Samsung DDR3 G-die

[suzuki]

Need to bookmark this topic. Thanks Stefan for the bump.

[Noxinite]

42 minutes ago, funsoul said:

Thanks for all the great testing Nox! I'm finally getting back to the z97ocf but can't for the life of me remember the voltages for samsung (iirc...psc is +0.25/+0.25/+0.25 but my samsungs hate those voltages). Been trying for a few days and totally stuck so figured to beg for some help. At 2800 9-12-12 (on air), can get to windows about 50% of the time but the least stress (like opening an app) and insta-bsod.

As an aside...is it worth running sammy or psc cold under ss or cascade? Seems much easier than hitting it with ln2 (iirc...ln2 seems most worthwhile for psc?).

Any ideas/tips/etc?

Tia!!

-Stefan

I had some progress testing alternative IMC voltages, but in the end I just stuck with the typical volts:

• SA = +0.350 to +0.450V.
• IOA = +0.150V.
• IOD = +0.200V.

PSC needs totally different IMC volts to max them out (at low CPU freq) IMO, aka "Bullant" IMC volts.

[funsoul]

Thanks dude! Got it running but +0.100/+0.0/+0.200 seems to work pretty well. Will play more, good to know what kind of boundaries to be trying! Thank you!

[ground]

I'm guessing this is said 3600+ gdie validation? https://hwbot.org/submission/3832580_noxinite_memory_frequency_ddr3_sdram_1841.2_mhz

I've finally gotten around to pick up a haswell system so I'll test my sticks pretty soon  

edit: brainfart, 4gb won't be sr gdie

Edited April 25, 2019 by ground1556

[Noxinite]

On 4/25/2019 at 5:33 PM, ground1556 said:

I'm guessing this is said 3600+ gdie validation? https://hwbot.org/submission/3832580_noxinite_memory_frequency_ddr3_sdram_1841.2_mhz

I've finally gotten around to pick up a haswell system so I'll test my sticks pretty soon  

edit: brainfart, 4gb won't be sr gdie

Heh, that was MFR.

[Noxinite]

Just now, Noxinite said:

Heh, that was MFR.

I have no idea where my results for 1GB freq testing went, but I also have no interest to test them further as MFR and CFR are still better.

[kicsipapucs]

On 5/25/2019 at 2:22 PM, Noxinite said:

I have no idea where my results for 1GB freq testing went, but I also have no interest to test them further as MFR and CFR are still better.

What results did you get with 2Gbit CFR?

[Noxinite]

Can't remember, but CFR clock worse on air then better on LN2 (than MFR).

[Samsarulz]

Great testing Nox, Maybe I can return to do some tests with DDR3 too, but Im just anamateur hahaha.

[ground]

Been trying around with my best 1GB SR Gdie stick (since for some reason my G3258 doesn't like dual rank at all - can't even post 1200) and this stuff is pretty fun. Still haven't even touched half the timings and voltages and its already doing close to 1800 for max valid at 1.8V (bios limited, probably gonna mod the board at some point). Gonna play with it some more over the next couple days and then take it cold and see where I end up. 

Couple notes already: 

- CL15 gave 10-15 MHz over CL14, CL16 doesn't work at these clocks.
- tras, trp, trcd scale negatively above 23
- Command Rate made no difference at all (at least not yet)
- Maxmem with 350 MB helps about as much as going with CL15, 10-15 MHz more (I'm quite unexperienced with mem-frequency (first time I'm trying it at all on any platform), so I'm not sure if this should be general knowledge. Are there any guides for max mem clock on Haswell out there?)

Edited June 9, 2019 by ground1556

[Noxinite]

Nice work!

[Noxinite]

I had to pull out my 1GB sticks to see if I could beat 3600MHz. ?

Seems I could.

https://hwbot.org/submission/4173147

I think it's still IMC limited as I get freezes at higher. No voltage scaling for freq over 1.8V it seems.

[ground]

Now we are getting somewhere  This is promising for more  

 

Edit: Doing more detailed testing right now, it seems like some other timings have a range where they are the most stable at high clocks, at least with my best stick:

CL: 15
tRAS: 22-23
tRP: 17-23
tRCD: 23-29

tWR: 12-15
tWCL: 7-11

Any higher and they refuse posting, with any of these settings. 

 

I did manage 3600, but I think I'm running into the IMC limit now. (again, this IMC can't do 2400 with DR sticks)

Edited June 10, 2019 by ground1556

[ground]

https://hwbot.org/submission/4179572_ground1556_memory_frequency_ddr3_sdram_1898.5_mhz

I've thrown some dice on my sticks and the CPU, got me almost another 100 MHz. Is that realistic scaling for the IMC (water -> dry ice) or is it likely I'm at the limit of the memory? Sadly I completely forgot to test ambient memory + cold CPU to compare  

[Mahameru]

6 hours ago, ground1556 said:

https://hwbot.org/submission/4179572_ground1556_memory_frequency_ddr3_sdram_1898.5_mhz

I've thrown some dice on my sticks and the CPU, got me almost another 100 MHz. Is that realistic scaling for the IMC (water -> dry ice) or is it likely I'm at the limit of the memory? Sadly I completely forgot to test ambient memory + cold CPU to compare  

any CB/CBB with your sticks? 

