Install Steam
login
|
language
简体中文 (Simplified Chinese)
繁體中文 (Traditional Chinese)
日本語 (Japanese)
한국어 (Korean)
ไทย (Thai)
Български (Bulgarian)
Čeština (Czech)
Dansk (Danish)
Deutsch (German)
Español - España (Spanish - Spain)
Español - Latinoamérica (Spanish - Latin America)
Ελληνικά (Greek)
Français (French)
Italiano (Italian)
Bahasa Indonesia (Indonesian)
Magyar (Hungarian)
Nederlands (Dutch)
Norsk (Norwegian)
Polski (Polish)
Português (Portuguese - Portugal)
Português - Brasil (Portuguese - Brazil)
Română (Romanian)
Русский (Russian)
Suomi (Finnish)
Svenska (Swedish)
Türkçe (Turkish)
Tiếng Việt (Vietnamese)
Українська (Ukrainian)
Report a translation problem
Discussions Rules and Guidelines
Report this post
It would be interesting if u could replicate it with setting cpu very low or ram very low, in such a way it should be guaranteed stability. If it still happens then you might be unto something.
I also use max pre-render frames set to 1. So since u can't use that, yeah this isnt an nvidia issue, it's just something specific to your pc setup.
edit: im going to test it a bit. it seems with it off, fps rates are more stable, but way lower.
It's hard to say without demo recording and thus benchmark options via running rivatuner, however it felt like fps was perhaps more consistent with it off, but ah. It's hard to say.
Anyway, for me both on and off work just fine.
I've seen a lot of users having this crash.. I doubt all have now system stability all the sudden. e.g. latest COD's are more demanding, BF also and both use more system ram.
My cpu voltages are fine, ram also. PSU as well.
4770k @ 4.6ghz 1.245v, when I say fine Im stable at 1.235v too.
mobo is z87deluxe, I own this combo for over 5 years, I know it in detail by now. Also made a haswell OC thread long ago @ guru3d forums.
ram is gskill tridentx 2400mhz 1.65v, also with extra mobo vcsaa for better pcie/dram stability
*I do have command rate 1T, instead of default 2T, haven't changed that ever. Other adv dram timings are mobo optimized default, didn't tweak those at all, all are rather more loose.
psu has single rail 52a, and stable 12v+ line.
gpu is also stable 980TI 1405mhz factory OC, vram stock. atm still at 417.35 driver, win10 RS5.
I play the game at higher then desktop 1080p resolution (DSR 2880x1620p), but for me fps is better now after this latest update from yesterday.
I use to think its something with new 41x,.xx driver since 399.xx or specifially 397.62? was fine, I always had threaded optimization -ON back then.
Speaking of frames to render ahead, I have 144HZ and don't see/feel any lag @ 2, also fastsync eliminates extra vsync lag.
--------------
I posted this @ nvidia driver thread - guru3d forums just now, migh as well quote it here.. And all my findings that I didnt mention here.
PS I play the game @ default Ultra, except effects, postprocess to high, and postAA @ TAA.
Under the term stability it must be enough for low temperature as well as high temperature conditions as performance/stability changes obviously due to these conditions. Which is why a cold pc can run stable in tests all day long, but once under proper pressure it kneels under and crashes as temperatures are higher.
Often one can read that "I set +.002 for rail stability" for example. There is no such thing. If you found it was stable it's stable. Adding extra voltage is often done due to lazyness just to "make sure". Also, untweaked ram timings, yuk!
I tweaked adv ram timings enough to find out no practial difference in terms of speed 36GB/s and latency 42ns, it already is as good as it can get.
Tested aida64 memory benchmark and cinebench and various cpu benchmarks e.g. 3dmark physics tests.
Well I just played QC again and it happened again, so idk.. the only thing I can suspect next is either this ram 1T or cpu cache voltage - still too low.
Or just uninstall this EA pos and stop wasting my time.
I've currently set 3400mhz for ram which beats my previous 3800mhz at same primary timings in real world tests such as cinebench and realbench, especially encoding and ingame frametimes are better. Boot times are faster as well, it's quite something once u get really into it. Ultimately i could make a 3800/4000 setting faster, but in order to beat the current preset i made i would need to go way beyond 1.5v and it wouldnt be that much to gain but a lot of constant heat.
I run the ram at 0ns IOL, 0ns being as close to 0ns, but detected in bios as 0.
Memtweakit reports 1.4x million points, my 3800 setting at about 0.4x million. Memtweakit cannot handle 0ns for some reason, aida 64 reads it ok.
One thing that is great with achieving lowest possible IOL, is that tweaking ram gets easier, as errors are detected faster, so you in turn have an easier time dialling in perfect ram settings.
Cpu cache is also heavily dependent on ram timings. Especially write timings, anything with rw, wr, or ww. You want lowest possible write timings, and you do not want to lower read timings and then raise write timings to get write stable. I had about 37ns on 7700k with a poor kit and a weak msi mobo, but knowing what i know now i think i could get it lower.
