Overclock.net banner

1 - 13 of 13 Posts

·
Registered
Joined
·
28 Posts
Discussion Starter #1
Hi
I am working on a stable 24/7 memory clock. I bought a high end Trident Z 4000c17 17 17 37 2T kit (2x8gb). I had high expectations that i could either push frequency much higher or timings a good bit lower. So far not having a great deal of luck. I presume its samsung Bdie

Im using memtest86 version 8.1 boot from USB in bios which seems to be the most accurate and recommended way of finding errors. Memtest64 in windows does not. Aida 64 benchmark doesnt either.

Ive managed to get 4000mhz c16 16 16 37 2T no errors. i set Dram to 1.49v as im happy to run that daily. VCCCIO and VCCSA to 1.27v.

I can boot windows and run games and stress tests fine with 4133 c16 16 16 36 but memtest86 finds errors so doesnt seem like it would be a good daily clock.

I have tried pushing 4200mhz and over but cant seem to get a boot. even up to ridiculous timings eg 20 20 20 40 2T. Is this just the limit of my memory controller?

Any tips to help me get either higher frequency or lower timings?

Cpu: 9900k 5ghz all core
MB: Maximus XI formula
 

·
Memory Overclocker
Joined
·
468 Posts
Hi
I am working on a stable 24/7 memory clock. I bought a high end Trident Z 4000c17 17 17 37 2T kit (2x8gb). I had high expectations that i could either push frequency much higher or timings a good bit lower. So far not having a great deal of luck. I presume its samsung Bdie

Im using memtest86 version 8.1 boot from USB in bios which seems to be the most accurate and recommended way of finding errors. Memtest64 in windows does not. Aida 64 benchmark doesnt either.

Ive managed to get 4000mhz c16 16 16 37 2T no errors. i set Dram to 1.49v as im happy to run that daily. VCCCIO and VCCSA to 1.27v.

I can boot windows and run games and stress tests fine with 4133 c16 16 16 36 but memtest86 finds errors so doesnt seem like it would be a good daily clock.

I have tried pushing 4200mhz and over but cant seem to get a boot. even up to ridiculous timings eg 20 20 20 40 2T. Is this just the limit of my memory controller?

Any tips to help me get either higher frequency or lower timings?

Cpu: 9900k 5ghz all core
MB: Maximus XI formula
I have Samsung B-die TridentZ 2x8 sticks (XMP rated for [email protected]) on a ASRock Z390 Taichi with i7-9700k. I can do [email protected] with very tight secondary and tertiary timings at DRAMV 1.45V and VCCIO and VCCSA at 1.18V. I can do [email protected] at DRAMV 1.50V and VCCIO and VCCSA at 1.25V.

Both configurations are stable with RAMTest to 20,000% (which is an overnight memory test) and with PassMark's MemTest86 (booted from USB) for 8 passes, all tests, all CPUs (not running in parallel). It was definitely harder to get stability at 4133 than at 4000 and, in fact, my AIDA64 memory benchmark numbers are better at 4000 so that's where I'm running for 24/7. I could not find any configuration that would boot at 4200. I suspect it might run at 4200, but the motherboard doesn't train properly at 4200 to get things started (it has a hard time training at 4133 too). I may try again sometime with more manual settings (just for fun).

Going above the XMP rating on your sticks is just silicon lottery and your motherboard's ability to figure out secondary and tertiary timings that will work. Sometimes you can go way above what the sticks are rated at and sometimes only a bit higher - it all depends upon what sort of chips you got. With memory sticks, it only takes one memory chip out of all the ones on a stick that's holding things back to hold the whole set back. You can nearly always improve benchmark numbers by tightening secondary and tertiary timings, but that is a real time sink activity because you've got to run long stability tests over and over again to find how far you can go on each of the important timing numbers.
 

·
Registered
Joined
·
28 Posts
Discussion Starter #3
Hi
I am working on a stable 24/7 memory clock. I bought a high end Trident Z 4000c17 17 17 37 2T kit (2x8gb). I had high expectations that i could either push frequency much higher or timings a good bit lower. So far not having a great deal of luck. I presume its samsung Bdie

Im using memtest86 version 8.1 boot from USB in bios which seems to be the most accurate and recommended way of finding errors. Memtest64 in windows does not. Aida 64 benchmark doesnt either.

