Overclock.net banner

Limited Edition GALAX HOF Tempest DDR4-5000 CL19 2x8 GB Ultra Binned Samsung B-Die Kit Review

11K views 111 replies 14 participants last post by  Ichirou 
#1 · (Edited)
Latest Update: Feb, 3, 2022

I was fortunately given the opportunity to acquire this kit from GALAX, which had to be imported from Hong Kong via FedEx.
The total cost came to $439 USD for the RAM and shipping, and the customs ended up costing about $66 USD, for a total of about $505 USD.
(Yes, the cost is insane. And yes, I'm insane enough to fork out the money just to have some fun with memory overclocking.)

After buying my ultra binned Micron B-Die kit (in my signature) a while back and pushing it to the limit on the Z390 platform, the itch for more never went away.
That kit in particular is no longer available for sale from what I can tell, so it is special if only for that status alone.
Of course, I could've purchased any random Samsung B-Die kit like anyone else, but I've become a bit of a connoisseur for exotic and unique hardware.
There are few (if any) Samsung B-Die kits prebinned to 5,000+ MHz on XMP, so considering that it is limited edition, I figured I would bite.

Attached are initial photos of the product. Since I just received my binned 12900k in the mail, I will start testing once the system is fully set up.
For now, the system specs for the new test bench that I will be using is detailed below.
The RAM will be tested with XMP and then manually overclocked in both Gear 1 and 2. The CPU will remain at stock for now, with E-cores disabled.

CPUIntel Core i9-12900K (Prebinned w/P-core SP 98; E-cores disabled)
CPU CoolerNoctua NH-D15 (Brand new; dual fan setup) with Thermal Grizzly Kryonaut and LGA1700 bracket
GPUIntegrated GPU
MotherboardASUS ROG STRIX Z690-A Gaming Wi-Fi D4 (BIOS version 0812) & MSI MPG Z690 Edge Wi-FI DDR4 (BIOS version 122 beta)
RAMGALAX HOF OC LAB Tempest DDR4-5000 2x8GB (Prebinned Samsung B-die) with table fan cooling
PSUEVGA SuperNOVA 850W G6 80+ Gold (Brand new)
StorageWD Black SN750 SE 1 TB Gen. 4 NVMe SSD (Brand new)
OSWindows 10 Pro
CaseOpen Test Bench
 

Attachments

See less See more
9
#2 ·
Can't wait to see what they can do!
 
  • Rep+
Reactions: Ichirou
#4 · (Edited)
Yeah, I'm probably going to have to buy another PSU as it would be inconvenient to run both systems one after the other. Still need this one for work, after all.
Beauties. Shame you are not putting that binned 12900k on water.
Actually, it will be, but the water loop parts haven't all arrived yet. Still waiting on the MO-RA3 order and then I'd have to order fittings once I do physical measurements in the case. Spent about $1700 on the water loop parts so far (although a significant portion of that was shipping and customs); will have to spend more once the fittings are factored in. I've been doing very, very strict calculations on everything I'd need for the loop so as not to waste money on unnecessary components (as best as I can), but it's a pretty complex loop with various aesthetics involved, so it ended up costing more than the typical water loop configuration. Otherwise it would probably be like half the cost. But, I'd rather spend the money on the water loop than on some insanely scalped GPU that I'd probably only use for a year at most and even then never fully capitalize the performance on, so... In any case, I just figured I'd get a head start in benching some stuff on this new motherboard.

I will be reporting back a bit later, probably with a different PSU. In the meanwhile, how should I jump the motherboard to start it? Take a screwdriver and stick it between the two POWER SW pins? Is it safe with a magnetized screwdriver? On this motherboard, the POWER SW pins aren't marked as + and - but as something else entirely, so I'm not sure whether that makes a difference.
 
  • Rep+
Reactions: Jedi Mind Trick
#7 ·
I was fortunately given the opportunity to acquire this kit from GALAX, which had to be imported from Hong Kong via FedEx.
The total cost came to $439 USD for the RAM and shipping, and the customs ended up costing about $66 USD, for a total of about $505 USD.
(Yes, the cost is insane. And yes, I'm insane enough to fork out the money just to have some fun with memory overclocking.)

