Set of timings for DDR3 Elpida Hyper

[unityofsaints]

On 1/19/2019 at 12:12 AM, TaPaKaH said:

On tRCD - 1020 hardly tells you anything. tRCD 6 would be extremely impressive (only Hicookie's GTX2 were capable of such a feat), tRCD 7 is standard on 90-95% of sticks. If tRCD 8 is necessary at 1020 then kit is garbage.

But you are comparing LN2 RAM vs. ambient here?  On LN2, both Luumi and Cookie's have 32M runs at those settings or above.

[TaPaKaH]

I don't know why Luumi ran tRCD 8, usually this is not necessary even on air.

[unityofsaints]

17 hours ago, TaPaKaH said:

I don't know why Luumi ran tRCD 8, usually this is not necessary even on air.

More likely to be an IMC rather than a stick limitation. @Luumi, do you remember?

[ground]

did a quick test on my kit at 2.1V, they keep on scaling, 6-6-5 at 976 is doable at that.

Might manage 6-6-5 at 986 (it did manage half a run before I got too greedy and went to 1000, but it doesn't scale anymore with voltage from here (sticks might be getting too hot too, though I have plenty of airflow right now)

 

Next steps are  -optimising subtimings (still using the same as at 1107) -trying if it goes any further at 6-7-5/6-7-6/6-6-6

[ObscureParadox]

26 minutes ago, ground1556 said:

did a quick test on my kit at 2.1V, they keep on scaling, 6-6-5 at 976 is doable at that.

Might manage 6-6-5 at 986 (it did manage half a run before I got too greedy and went to 1000, but it doesn't scale anymore with voltage from here (sticks might be getting too hot too, though I have plenty of airflow right now)

 

Next steps are  -optimising subtimings (still using the same as at 1107) -trying if it goes any further at 6-7-5/6-7-6/6-6-6

I probably wouldn't advise running hypers at that kinda voltage for too long, it's not going to do the longevity of them any good.

[ground]

11 hours ago, ObscureParadox said:

I probably wouldn't advise running hypers at that kinda voltage for too long, it's not going to do the longevity of them any good.

Its a kit as bad as they come, planning on getting a new one asap. For most of my benching I’ll keep them below 1.95 most likely. Haven’t confirmed, but I’m almost sure they can’t do spec.

 

...they do spec, and now that they have taken 2.1V for maybe an hour of testing they suddenly work much better then before, currently running 1050 7-7-6 at 1.775V, 1000 7-7-6 seems to be good for 32m at 1.68V and 1100 7-7-6 at 1.85V...

Edited February 9, 2019 by ground1556

[ground]

Even bad hypers like running a little colder  1002 MHz tRCD 6

(ignore subtimings & time - I was mainly testing the basic effect of running them at around 0°C (pair of peltiers cooling the RAM under a RAM block)

[M1RR0R]

18 hours ago, ground1556 said:

Even bad hypers like running a little colder  1002 MHz tRCD 6

 (ignore subtimings & time - I was mainly testing the basic effect of running them at around 0°C (pair of peltiers cooling the RAM under a RAM block)

can you try this mem under SS/ln2? 

[ground]

2 hours ago, M1RR0R said:

can you try this mem under SS/ln2? 

If I get around buying a memory pot I’ll most likely run it on dry ice. Will also try some more mounting pressure on the peltiers, maybe I can get to a little lower with this setup already. I also have a feeling I degraded the imc a bit with the high memory voltage, need to grab one of my remaining high memory clock capable chips for more testing...

[Noxinite]

I tried hypers on SS ages back, did 2000 6-6-4 @2V in 32M on Z97. Those sticks were ~1.66V 2000C7 I think.

[M1RR0R]

9 hours ago, Noxinite said:

did 2000 6-6-4 @2V in 32M on Z97

what about CL7? 2400+ with 7-7-7/7-8-7 is possible? 

[GtiJason]

On 2/26/2019 at 1:19 AM, M1RR0R said:

what about CL7? 2400+ with 7-7-7/7-8-7 is possible? 

Pretty sure I've only seen l0ud_sil3nc3 pull this off. At least at ambient temps that is

[websmile]

Hiwa ran G.Skill Perfect Storm Hypers on cold 2600 7-7-7 lots of years ago iirc (Ivy Bridge), but I am not sure this makes sense on most newer platforms

[Guest]

@websmile Do you have any links to that?

[websmile]

No, maybe you find it at xs where he posted that

[GtiJason]

12 hours ago, websmile said:

No, maybe you find it at xs where he posted that

Thanks, I went to check and 5 hours later I made it back to this thread. Never send a ddr3 guy to xtreme pi alone, very easy to get lost in there reading about XDZ, XDX, XDY, XDT, XDR, XEA, XEB, XED, XEE, XEF, XEJ, XEV, XEX, XEY, XFD and XFE to name a few.

