Du verwendest einen veralteten Browser. Es ist möglich, dass diese oder andere Websites nicht korrekt angezeigt werden. Du solltest ein Upgrade durchführen oder einen alternativen Browser verwenden.
ICs kannst du mit Typhoonburner auslesen. Da du ein 3200MHz kit gekauft hast, schätze ich mal, dass 2933-3200Mhz auch möglich sind. kommt ein bisschen auf dein MB an.
also du kannst schon 1,4V machen, das ist nicht das Ding. 1,45V musst du schauen wie es mit der stabilität durch gehäusetemp und luftstrom aussieht. Dein RAM zieht keine 5W pro Riegel. Das ist vollkommen ungefählich. Zumal du mit dem 1700 schon viel früher limitiert bist. Spannung hoch und schauen wie weit du die Frequenz bekommst. evt. mal die timings etwas hoch so 18-22-22 und dann ziehst du die runter und dann die Spannung auch wieder runter.
Im C-Die Thread zeichnet sich ab, dass die sehr Temperaturempfindlich sind und daher nicht gut auf Spannungserhöhung reagieren.
Ein Lüfter auf die RAMs gerichtet scheint abhilfe zu bringen.
Hi guys!
First of all I would like to apologize for using English. I understand what you are talking about but can't use German for explain myself. I learn a lot from you and like this conference and hardware thread in particular, that's why I come to you with my question. Thanks.
I got 2x16Gb of CMK32GX4M2B3200C16R 4.32 with Samsung C-die onboard and it runs stable at 1.36v3400@16-19-36CR1 on my Maximus XI Hero. I can start it higher to the point of 3600@18-20-20-38 but I was not able to commit any of results with tests.
Recently I've got similar HyperX Fury kit as Olan (HX432C16FB3K4 / 32) unless I have kit of 4 single rank modules. I bought it relatively cheap with one goal in my mind - to get 3600 or similar on my motherboard, as far as I found that T-topology allow you to get higher results using all DRAM slots. But I was surprised that with new Kingston kit I can't even able to get 3300. In point of fact, it get unstable even if I reset BIOS to defaults, set Manual overclocking profile and copy some or all timings using Asrock Timings Configurator from stable XMP run on 1.35v3200c16.
I also tried to use Rysen DRAM Calculator with Samsung OEM layout, tried to play with timings/subtimings, DRAM voltage and VCCIO/VCCSA in range of 1.05-1.35, with RTL IOL block, and it did not bring any results for me in matter of 3300+.
I know that the most likely problem is that I'm doing something wrong. Thank you again.
Try 1,45V DRAM 18 22 22 42 3600mhz. If that is helpfull you can lower timings and voltage to your liking. I would try to get a higher frequenzy and lower timings first, before you lower voltage
Beitrag automatisch zusammengeführt:
The hero indeed overclocks better with 4 than 2 DIMMs but you only really need that at 4000MHz and beyond
I tried, but no joy. It frequently unable to boot and often throws "7F" Q-code. Sometimes OK(A0) but never boot to OS due to errors. I also added up to 1.48v and tried at 3600 and 3500.
BIOS was reset to default, all settings was in auto except primary timings, voltage and clock. Is there a chance that AUTOs is irreparably wrong on Hero?
Maybe I should swap modules? Or try it with one pair instead of two?
Several fails and finally it's worked after I changed Command rate from Auto to 2N, lol. Looks to me "Memory_OK" jumper is just for wasting time.
Now there is something to play with.
That's a good news since according to Rysen DRAM Calculator, Hynix CJR can achieve 3800 while Samsung OEM is only 3466 on a safe preset
Beitrag automatisch zusammengeführt:
I can't believe it but there are 4000 for now with @Maurise preset, and seemingly stable. I'm afraid to raise further to be honest
The question is why that can't be achieved on 2 pairs.
@pau1er dont be afraid of higher clocks. If you cant get it up a secound time its not stable anyways. I would say you top out at 4200MHz so or so, because thats whay CJRs max is more or less and with only 2 DIMMs its also the MB that cant get higher. I would try to get 4133 running.
Do you testet both and they go up that high or only one? If the ICs are indeed the same they can still have a huge quality difference with a bin that low quality. 3200 CL 16-18-18 is not the worst but the worst of the usable stuff (3000MHz and up. under that you get the same ICs or something terrible that cant OC much higher in the first place). I remember my HyperX 3333 16-18-18 would not even clock over 3400.
If you still want 32GB i would test with a 3600/3733Mhz kita if you dont want to swap out a lot of kits until you find one that can do 4000. HX437C19FB3K2/16
Beitrag automatisch zusammengeführt:
PVS416G400C9K or you get a pair of these. Best in class by far. Samsung B-Die should do about 4266/4300 17-18-18 1,5V or better.
you should also disable fast boot (in the lowest section of the dram timings) and enable Full MCH Check (below fast boot) for better training. the Mem OK jumper should be off.
you should also disable fast boot (in the lowest section of the dram timings) and enable Full MCH Check (below fast boot) for better training. the Mem OK jumper should be off.