[ground]

4 hours ago, ivaneza said:

any CB/CBB with your sticks? 

Nt on dice at least. 

[Guest]

https://www.ebay.com/itm/SAMSUNG-8GB-4-x-2GB-2Rx8-PC3-10600E-WORKSTATION-MEMORY-RAM-M391B5673GB0-CH9/273598258321?hash=item3fb3ba1c91:g:9rQAAOSwnWpcUzBi:sc:USPSFirstClass!35405!US!-1

Here's a solid option.

[Noxinite]

Update time!

I used my last litre of LN2 from my delivery this week to (very) quickly test the 2GB g-die (in single stick config) - there is good news and bad news.

The good news is that:

• No CB.
• CBB around -120C.
• 2800C7 trained at 2.4V and full pot with 39/39/4/4 RTLs/IOLs.
• RFC scales a tiny bit with temperature (80 @ -120C).

The bad news is that:

• 2800C7 insta-failed 32M at 2.4V and full pot.
• RCD and RP do not scale with temperature (still can only train 12-12 @ 2800MHZ and full pot).

The caveats are:

• I was forced to use MX-4 thermal paste between the ICs and the heatspreader, and again between the heatspreader and the pot.
• The CPU was on the Intel stock cooler.
• I had issues with my Impact just the day before trying to run cold PSC (but I'm now 99% sure it was my CPU playing up).

[Noxinite]

Can confirm that e-die is bad, bought a stick by accident and it could only POST 2200C9 (CWL6, RDRD5) at 2V.

[Noxinite]

Tried g-die on LN2 again, still no CB and CBB around -120C. But there seemed to very little scaling and I was having issues running on the Z97 OCF. The Impact 7 seems to run better with g-die, but it's a pain with cold mem due to how small it is.

[SparkysAdventure]

Has anyone found any ds g-die on any other pcbs, such as the K0 or ST stuff?

Also, how has the binning gone?

Edited September 8, 2020 by Sparky's__Adventure

[kicsipapucs]

Has anyone ever tried G-die on X79 in quad chan?

[ground]

been playing some max memory frequency on x58, Gdie+x58a-oc feels like cheating.

http://valid.x86.fr/81dluy

1.5V VTT, 2.2V Vdimm + mostly manually tweaked timings

Might hook up the peltiers in a couple days and hope the memory scales further... I'm pretty sure I'm trp limited now.

[kicsipapucs]

Ever since I got pics from nox about gdies and then him making this topic a bit later, I felt the same as some other guys commented before me. Very insane how vendors missed this big of an opportunity for epic oc kits back in the day by not finding out. After almost 2 years of searching, I managed to snipe and pick up a tray of G-die locally a few weeks ago. I did't have much luck, all 19 sticks need 2V+ for 2800C9 tight, many 2.15-2.20V. Noxinite has multiple sub2 sticks which didn't let me go past it, so I started going lower in frequency and working my way up to try and find where does scaling stop being linear and has a big dent in it. I compared it to loose profiles results and the first stick in my binning order gave me these 2 graphs:

Could have been my first <2V 2800C9 G-die if tight profile didn't skyrocket in voltage requirement at 2788+.

I did my testings in Z97 OCF, with 4770K, xp no waza spi 32m, single channel. No cooling on the mems, just a fan sitting on top.

Edited September 3 by kicsipapucs

[Mahameru]

8 hours ago, kicsipapucs said:

Ever since I got pics from nox about gdies and then him making this topic a bit later, I felt the same as some other guys commented before me. Very insane how vendors missed this big of an opportunity for epic oc kits back in the day by not finding out. After almost 2 years of searching, I managed to snipe and pick up a tray of G-die locally a few weeks ago. I did't have much luck, all 19 sticks need 2V+ for 2800C9 tight, many 2.15-2.20V. Noxinite has multiple sub2 sticks which didn't let me go past it, so I started going lower in frequency and working my way up to try and find where does scaling stop being linear and has a big dent in it. I compared it to loose profiles results and the first stick in my binning order gave me these 2 graphs:

Could have been my first <2V 2800C9 G-die if tight profile didn't skyrocket in voltage requirement at 2788+.

I did my testings in Z97 OCF, with 4770K, xp no waza spi 32m, single channel. No cooling on the mems, just a fan sittnig on top.

Nice finding, any pics of them especially IC mark?