Samsung DDR3 G-die

[redux]

I have a whole tray of this stuff...maybe I should test it sometime ?

 

Edit: two trays!

Edited May 20, 2021 by redux

[SparkysAdventure]

please do

[kicsipapucs]

On 5/16/2021 at 8:17 AM, ivaneza said:

Nice finding, any pics of them especially IC mark?

Here you go:

[Noxinite]

Tried LN2 again, and once more almost zero scaling.

Ended up testing C8 max 32M freq (set RCD, RP, RFC, RDWR all high) and got:

[kicsipapucs]

I thought I'd give full dimm X79 quad channel a go with Sandy-E, but so far not so good, 2400 loose doesn't want to train with 8 dimms at all with latest bios, 4 dimms work very well on the other hand (almost managed to get 2666 full tight to pass geek3)

Edited August 7, 2021 by kicsipapucs

[kicsipapucs]

On 8/4/2021 at 6:05 PM, kicsipapucs said:

I thought I'd give full dimm X79 quad channel a go with Sandy-E, but so far not so good, 2400 loose doesn't want to train with 8 dimms at all with latest bios, 4 dimms work very well on the other hand (almost managed to get 2600 full tight to pass geek3)

Okay, so there is some progress.

With Rampage IV Extreme you can set a maximum of 2.100V of vdimm in bios, but the board doesn't increase vdimm above 1.925V, no matter what. 1.920V set vdimm is the highest it reacts to, anything above that give the same output (with LN2 Mode and Extreme OV both enabled too).

I managed to validate 2797.4MHz with single to quad channel, every channel individually. More loose timings didn't help, more or less IMC volts either spent many hours without luck, same exact spot always insta freezes no matter what. Temp or platform limit I guess, still got overall Sandy freq valid #1:

I also tried Geek3 and I got 2666C9-12-13-20-1T tight subs to pass in quad channel, got me a lovely 8922 multi mem score 2c2t, highest I have seen with Sandy(-E).

Full core 5G run:

AIDA gave me this:

80GB+ AIDA write is possible and so it 9k+ geek mem, but I need to either vmod the board or a Rampage IV Black Edition (that can set up to 2.400V vdimm in bios, hopefully not capped at same 1.925V real as regular R4E) for more volts and/or cold CPU for maxing the IMC if it scales.

Edited July 11, 2022 by kicsipapucs

[kicsipapucs]

I purchased a Rampage IV Black Edition, in YouTube reviews I have seen that its bios can go up to 2.400V vdimm. I am not sure if it has the same behavior as regular R4E (as in not applying volts above 1.920V), hopefully not.

Should come within a week and then I'll aim for 2700C8-11-12-13-1T?

Edited August 17, 2021 by kicsipapucs

[SparkysAdventure]

JEDEC bins

Voltage:

C__ is 1.5v

Y__ is 1.35v

Speed:

_F8 is 1066 7-7-7

_H9 is 1333 9-9-9

_K0 is 1600 11-11-11

_MA is 1866 13-13-13

I know these all exist in the dual rank, ecc unbuffered udimms, and should also exist in other formats. There may exist an 800 "F7" or 2133 "NB" variant, but I have not seen any documentation of either.

Edited April 2, 2022 by Sparky's__Adventure

[SparkysAdventure]

Managed to obtain 89 (or was it 87?) sticks of M391B2873GB0-YH9. Binning data coming Soon™.

If I can find a financially viable way to get these ICs swapped to a better PCB, I absolutely will. That'll come after binning, though.

I'll need to keep tRCD, tRP, and tCL binning all separate.

Edited March 24, 2022 by Sparky's__Adventure

[Mahameru]

12 hours ago, Sparky's__Adventure said:

Managed to obtain 89 sticks of M391B2873GBO-YH9. Binning data coming Soon™.

If I can find a financially viable way to get these ICs swapped to a better PCB, I absolutely will. That'll come after binning, though.

I'll need to keep tRCD, tRP, and tCL binning all separate.

Charles, can you post pics of label and IC mark of them? 

[redux]

Finally began binning this stuff this month, so watch this space . So far my voltage findings agree with @kicsipapucs , hardly guaranteed to get 1400MHz CL9 bench settings to run at ≤2V on a random stick, but still much more predictable than 2Gb D for high-frequency tests. My best sticks are all from the same sticker date (though I haven't checked IC batch codes).

Once done I should have good CL8 and CL9 data on ~150 sticks (125 SR / 25 DR, but still!). Will also mess around with validation when I get the chance.

[Noxinite]

SR. ?

Although, tbf, it runs quite nice on the LN2 board. There might even be some decent effi at 3333C11 or something.

[SparkysAdventure]

On 1/31/2023 at 4:00 PM, Noxinite said:

SR. ?

You rang?

Nearly 100 sticks to bin myself

[Noxinite]

You said that a year ago.