Thank you for your advices.
I can't get stable boot to OS with c18-22-42cr2 4100v1.45 and want to stay at safe voltage on the other hand, so I decided to play around 4000 testing only first RAM pair. My current problem: after 5-8 minutes of bench errors arising and if you give it some rest you can repeat test with similar behaviour. Looks like 1.45v is not operable for first pair even with fan. This suitable for all layouts I managed to test, i.e. 4000c18-22-42 and 3600c16-20-32. Current power draw is 4.770W.
there is no "save" voltage for DDR 4. other than 1.2V you can go to 1.5V if you like. and/or you make your timings loser. 20-24-24-44. 1.5 is the max i recommend für under 4400MHz because you dont get a benefit from more in my experience. given you have decent enough RAM that scales like most other. You dont need voltage for higher frequenzy. you need it to keep the timings down. at least where you are. 4400 i need more than 1.5V to even boot reliable. under that even 1.35 is fine. as you see your RAM doesnt take much watts. if you cant keep it cool, than you have too much voltage. other than that even 1.65V is fine. at least for modern ICs. there are older ones that wont even boot with 1.45V and die at higher. but yours should be more than fine with up to 1.5V
Beitrag automatisch zusammengeführt:
i have not seen a kit that was damaged by to much voltage alone. if you short it yea. of you damage it physicly yea. if you have it at over 70°C possible. you will damage your IMC in the CPU befor the RAM. because that takes most of the watts from the dramvoltage rail.
@Maurise Why ppl tend to raise clocks then, if sweetspot is between timing tension? Is it not possible to find best output between 2133 and 3000 for examlpe?
Beitrag automatisch zusammengeführt:
I can assume, higher clock expanding output tuning window?
sweetspot is the wrong term. your RAM is build to be at 1.2V 2133 or 2400 or 2666 or 2933 or 3200Mhz with timings like 19-26-26-46. its called the JEDEC standard and samsung, sk hynix, micron and nanya make the RAM for that specification. if you dont do OC or XMP, thats your JEDEC Profil. G.Skill, Corsair, Kingston HyperX, Crucial, ADATA, Patriot and so on, they just buy a truck load of chips from the 4 manufacturers and look what happens if they do 1.35/1.4/1.45/1.5/1.55V with them and how low the Timings go. then they sort them and sell them. if you buy the good shit, they will have a custom PCB thats better than what the standard is. crucial balistix max, the 3600 and 4000mhz bin have an other pcb then their lower end stuff.
So why 1.2V? because servers. your and my and every ones PC is a drop in the ocean comparred to what servers need. and they want 1 thing above all else: reliablility and density. they dont need 4000MHz. because if you want more speed you just make a motherboard with 2 more channel. you can even increase the storage that way. the secound thing is that you RAM is not really stable. even at 1.2V, you will have 1 error for 4GB per week run time. so you use ECC. also you want a DIMM thats bigger than 32GB. then you use load reduced dimms. they have a contoller so that the IMC is not hit that hard. and so on and so on. that stuff does not OC well. and why would you even try. its not testet. a standard is a standard for a reason.
Beitrag automatisch zusammengeführt:
also, frequency is better for performance than super low timings. if you have a given voltage, your RAM will have an increase of timings very regular. if you actually would calculate the real delay in nano secounds it should about stay the same. thats because your transistors in the IC only can run so fast for a given voltage. https://docs.google.com/spreadsheets/d/1rqnHgPrG059fpOjArpc_nFSDVG1dsjw6y3o9ujf4hro/edit#gid=0 about that. why run faster than? because that only for your main timings. the subtimings can go to abaut what your JEDEC is sometimes. and they dont increase with frequency that much. so you still gain performance. thats why its almost non benefical to go out and buy RAM thats faster than 3600-3800Mhz for XMP. you have to lower your subtimings because your MB just dosnt know whats stable and what not. and the memory is much to smal to have settings for all ICs.
you can run 4000MHz and above because why not. you did not buy it for a lot of money. i bought a 4133 17-17-17 1.4V because i did not wanted to buy 10 low end kits like 3200 14-14-14 and look for the best. and i must say my kit is Ok. its not the best but far from the worst. the only XMP profile my kit cant do is 4000 15-16-16 1.5V and 4400 17-18-18 1.5V. the later probably is my MB, but hey. still very good.
I thought server and customer standards goes it own ways when it is possible, like ATX and EPS when we comes to power supply, or ECC and non-ECC with RAM. But I've got you.
So it happens that way to overclock specific pair of DRAM is
1) to determine their best current latencies in nanoseconds
2) match it to the other clock
3) deal with related problems like overheating
4) ???
5) profit
Sounds good
3200c16 = 10ns = 3800c19. Let's try
dont forget the other 2. if you dont have sambsung B-Die 18 in your 16-18-18-38 is a timing(s) that climbs with frequency. the other you must test how far you get it down. lowest ist 28. but 32-34 is pretty good.
If I able to run 4100c18 according to that table, 3600c16 shouldn't be a problem in matter to lower voltage? What timings I have to adjust here at your opinion?