I just checked again my cachev and I think that was the culprint afterall, meh, I've raised it further, forogt that I lowered a bit again few days ago.. Last bios was 2years ago, and it did something to this cache.. I remember earlier beta seemed more stable as overall with lower volts across the board.
btw;
Im still on DDR3 so it's a bit rough. Best I can do with this 2x8GB kit is 2600MHZ and 39GB read, write 40GB, copy ~ 38GB.. latency is somewhat the same think it was ~ 40ns. with tweaked third timings. At auto it sets them too loose and I actually get negative bandwidth. But its very unstable at those timings. Think I hit the wall there with cache+cpu freq and that ram freq.
I can do 4.7GHz @ 1.29v, but for that I have to sacrifice cache back to 4GHz with 2400mhz ram, atm 4.2GHz with 4.6GHz cpu. which in end is like 4.7Ghz +4ghz cache.
I dont think I have this RTL/IOL on my board.
EDIT: actually I do (for channelA and B), but idk how to set those, I never tweaked them ever :), only secondary and third timings - namely yes only this trdrd and twrrd.
But yeah the funny thing is, QC was 1000% stable prior March 2018 patch before they did something for nvidia + intel or at least so I've read, I didn't see any intel cpu mentioning in that patch though.
EDIT2:
This crash now happened after I wakeup pc from sleep, so it could very well be connection with that, I saw sometimes nv driver can have some glitches if you wakeup from sleep instead of fresh reboot.. so idk,
I run it so each channel have the same response, in this case 0/0/0/0. Unfortunately I dont have a way of measuring how close to 0 it really is, but i suppose setting up a meter at each end and putting a current on there would do the trick alltho i have no intention of trying to fry the motherboard ;)
At 3800 I could have 2/2/2/2 but very hard to attain and not really functional benefit in any real world tests, especially since its 64gb and double rank double sided dimms.
What benefit you get is proportional to motherboards electric pathways to and from cpu/chipset to ram and vice versa. My motherboard is specifically optimized for this purpose hence i can achieve good results. There is also a distinct difference between every other multiplier for ram bus speed, ie easier to train on 3400
If I were to choose between optimizing for lowest ram timings or good RTL/IOL values, i'd go for the latter, as it in my experience helps again unnevenness of frametime/response on a large scale. While optimized ram timings is more for crunching numbers so to speak. Consequently if you can't get some timings stable increasing RTL can also work and so relaxing that to higher values will give your system more time to make a low read timing f.example stable. At the cost of fractional delay added.
So there is a lot to consider and this doesn't even cover the tip of the mountain so to speak.
2600MHZ is not optimized for my mobo, I remember early bios wasn't auto optimized either and 2400 had lower performance then 2133MHz
This is @ 2400MHz
[img]https://i.postimg.cc/nc7697j8/timing.png[/img]
This is @ 2600MHz with maunally tweaked third timings, but yeah its not 100% stable, can get bsod 0x101 or 0x124 or just hard reset.
[img]https://i.postimg.cc/TYgR6mVB/FIXED-26000-top.png[/img]
[img]https://i.postimg.cc/15wcGsbx/3.png[/img]
and for some reason, one of the channel readings dissapears, I think when I pass tWTR over certain value. Not sure anymore, I've stopped tweaking..
As you can see RTL is a bit higher here, and this is at auto mode.. I didnt see any real benefit, from what I've read its more of a DDR4 thing., but bandwitdh is what I get above, if I leave at auto it raises tRTR and tWTW too much and it gets slow again.. So I need tRTR 4-6-6 and at least max tWTW 5 (4) -7-7. Mobo auto sets them to 5-8-8 and 6-9-9 or so..
[img]https://i.postimg.cc/Yqgdq13w/memory-2600.png[/img]
If I tweak just tRTR I get ok read, but write and copy suffers then, acts as if I had 1866 or 2133MHz ram not @ 2600MHz.. as seen on this last pic.
Why is this important?
More constant fps, faster response time, basically everything. You're getting practically zero real value out of a very low tWCL with a considerably higher tWR. You need to take syntetic benchmarks or actually anything that tells you otherwise with huge bucket of salt. It simply isn't the case. It's also a huge problem in the event of unexpected power shutdown, timing of reset goes to fast and doesnt get last write done properly, leading to unstable boot of windows over time and possible corrupt files. Alltho Windows has become much better at this over the years, this would still be an issue. A lot of the tertiary timings can prolly be much lower.
I'm getting closer to my end result 3400 preset now and have 1 t_RRDR which is the read timing for dimm ranks, not other dimms (t_rrdd) So its 1-5 for read atm, and testet ok for an hour ish on ramtest.
You're right there's some differences between ddr3/ddr4, obviously, but much of this stuff if not all of it is pretty much the same. Especially with consideration to tWCL/tWR synergy.
You should get RamMon and check what it reports the lowest default specs are there for your kit, providing it's recognized.
If you have an xmp profile, what's the lowest reported tWR on that? I'd start with that as a starting point, unless the case is that it's a very badly designed xmp profile to begin with.