I've barely seen anyone actually post any runs/scores with g-die. Is it just the "thing to do" to tell everyone you are binning it so that you seem to be on trend?? XD

[redux]

4 hours ago, Noxinite said:

 Is it just the "thing to do" to tell everyone you are binning it so that you seem to be on trend?? XD

Apparently, owning G-die is like a status symbol or something

I for one am nearly done binning it and I'm hoping to do some proper dual channel tuning this month. Unfortunately most of my dual rank sticks are pretty shit, but we'll see what I can do.

[Noxinite]

On 6/13/2023 at 12:33 AM, redux said:

Apparently, owning G-die is like a status symbol or something

I for one am nearly done binning it and I'm hoping to do some proper dual channel tuning this month. Unfortunately most of my dual rank sticks are pretty shit, but we'll see what I can do.

Define bunnyextraction? No 2800C9 under 2V? Are you binning with tight/loose IOLs/RTLs? What PCB is this?

[l0ud_sil3nc3]

What is the fastest G-Die 32M posted guys?

[SparkysAdventure]

On 6/12/2023 at 1:32 PM, Noxinite said:

You said that a year ago.

I've barely seen anyone actually post any runs/scores with g-die. Is it just the "thing to do" to tell everyone you are binning it so that you seem to be on trend?? XD

College has kept me busy and put me nearly 2 years behind where I intended on being, I have some catching up to do.

6 hours ago, l0ud_sil3nc3 said:

What is the fastest G-Die 32M posted guys?

By raw freq, freq at timings, efficiency at freq, or raw lowest time?

Edited June 16, 2023 by Sparky's__Adventure

[l0ud_sil3nc3]

11 hours ago, Sparky's__Adventure said:

College has kept me busy and put me nearly 2 years behind where I intended on being, I have some catching up to do.

By raw freq, freq at timings, efficiency at freq, or raw lowest time?

Fastest 4/5Ghz time, you know I don't care about loosey goosey ?

[Noxinite]

11 minutes ago, l0ud_sil3nc3 said:

Fastest 4/5Ghz time, you know I don't care about loosey goosey ?

Not seen anyone else post any? So I guess by default (4GHz): https://hwbot.org/submission/4070228

 

Also, cross-post from Z97 LN2 thread as I was looking for this earlier and counldn't find (I think bad OS here, also pentium).

Edited June 16, 2023 by Noxinite

[SparkysAdventure]

3 hours ago, l0ud_sil3nc3 said:

Fastest 4/5Ghz time, you know I don't care about loosey goosey ?

If you're willing to consider theoretics for 5g:

The best S2D result documented was with ambient memory and a Z97 Impact. Alex's 3k 9-12-12 was in the mid-low 5.56s and some potential to push further.

S1G strong enough to run comparable settings is already known to exist. Even with tRP being weaker, the stronger tRFC, tRDRD, and tWRRD should make ambient 5.55s viable. Sammy PCB likely wouldn't be used for this because of RTL and mild tRDWR weakness, however even B63 sticks should be free of this issue. Basically, golden ambient S1G should be in cold H2C territory.

On top of that Nox showed that tCL does scale with phase, and tRFC scales further. Combined with with a Z97 LN2 to make sure tRDWRs and RTLs stay low, I think even a good chiller or weak phase unit should be enough to push into the mid-low 5.55s. It should also be worth exploring if 3200ish 9-13-14 or similar higher frequency settings are worth running.

One more path would be running quad rank, which would probably be ideal on a T-top board like the 4-dimmers from Asus and Giga. I'm thinking with the same frequency and tRCD/tRP, the benefits from quad rank should about match the drop in time from the lower tCL and tRFC cold brings and lower RTLs a good 1dpc board provides. If someone could get phase + quad rank running right, 5.54s should be in range, but I would expect that to be quite difficult.

Either way, I doubt we're going to see 5.54s until someone starts PCB swapping in mass. A couple users here have demonstrated that it is possible without special tools, but even one IC is a pretty big time investment, let alone multiple dual rank sticks. Finding quality ICs isn't the problem, as multiple people here have large quantities of single rank sticks to bin through.

I think there is a future for S1G as a viable alternative to PSC/BDBG/BBSE, though it is still a ways away. As far as other ICs go, I think it's worth looking into S1E and maybe S1D, as well as revisiting 1GBit Micron and Hynix on Z97.

Edited June 16, 2023 by Sparky's__Adventure

[Noxinite]

Those are... definitely a lot of words.

Now prove it! Gimme some cold mem results!

Edited June 16, 2023 by Noxinite

[Noxinite]

Found another interesting screen that I don't think I posted - pretty sure this was air.

[SparkysAdventure]

On 6/18/2023 at 10:56 AM, Noxinite said:

Found another interesting screen that I don't think I posted - pretty sure this was air.

pic dead, re-up?

Edited July 6, 2023 by Sparky's__Adventure

[Noxinite]

On 7/6/2023 at 10:21 PM, Sparky's__Adventure said:

pic dead, re-up?

That's weird, not sure how the bot broke that. Also, I cannot remember where I found it, haha.

I think it was previously posted in this thread also it was a 4x2GB score.