Ive managed to get 4000mhz c16 16 16 37 2T no errors. i set Dram to 1.49v as im happy to run that daily. VCCCIO and VCCSA to 1.27v.

I can boot windows and run games and stress tests fine with 4133 c16 16 16 36 but memtest86 finds errors so doesnt seem like it would be a good daily clock.

I have tried pushing 4200mhz and over but cant seem to get a boot. even up to ridiculous timings eg 20 20 20 40 2T. Is this just the limit of my memory controller?

Any tips to help me get either higher frequency or lower timings?

Cpu: 9900k 5ghz all core
MB: Maximus XI formula
I have Samsung B-die TridentZ 2x8 sticks (XMP rated for [email protected]) on a ASRock Z390 Taichi with i7-9700k. I can do [email protected] with very tight secondary and tertiary timings at DRAMV 1.45V and VCCIO and VCCSA at 1.18V. I can do [email protected] at DRAMV 1.50V and VCCIO and VCCSA at 1.25V.

Both configurations are stable with RAMTest to 20,000% (which is an overnight memory test) and with PassMark's MemTest86 (booted from USB) for 8 passes, all tests, all CPUs (not running in parallel). It was definitely harder to get stability at 4133 than at 4000 and, in fact, my AIDA64 memory benchmark numbers are better at 4000 so that's where I'm running for 24/7. I could not find any configuration that would boot at 4200. I suspect it might run at 4200, but the motherboard doesn't train properly at 4200 to get things started (it has a hard time training at 4133 too). I may try again sometime with more manual settings (just for fun).

Going above the XMP rating on your sticks is just silicon lottery and your motherboard's ability to figure out secondary and tertiary timings that will work. Sometimes you can go way above what the sticks are rated at and sometimes only a bit higher - it all depends upon what sort of chips you got. With memory sticks, it only takes one memory chip out of all the ones on a stick that's holding things back to hold the whole set back. You can nearly always improve benchmark numbers by tightening secondary and tertiary timings, but that is a real time sink activity because you've got to run long stability tests over and over again to find how far you can go on each of the important timing numbers.
Thanks for the info. Make a me feel a bit better you are stuck around the same numbers as me.
 

·
Registered
Joined
·
1,010 Posts
I believe it is mostly a motherboard limit. On top of this there is a silicon lottery among b-die chips (and this particular kit you got is known to be petty good).
You are at limits for daily of both vdimm, vccio and vssa, so I would eventually play only with subtimings to squeeze a little more in benchmarks and 0,something% in real world applications then call it a day :)
 

·
Registered
Joined
·
28 Posts
Discussion Starter #5
I believe it is mostly a motherboard limit. On top of this there is a silicon lottery among b-die chips (and this particular kit you got is known to be petty good).
You are at limits for daily of both vdimm, vccio and vssa, so I would eventually play only with subtimings to squeeze a little more in benchmarks and 0,something% in real world applications then call it a day /forum/images/smilies/smile.gif
Ok thanks. Yeh I think I will tighten trefi and tras and leave it at that for daily
 

·
Premium Member
Joined
·
632 Posts
It's not about the limit of the memory controller here. Asus's advertised 4400MHz+ memory on 4 DIMM Maximus boards refer to a QVL entry with all 4 DIMM slots populated since these boards are on the T topology where 4 DIMMs will clock better than 2.

I do recall 4133MHz being the end of potential frequency scaling on these on 2 sticks.

MSI's Z390 boards are running daisy chain topology however, you'd have better 2 DIMM clocking headroom on those.
 

·
Registered
Joined
·
1,010 Posts
It's not about the limit of the memory controller here.
With "limits" I meant to not ruin the cpu. Personally, I wouldnt go above 1,25 for long term use.
It has been said several times that 9xxx cpu have a great IMC and the limits are either the mainboard or the dam dies themselves.
 

·
Memory Overclocker
Joined
·
468 Posts
It's not about the limit of the memory controller here. Asus's advertised 4400MHz+ memory on 4 DIMM Maximus boards refer to a QVL entry with all 4 DIMM slots populated since these boards are on the T topology where 4 DIMMs will clock better than 2.

