable, and i assume 180 to be quite a margin away from that,
it doesn't explain why the RAM is unstable with XMP as it only increases the BCLK a bit.
Which would mean (i guess) that the unstability their lies directly on the RAM itself.
Will do some testing at the XMP profile state and give a screenshot of it if i can't solve it.
EDIT:
Okay it failed quite quicky on Memtest, here's screenshots of the BIOS settings.
Worth Noting, the status voltages is from the last setting, so they do not reflect what's being show here*
I seem to remember reading how the PCH voltage needs to be within 0.5 volts of ram.
instead of 1.05, try bumping that to 1.15 (0.1 bump) and re testing. dont go higher, as I am not even close to 100% sure about that
If you get an extension in playing time, I would set the ram timings a bit looser, to the XMP settings (7-8-etc..) so you can afford to drop ram voltage a tad.
Also, bump the cpu pll from 1.8 to 1.82 if you can. this is the clock timer voltage, and SOME motherboards respond with more stable bus clock settings when overclocked.
Didn't touch the timings, but bumped the voltages, but sadly still got errors.
EDIT 2:
Hmm, kinda weird when i play stress test,
i expected Stressapptest to be better/faster than HCI Memtest, but i could run it for 4 hours with -W to no avail,
but HCI Memtest got it quite fast, ran it for maybe 30-60 min while playing:S
Am i doing something wrong with the Stressapptest?
EDIT 3:
Worth noting, HCI Memtest can fail at ca 1900% when ran as 1700x4 and left alone.
But if i play Overwatch when testing at 700x4 it fails way faster.
Will have to find a more solid way to detect the errors.
I sometimes use standard setting of intel burn test with half core count number of memtest instances open with approx 10% memory each.
Then attempt to browse the internet, forcing myself to not punch a hole into the monitor for how extremely laggy it is.
Seems to make the errors pop up stupid fast
Have you tried adjusting the QPI frequency (under ram freq setting)??
Try removing that from auto and just change it to whatever gives you a frequency roughly double ram speed. Seems to change quite a bit motherboard to motherboard.
That frequency is unfamiliar to me, but it appears to be similar to AMD's northbridge, acting as a data highway to internal components. Too high can induce errors under random loading, according to a very brief google search just now for me.
I found Stressapptest to be better/faster than HCI Memtest for finding memory errors but HCI was faster for finding chipset\memory controller related errors on my system anyway
Wait, you use intel burn test with half core, and memtest with half core 10% of ram?
Or do you use intel burn test fully and memtest half core with 10%?
Dasa, to me Stressapptest doesn't really find the errors on RAM weirdly enough, at least not quick, can run for 10 hours without issue,
and with HCI it goes way faster, and the solution was bumping DRAM Voltage so i would assume it's purely memory related in those cases.
Think the QPI is generally the lowest, which would be double DRAM, could do a recheck but pretty sure it won't matter (played around with that one before i think).
That ram you have are pretty good uses PSC chips. Around 2000Mhz@CL8-9-8 / 9-10-9 is possible or even more depending on the quality of the ram and PCB.
with my 8 core, I would start 4 instances of memtest with approx 1600mb each (16gb overall in system), let them run
THEN start intelburn test, initiate the standard 1024mb test.
With those 5 things running, Ill waste some time on the internet, and cross my fingers that I dont rage quite with a fist or foot put through the monitor because of the laggyness lol
Loosen the timings a little more first, see if you can get the memory stable at the speed you are looking for. If you cant do it with loose timings (9,10,9,28 2t), then you may have instability elsewhere. If you have stability at a lower memory speed, then it should be IMC voltage that may need a bump. Dont think it would need any attention at 1600 though.
Once you get it stable at the speed you want, then gradually lower the timings and see where you lose stability. I have some PSC 2000 running 2400 9-10-9,28 2t... its pretty quick stuff, just need to feel it out and learn what the modules like. I could run 1t up to 1866, but beyond was no go with my x79 setup. On my old x58, I could run 1t up to 2000ish.
slack timings, was able to get the old i7 to boot like 2000, but I think it kept freezing with the hynix ram. I could get 1600-1700mhz stable though. they want speed for sure.
find a cpu limited section of game and compare the fps
or export images in lightroom and compare the time it takes
depending on what your using the system for you just need to compare how quick it is before and after
there is no synthetic benchmark that can do this for you as some programs require more files from outside the cpu cache than others
Btw, on the CPU front, is Manual or Offset prefered?
It seems Offset allows adaptive voltages (which i guess is default?),
and Manual will enforce one voltage at all times.
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Related Threads
?
?
?
?
?
Ask a question
Ask a question
Overclock.net
27.8M posts
541.2K members
Since 2004
A forum community dedicated to overclocking enthusiasts and testing the limits of computing. Come join the discussion about computing, builds, collections, displays, models, styles, scales, specifications, reviews, accessories, classifieds, and more!