How to Calculate tREFI?

[owikh8464]

As per title, how to actually set DRAM Refresh Intervals (tREFI)? Is there any formula to calculate this?
Thanks in advanced.

[HiVizMan]

tREFI = (Static Refresh ÷ Number of Row Addresses)÷ 2

for example. tREFI (1Gb) = (64ms ÷ 16,368) ÷ 2 = 7.81μs

[owikh8464]

HiVizMan wrote:
tREFI = (Static Refresh ÷ Number of Row Addresses)÷ 2

for example. tREFI (1Gb) = (64ms ÷ 16,368) ÷ 2 = 7.81μs

Thanks for replying.
I'm sorry I don't get you. So let's say if I got a 2 x 2GB Corsair GTX4 clocked to 2400MHz 7-11-7-28-1T as shown in the picture below.

What tREFI should I set? Is 10360 correct?

[HiVizMan]

The minimum interval between refreshes varies according to DRAM ICs and the density of the modules – higher density modules need refreshing more often (lower value for tREFI). Refresh interval requirements also vary according to DRAM temperature (changes in cell leakage rates) and VDIMM. As you can see there are quite a number of factors that you need to include in the permutations. The best thing you could do is experiment and find out for yourself what works on your system and what does not.

http://i4memory.com/f2/ Have a look over there these guys are very into their memory and may be able to walk you through what you want to know I am sure.

[owikh8464]

I think I need someone like Raja or Shamino to answer this.

Thanks anyway.

[johndamon]

I'd really be grateful if some RAM Guru(s) could give us a bit of a lesson on this subject. I would like to know how low one can set this without risking sudden corruption syndrome, or some other horribleness. I now understand that lowering this setting is an easy way to tweak one's MemTweakit score. But I don't know if this would provide a corresponding boost to real world performance. All things RAM fascinate me and since the basic overclocking on Ivy Bridge is so straightforward, I have enjoyed the added challenge of figuring out how to push my RAM as far as possible. Thanks for any info/advice!

[Raja]

Memtweakit is counter-intuitive. A larger value = higher performance. A lower value = less performance. Why? Because during a refresh operation, the bank is inaccessible for read/write transactions. Setting a lower value for tREFI results in more frequent refreshes - you can figure what that means easily enough. Setting a value that is too large for the DRAM will result in data becoming corrupt because the charge in the DRAM cells diminishes over time.

A refresh recharges the cells to either a logical 1or 0, whichever value was written to them in the first place. That's above 50% of VDIMM for a logical 1 and below 50% of VDIMM for a logical 0. If the charge in a cell containing a logical 1 falls below 50% of VDIMM, it would be read as a logical 0, and there's your data corruption.

Overall, you wont get much gain by increasing the default value of tREFI in real-world apps. If you're already at 10360 or if that is default for the DIMMs, I'd focus more on the third timings than tREFI.

[johndamon]

Thank you for filling in some critical gaps in my understanding of the recharging process, Raja. I very much enjoy scoring 60000+ on MemTweakit, but I think I'll raise the tREFI back to 6936 and avoid any annoying disasters.

I am trying to understand the possible reasons that my mixed Kits run so smoothly together... I happen to be one of those very lucky souls who has managed to mix two different RAM Kits with very positive results. In my attempts to get my other MOBO, an MVE, to run more than 1 stick of RAM I foolishly bought several different RAM Kits instead of RMAing the board. So I found myself in possession of over a grand worth of DIMMS. Despite knowing the risks and the small chance of joyful results, I ran through all of the possible pairings and was grateful to discover that my Corsair Dominator Platinums 2666 CL11 (2x4GB), and my G. Skill RipJaws Z 2400 CL10 (2x4GB) got along swimmingly! I have been running them as a 16GB (4x4GB) 2400 9-12-11-31 CR1 Latency Boundary (5) 1.725v Kit for a few months now without any problems. I get Bangin scores in all of the Memory Benchmarks (AIDA64 Extreme Pro, SANDRA, MAXXMem, etc.) and my gaming and A/V performance is beyond awesome! I purposely placed the G. Skill DIMMs in slot A0/B0 so that the MOBO would read the RipJaws Z's 2 XMP3 configurations rather than the DOMs' 1. I am curious if the fact that the G. Skill DIMMS are double the density of the Corsair DIMMS is an important factor to consider when running them together. Bearing in mind the explanation you provided in your Post Raja, do you think that the fact that the tREFI setting affixed by the MOBO to all of the DIMMS based on the RipJawZ XMP3 is part of the reason that these DIMMS are working so well together? Since the Refresh Rate determined by tREFI has been arrived at according to the RipJawZ XMP3 numbers, did I dodge a bullet by placing the RipJawZ in A0? I know that this must be painfully uninteresting to everyone else, but I'm fascinated by the fact that I am able to run mixed Kits at a speed which is quite high for most IB MCs, considering they are of different densities, different ICs, different SPDs, and populating all four slots! What exactly is working in their favor, aside from the fact that they are Binned, high performance DIIMS from two very serious RAM sellers? I guess that I am hoping to figure out what is allowing them to work so well so that I'll have a better chance of getting two of my other Kits to do the same in the second Rig I am currently putting together. Thanks Raja, owikh, and anyone else who has the patience to read through all the babbling! 😮

[johndamon]

BTW owikh8464... Can I ask you how you like your EKWB components? I am going to build my first Loop as soon as Amazon manages to get their hands on my damn Corsair 900D Case, and I am using the updated EKWB H30 360 HFX Kit. I have read both good and bad about EK, though almost all of the bad had to do with complaints about the Nickel Plating wearing off various blocks. Is the EK stuff generally easy or hard to work with? Any tips would be greatly appreciated! Thanks man!