I do recall 4133MHz being the end of potential frequency scaling on these on 2 sticks.

MSI's Z390 boards are running daisy chain topology however, you'd have better 2 DIMM clocking headroom on those.
T-Topology does not mean that it will clock 4 sticks better than 2. It means it may clock 4 sticks better than a daisy chain motherboard will clock 4 sticks and in fact, it may not clock 2 sticks as well as a daisy-chain board will clock 2 sticks.

That's the idea behind T-topology, but I've also read a number of very technical articles that tout how daisy-chain can be tweaked (with timings) to be better for 2 sticks and just as good for 4 sticks vs. t-topology so some articles say it is more about the actual details of the implementation than it is about which topology is chosen.
 

·
Premium Member
Joined
·
632 Posts
Double.
 

·
Premium Member
Joined
·
632 Posts

·
Memory Overclocker
Joined
·
468 Posts
Is there any English in those videos? I went to the 3 minute mark in each video per your suggestion and did not find any language I can understand. Perhaps some text describing what the point here is would be useful. Far too many link-only posts these days that don't offer any context for what this is about or even quote what they are responding to.
 

·
Premium Member
Joined
·
632 Posts
There are english captions for the first video, I referred to 3:00 onwards for the specific portion on memory topology.

Video #2, I've paraphrased the content on the comment #6 and that both QVLs of the Godlike Gaming and Extreme are referenced (T vs. Daisy chain on 2 DIMMs)
Toppc goes on to mention On-Die Termination's (of DDR4 ICs) in relation to T and Daisy chain layouts and that both slots on T need to be populated for extended OC margin for a "perfect signal" since there's a trace related to ODT on both A1 A2 and B1 B2 slots. That's unlike the daisy chain topology.

Video #3, just a video on "Dragon Alliance"... A couple of DRAM models with a stable IC supply (e.g. Hynix-A only, Hynix-C only) are picked from RAM vendors for demonstrating the benefits of daisy chain in a 2 DIMM config, MRC tuning were done on a number of these kits for an MSI certified auto overclock beyond XMP rating. e.g. A Corsair XMP-3600 kit with a designated IC running 4000MHz instead.

P19 onwards
https://download.msi.com/archive/mnu_exe/pdf/MSI-Z390-Overclocking-Guide.pdf


Video #3 is related to another topic in his motherboard series on DRAM sorting from "module houses" (corsair, gskill etc) and how were the QVLs built on 4/6/8 PCB layer T-top 2DPC or 1DPC QVLs etc. Regardless, both Overall, Toppc and Raja referred to 4 DIMM being the ideal config for a T-topology motherboard for an extended OC margin.

Daisy chain on Ryzen: Crosshair VII Hero, on an article authored by The Stilt
https://www.io-tech.fi/artikkelit/testissa-amd-x470-emolevyt-asus-gigabyte-msi/


Elmor's C7H PDF extract on memory OC margin for this particular daisy chain board
 

·
Registered
Joined
·
56 Posts
Hello, I have a kit similar to yours : 4x8 GB TridentZ 4000 c18 19 19 39.

I tried to change the 4 main timings to 17-17-17-37 and a few other ones but my system is not 100 % stable with these timings, because every timing must be changed. Consequently, may you give me your default values for the following timings, please (on Gigabyte Z390 Aorus Pro) :

t CCD_S
t CCD_L
t REFIx9
t XP
t XPDLL
t PRPDEN
t RDPDEN
t WRPDEN
t CPDED
t AONPD
RFR delay
t XSDLL
t XS offset
t ZQOper
t MOD
round trip latency (Dimm0/rank0)
round trip latency (Dimm0/rank1)
round trip latency (Dimm1/rank0)
round trip latency (Dimm1/rank1)
IoLatR0D0
IoLatR1D0
IoLatR0D1
IoLatR1D1
Rtt Nom (dimm0)
Rtt Nom (dimm1)
Rtt Park (dimm0)
Rtt Park (dimm1)
Rtt Wr (dimm0)
Rtt Wr (dimm1)

Thank you.
 
1 - 13 of 13 Posts
Top