After buying my ultra binned Micron B-Die kit (in my signature) a while back and pushing it to the limit on the Z390 platform, the itch for more never went away.
That kit in particular is no longer available for sale from what I can tell, so it is special if only for that status alone.
Of course, I could've purchased any random Samsung B-Die kit like anyone else, but I'm become a bit of a connoisseur for exotic and unique hardware.
There are few (if any) Samsung B-Die kits prebinned to 5,000+ MHz on XMP, so considering that it is limited edition, I figured I'd bite.

Attached are initial photos of the product. Since I just received my binned 12900k in the mail today, I will report back once the system is fully set up.
For now, the system specs for the new test bench that I will be using is detailed below.
The RAM will be tested with XMP and then manually overclocked in both Gear 1 and 2. The CPU will remain at stock for now, with E-cores disabled.

CPUIntel Core i9-12900K (Prebinned w/P-core SP 98; E-cores disabled)
CPU CoolerNoctua NH-D15 (Brand new; dual fan setup) with Thermal Grizzly Kryonaut
GPUIntegrated GPU
MotherboardASUS ROG STRIX Z690-A Gaming Wi-Fi D4 (BIOS version 0901)
RAMGALAX HOF OC LAB Tempest DDR4-5000 2x8GB (Prebinned Samsung B-die) with table fan cooling
PSUPending a new one
StorageWD Black SN750 SE 1 TB Gen. 4 NVMe SSD (Brand new)
OSWindows 10 Pro
CaseOpen Test Bench
Afaik strix isn't compatible with noctua
 
#8 ·
Already installed the cooler just fine ;)
Didn't even need the LGA1700 bracket. Just used whatever old socket one it had included in the box. (Apparently, this NH-D15 isn't even up to date for LGA1200, lol)
I directed the fans to the left (back). Dual fans. Just barely, but doesn't, touch the RAM.
 
#11 · (Edited)
Still waiting on the PSU to arrive in the mail, and then I'll be able to test and see. Naturally, the goal's to beat the 4,800 CL17 kit.
Honestly, if it can even do 5,000 CL18, that's kind of good enough. But we shall see. According to latency calculator, I'd need 5085 MHz CL18 to beat it exactly.
This kit is binned for high voltage, so that should yield some promising results? Maybe, I don't know.

Frankly, I'm more optimistic about something like 4,000 CL13 on Gear 1 rather than 5,000 CL17 on Gear 2.
We want Taiphoon Burner report once it's running . Curious to see the production date. (y)
Of course :)
 
  • Rep+
Reactions: Jedi Mind Trick
#10 ·
We want Taiphoon Burner report once it's running . Curious to see the production date. (y)
 
#16 ·
Sadly, I didn't invest in any platform after Z390 and prior to Z690; I'm not really one of those people who upgrade with every new generation that releases :p
If only Purolator worked on the weekend, I'd probably have the PSU by now... But I guess it'll have to wait until Monday.

On a side note, I might end up swapping the motherboard for the MSI Edge, as I've heard from Carillo (who prebinned this CPU) that he achieved 4,300 CL14 1T on that but not on the Strix. That'll come later I suppose. For now, let's see what BIOS 0901 does for me.
 
  • Rep+
Reactions: YoungChris
#17 ·
Finally got the PSU today, and after some workspace clean-up, I have the test bench up and running. Jump starting with the PWR SW headers works like a charm.

Initial observations:
  • The Strix Z690 refused to work with my 4K monitor, so I had to use my 1440p monitor instead. Annoying and caused me some worry when I saw that the PC wasn't responding on boot, but apparently it is a known issue. I hope ASUS has addressed/will address this in a future pre-bundled BIOS for their Strix or other Z690 motherboards, and maybe address it in their FAQs online somewhere.
  • The new EVGA 850W G6 PSU "clicks" every time the system boots up or restarts. Not something that I'm used to hearing and not sure if it's any cause for concern, but maybe it's just what new PSUs are like nowadays. My old G2 doesn't do such a thing.

No Internet connected at the moment; Windows 10 Pro installed without license (since it's a test bench). All tracking garbage refused on installation.
It's now time to test XMP, with voltage tweaks.

