Tertiary timings for F3-2400C10D-16GTX

[LeetMiniWheat]

Are there any spec sheets or official timings listed anywhere for G.Skill F3-2400C10D-16GTX? My Asus board had really loose secondary timings I was able to lower to SPD spec, but I'd like to tweak the tertiary timings too. Screenshot below of my current timings. (10-12-12-31-2N tweaked down to 10-12-11-31-1N stable though) There's no tCKE listed either in SPD but my board defaults to 6, and I can't find where RC is in my BIOS.




Also, any advice on which timings to raise to reach past the 2400MHz barrier? 2600 was not stable with 11-13-13-35-2N even with cpu IMC voltages raised.


AIDA64 reported SPD here:

[emissary42]

tFAW = FAW (aida64) = 26 in the XMP.
tRC or RC is not available on current Intel platforms, it still gets used on AMD though.
What timing does the board default to 6? You stated FAW, but that is very unlikely.

Tertiary timings are not part of the SPD/XMP in the first place. The mainboard determines suitable tertiaries during memory training. You can try to close in on 5-5-5-15-4-4-4-4-4-12-12-12-14 bit by bit and see how far you can get with that. If you get at least halfway there, that should already improve memory efficiency quite a bit.

With those modules you might not be able to hit DDR3-2600 stable regardless, based upon the ICs that are probably used.

[LeetMiniWheat]

tFAW = FAW (aida64) = 26 in the XMP.
tRC or RC is not available on current Intel platforms, it still gets used on AMD though.
What timing does the board default to 6? You stated FAW, but that is very unlikely.

Tertiary timings are not part of the SPD/XMP in the first place. The mainboard determines suitable tertiaries during memory training. You can try to close in on 5-5-5-15-4-4-4-4-4-12-12-12-14 bit by bit and see how far you can get with that. If you get at least halfway there, that should already improve memory efficiency quite a bit.

With those modules you might not be able to hit DDR3-2600 stable regardless, based upon the ICs that are probably used.

Hi sorry, I meant tCKE defaults to 6 and is not listed in SPD. Good to know RC is not relevant to intel. And yeah I know what I was buying, overclocking is never guaranteed. I seem to see 16 tertiary timings, but you list 14. My board is overly conservative with them.

Thanks for the reply

[emissary42]

My sequence was sorted in ASRock Timing Configurator order, because that is what most people use (even with ASUS, Gigabyte or MSI mainboards). It should translate into 5-5-5-15-4-4-4-4-4-DN-14-12-12-12 in MemTweakIt order. Most of the timings should be similarly named in the UEFI.

[LeetMiniWheat]

Thanks, I did some reading and with your suggestions I managed to lower the Tertiary timings. tRDRD at 4 compared to 5 can have a noticeable impact I've heard, but can be finicky above 2400. This might be what's holding me back from going to 2600. (you specified 5, and my board auto'd to 4 which seems to work at 2400)

Anyways here's my new timings so far. tWRRD wasn't bootable at 15 but 16 works. I'm still testing with HCI MemTest. Not entirely sure if tRP is stable yet either. I need to read the datasheets for these 4gbit samsung ICs, I know there has to be some kind of formula... like I know there's a relation between CAS and tWCL or CWL whatever you want to call it.




Next I want to relax some and try for 2600

[emissary42]

The tertiaries look a lot better now. That should have given you a bit more bandwidth as well as a reduction in overall latency. You can still try to reduce tRRD (5/4), tWR* (14/12/10), tRTP (8,7,6) and lower tRAS (28/25/22/21) but that will probably only have negligible impact on overall performance.

*MemTweakIt reports wrong values

Getting to 2600 might not be an issue of relaxing timings, just saying. Some ICs stop scaling past a certain point and Samsung 4Gbit B-die are known for that.

Can you run 10-11-12 / 10-10-12 / 10-11-11 / 10-10-11 @ 2400?

[LeetMiniWheat]

Thanks for the help. Yeah I'm currently at 16 tWR I may try some of your suggestions next. I've been testing lower tREF. Unsure about tRAS though, there's so many formulas people state but doesn't seem to be any real consensus on it, nor does it seem to affect performance much.

Edit: tRCD refuses to go below 12, but 10-12-11 works.

[emissary42]

You might not get much lower with tREF because of the higher density ICs used on 8GB modules that need larger tREF values than their 2Gbit counterparts found on 4GB sticks.

Samsung 4Gbit B-Die usually does CL+2=tRCD+1=tRP or CL+2=tRCD+2=tRP, so you probably got some other revision of 4Gbit Samsung on your TridentX.

[LeetMiniWheat]

well following your formula (if I understand it correctly) my 10-12-11 is your first one (CL+2=tRCD+1=tRP)

can't seem to get any of the main timings any lower and I'm afraid to go lower than 30 tRAS since lower is not always better. working on more secondary timings now since it passed both memtests

[emissary42]

CL+2=tRCD+1=tRP would be 10-11-12 at DDR3-2400 (or 9-10-11 @ 2133).
CL+2=tRCD+2=tRP would be 10-10-12 at DDR3-2400 (or 9-9-11 @ 2133).

[LeetMiniWheat]

I give up, Asus wins. it's auto secondary/tertiary timings are somehow faster than the ones i spent so much time tweaking (and I even managed to get them stable even lower). primary timings won't budge lower than spec 11-12-12-XX though without failing HCI MemTest

Edit: I noticed the board autos to looser secondaries than the SPD specifies but tight tertiaries. If I set secondaries to SPD the tertiary timings loosen. If I set the same tertiary it auto's to along with gskill SPD secondaries, it doesn't boot. perhaps Asus has some magic going on here to optimize better. on auto synthetic benches don't score higher but real world does. being a gaming board, I suppose that makes sense. that said, I think I can still tweak a little, such as tRRD and tREF

[emissary42]

I give up, Asus wins. it's auto secondary/tertiary timings are somehow faster than the ones i spent so much time tweaking

Unless you did something wrong the MB might relax other timings like RTLs and / or IOLs to compensate. Tuning those manually is however a little bit to complex for a drive-by attempt. There should be a tutorial available on this for the Maximus VII models at some of the bigger OC related communities.

How to do that right varies from one model to another and as a Gigabyte Z97X-SOC Force user, i can't really help you with that.

[LeetMiniWheat]

Unless you did something wrong the MB might relax other timings like RTLs and / or IOLs to compensate. Tuning those manually is however a little bit to complex for a drive-by attempt. There should be a tutorial available on this for the Maximus VII models at some of the bigger OC related communities.

How to do that right varies from one model to another and as a Gigabyte Z97X-SOC Force user, i can't really help you with that.

already tried that, strangely enough I can't manually set the RTL/IOLs even with known-good settings and memory training off, MRC check off, etc. refuses to boot.

the two things that seemed to matter most were tRRD and tRFC lowering to 4 (from 7) and tREF to 250 (from 313) I think possibly having tighter tertiary timings and looser secondaries somehow makes a difference. either way, not a huge amount. Also you were right this RAM certainly does hit a brick wall after 2400, I managed to get it to boot at 2600 with MCH check off and extremely loose timings but it failed a quick memtest86+ on test #1 so I gave up on that. I get about 36GB/s-37GB/s read and write with 35GB/s-35.5GB/s copy and 40ns-42ns latency in AIDA64. wPrime benches were very similar but somehow faster with Asus settings and my tRRD/tRFC tweaks. I'm sure it'll go up once I get my cpu OC'd to 4.6GHz again.

Thanks for the help/suggestions, I may have another go at this on another date. working on lowering vDIMM now.