Only Ivy era run with hypers i saw was. . .  http://www.xtremesystems.org/forums/showthread.php?280903-SuperPi32m-5Ghz-Ivy-All-Out-Challenge!&p=5103700&viewfull=1#post5103700

And only hiwa runs were 2x2gb Trident BBSE and 4x2gb Trident PSC

http://www.xtremesystems.org/forums/showthread.php?280903-SuperPi32m-5Ghz-Ivy-All-Out-Challenge!&p=5177133&viewfull=1#post5177133

http://www.xtremesystems.org/forums/showthread.php?280903-SuperPi32m-5Ghz-Ivy-All-Out-Challenge!&p=5172912&viewfull=1#post5172912

[websmile]

No one posted decent results for the list bc Hypers are too slow for Ivy and haswell - I made a fast check myself and found that Hiwa ran only 1m at single stage mems -40 at 2600 7-8-7 with his Perfect Storm. Maybe GTX2 for example with more cold can do better, he mentioned that stability at divider level ended 2400 for him, which is not much considering that even GTR can do 2400c7 32m on cold

[GtiJason]

4 minutes ago, websmile said:

No one posted decent results for the list bc Hypers are too slow for Ivy and haswell - I made a fast check myself and found that Hiwa ran only 1m at single stage mems -40 at 2600 7-8-7 with his Perfect Storm. Maybe GTX2 for example with more cold can do better, he mentioned that stability at divider level ended 2400 for him, which is not much considering that even GTR can do 2400c7 32m on cold

I wasn't using the list as reference tho. Went through 20 pages of "hiwa posts". lol. What would you guess  best time would be for hypers on Haswell/DC at any popular cpu speed. ie 4.0/4.5/5.0 or even 4.7 uh hum Div VII. Looking for a time to shoot for, maybe sub 5.59.000 5g?

[websmile]

I am quite sure George already is around best you can get on Hypers at 1150 at 4G. Sub 6 should work for sure , if 5,59 sub I am unsure. Tbh, this is much more of a theoretical question as Hypers are pretty much frequency limited nd their reign ends with Sandy Bridge latest

[ground]

Has anyone here had Hypers under Dry ice? Any issues with coldbug? 

[TaPaKaH]

No coldbug up until -40 at least. MHz/volt doesn't improve much. tRCD drops by about 1 value (but not lower than CAS). Voltage scaling stops later.

[Chilli-Man]

I haven't had much of a play yet, but mine hit coldbug at -60

[ground]

Keep the contacts on the sticks clean, I thought one of my sticks had degraded badly, cleaned the contacts and it went from 1025 7-7-6 at 1.85V to 1100 7-7-6 at the same voltage, now testing less voltage (tRCD wall at 1100)

Edited March 14, 2019 by ground1556

[ground]

Oh, quick note too - if your sticks are dead/badly degraded the trick from d9 times of throwing them in the freezer over night or longer might work. Got 3 (out of 4, last remains unable to boot) 2000 c8 dimms that didn’t manage XMP, JEDEC boot or post at all before back to where they belong around 2000 7-7-6 at 1.66V. 

[ground]

I know there are way too few kits available for purchase for larger scale binning, but I would still like to know what observations others made about binning Hypers around here. I‘ve collected about a dozen sticks of 2000c8s from Corsair/Kingston over the past half year (and talked to some others who also picked up a couple), and noticed a couple things:

(testing done on x58)

- most 2000c8 could manage 2000 7-7-6 32m at <1.7V

- 2200 7-7-6 seems to be close to impossible on non Corsair sticks. Not sure if they binned differently or if its the pcb. Triple kits seemed to do better

- the bin is usually fairly consistant for voltage at 2000 7-7-6 (1.64-1.68V for the sticks that did alright)

- one kit of Dom gt performed particularly bad; pretty sure it was degraded badly by its previous owner. The best stick barely managed posting 2100 7-7-6 at 1.9V, couldn‘t do spec at 1.65V.

- 6-7-5 and 7-8-6 max MHz were fairly consistant (except for the kit mentioned above) across all sticks, anywhere from 2060-2080/2280-2340

- no noticable voltage requirement difference between 7-7-6/7-7-7 or 6-6-5/6-6-6 or 6-7-5/6-7-6

Can anyone see anything that appears weird / did anyone notice similar stuff / completely different behavior? Anything interesting to add? And before anyone asks, I‘ve already sold the kits I don‘t want to keep, nothing for sale.