The reason why we don't have perfectly trimmed ram timings out of the box is as you may know that JEDEC and i believe it's Intel dictate certain things which the manufacturers have to objey, which basically means that it's so cheap motherboards used in premade pc's for "i dont have a clue" are guaranteed working ram. At least that's my guess.
edit:
for default 2400mhz
the lowest reported tWR is 13.25ns, when i tested lower then 16 it just crashed, I didn't know there is a special tWCL relationship. min tWCL is 8 I think, can't seem to find it in this RamMon.
Another very important thing is that just because you can't set a timing low enough, doesn't mean it isn't possible, exactly because of dependencies of other timings, voltage, rtl/iol etc.
And it also doesn't mean just because a timing IS deemed to be stable that it is stable under all circumstances. This sounds probably a bit hard to grasp, but if you run ramtest with full cpu cache in effect, and system gets heated, everything is stable. You quit the test, reboot, computer doesnt boot. Now. The timings are still stable. But they are not responsible for taking on a full transistor refresh of power in/out as you had just quickly rebooted the computer after 100% use of the memory, and thus also heavily used cpu and motherboard at the same time.
So what to do? Well the options are either more voltage, or go back up that timing +1 and leave it. (And revisit it another time when u have looked at other timings(important!)) If you add more voltage, you also get the chance maybe u could decrease the timing even more. Until the same scenario repeats itself. And then the question becomes just exactly is enough enough? I've read some old threads on this topic and some weirdos would suggest at 80C dimm temperatures. Which is ludicrious, basically torture the ram chips to death, because it can technically be done. Well I never get my ram chips above 45c since they're watercooled so it is impossible for me to reach such a limit even under the most dire circumstances at extreme loads over many hours.
Anyway, finding lowest tWR makes all of this a lot easier. Try and base it on what primary timings u see fit, and make adjustments to everything else based on primary and tWR.
Another also very important thing here is tRTP which is also in relation to a lot of things, but you have that at 10, 10 should be fine. You want this one to be as low as possible too tho, think of it more as a stability setting, even tho it can have severe impact on getting from 97% performance to the absolute sweetspot of the maximum your config can handle.
tWCL should be in proximity of tCL but not be higher. tWR can be either, it can both be lower, same, and higher as your tCL. Alltho system chipsets make an impact here.
tWCL simply sets the bar for when writes have to be done before tCL refreshes new data in, if that write isn't done and you try to cache in new data, your going to have a bad time, given buffers or the data place is still in use.
The one thing I am not certain about here is how far u can stretch things given cpu cache, but I still think u can get pretty far. Im not familiar with the chip.
edit: Oh and very important, It also means that there is no point of having tWCL too low since your tCL doesnt need till it needs it. All that matters is that the write was ok as your tCL needs it. Just having a super low tWCL is pointless.
For ddr3 you're fine up to 1.8v if u have active cooling on them.
Oh and also, it can also be the case sometimes, rarely, that you dont have some timings actually low enough, because the other timings demand that X be done by now, and if that isnt by Y time passed, it'll fail.
your tBL, block burst length is 8 so i think we're good for the most part on my timings thinking here since it's also 8 for ddr4, however there is a slight difference even here from ddr3 to ddr4 as this is twice as fast "processed" for lack of a better word.
On https://postimg.cc/D8fXFk25 we can see that one of your ram chips simply can't handle it while the other one can, obviously. Also tRAS should be 37 in that one. At lowest 35, at highest 39.This will give you more headroom for your secondary and tertiary timings to breathe, you never want tRAS to be too low. Give your ram some time to act.
edit2, you should have halved your memory there.
I've read this certain ratio by some timings few times, but it always looked a bit too much for me.. and never really understood it correctly.
edit:
this is how it looked like half stable, last time i tested it.
ram was 1.65v still, i tried 1.68-1.7v but wasn't comfortable at that.. Temps should be ~ 40-45C or so.. my old 4x4 2133mhz cruciial balistix elite had temp sensor app and even at max 2666mhz test oc they never went over 45C.
this is with current gskill kit @ 2600
https://i.postimg.cc/yY9GdyrX/FIXED-26000-top.png
vs stock stock vanilla , minus 1T
https://i.postimg.cc/jdZN266z/2400-vanilla.png
That said, if I can kindly ask you, could you please edit values for 2600 in paint or something on this memtweakit 2600 picture above and maybe I could try those numbers first?
And for 2400 so i can tighten them up more :)
I see you have kept tWCL at 9, you should have set it to 10 or on 11, value of tCL and tried to lower tWR after that.
And you can't simply just plot in a whole new set of timings. This requires patience and you try 1 timing after the next. I would like to see if you could get tRRD any lower then 7,. Also tFAW is too high, you should be able to hit 28 tFAW or worst case 36 in there. 24 should be possible if u can lower trrd at all. Usually 4x trrd +4 is ok, it depends entirely on how good your RTL/IOL is.
You should start with tertiary at high timings, secondary as well, then plot in primary and go down from there.
-
But before you do anything you should always make sure your IOL are in check, if possible synced values. Often changing the ram multiplier here will give you a new timing. But I dont know what tools you have in that motherboard to change it.
I would need to look up your motherboard's manual to see exactly what can be done, and yeeee, im not going to ;)