  • PC fails to boot on XMP II (VDIMM is set to 1.60V as expected).
  • I set VDDQ to 1.30V and VCCSA to 1.40V. Training fails.
  • I set VDDQ to 1.40V. Training fails.
  • I set VDDQ to 1.45V and VCCSA to Auto. Fail.
  • I set VDDQ to 1.35V and VCCSA to 1.45V. Fail.
  • XMP I fails. Gear 2 fails.

So at this point, I decide to try Gear 1 4,000 MHz instead.

  • VCCSA/VDDQ at 1.35V, 4,000 MHz Gear 1 with 14-14-14-30 boots just fine.
  • Command Rate at 1N doesn't work; not sure how to get 1T working. N:1 and 1 doesn't work either, am I even doing it right?
  • RAM doesn't seem to want to do CL13 at 4,000 MHz.
  • 4,133 MHz @ CL14 boots just fine. Probably ASUS limitation here, as it seems many people hard-wall at 4,133 MHz for Gear 1.
  • BCLK 101.0 for 4,174 MHz @ CL14 boots just fine.
  • Just out of curiosity, 4,800 MHz with 17-19-19-39 G2 fails, even at 1.70V. For a supposedly binned kit, these results are disappointing.

I'm not sure whether or not these poor results are because of the motherboard or the CPU IMC. Or BIOS not being good enough (it can't do 5,000 on XMP after all). I'm definitely getting that suspicion, though. Thinking of swapping over to the MSI Edge at this rate. But I've got some more things to test before doing so.

Anyway, it's getting late, so I'll fiddle around with it more later. Feel free to offer suggestions for stuff to try, and I will.
 
  • Rep+
Reactions: J7SC
#20 · (Edited)
Try the MSI. Ill bet it doesnt do any better in gear 2. It has been like that every gen. 2 dim MB will clock an easy 2-300+mhz higher vs 4 dim mb.

It is the fact they are 4 dim board not the IMC or your memory bin. About gear 2 not doing 5000+.

Even on z590 Apex and Unify X you had to have a very good IMC and mem to clock over 4800 to 5000+ on b die.
Got it. So I was right to suspect that it is a motherboard/BIOS issue, not the CPU/RAM. I will try 0707 then. And 0812, since that's in your screenshots.
How do I enable 1T? Is it just Command Rate set to 1N? What primaries did you test for 4,600 MHz?

It definitely felt like the RAM wasn't the bottleneck here; it happily accepted ~4,133 MHz CL14 flat without any issue and simply didn't like going any further beyond.
Haven't tested out 4,600 MHz yet so that's something I should do.

I see, so I shouldn't even bother with swapping to the MSI Edge then, as results would likely be very similar.
I wonder why the Strix/Edge even bother to advertise 5,000+ MHz support then? Only specifically tested Hynix dies, I'm guessing?

On a side note, I forgot to fetch the Thaiphoon readouts. I'll do it the next time I test the system again.
 
#19 ·
Try the MSI. Ill bet it doesnt do any better in gear 2. It has been like that every gen. 2 dim MB will clock an easy 2-300+mhz higher vs 4 dim mb.

It is the fact they are 4 dim board not the IMC or your memory bin. About gear 2 not doing 5000+.

Even on z590 Apex and Unify X you had to have a very good IMC and mem to clock over 4800 to 5000+ on b die.
 
#27 ·
Hmmmm...

Now why would a beta BIOS do so well, I wonder ;)?
 
#34 ·
And yeah i would have never believed SR was harder to run. But some bios versions SR just doesnt want to work where DR will with no problems. I think they optimized more for DR since that is what most run but I could be wrong. Going by how the bioses worked for me in testing it seems true.
 
#35 ·
@bscool
If BIOS 0003 works out well, I will try out my current Micron B-die SR set just to see what the ceiling is on Gear 1. Would be genuinely curious, seeing as the Prime Z390-A is also limited not by the RAM but by the motherboard/CPU.

On a side note, I also came into possession another exotic RAM kit recently, although I haven't posted about it yet (wanted to finish up with this kit first). Hynix CJR (most likely) or DJR (unlikely) 4x16 GB 1.20V at flat CL15 no-XMP profile. Not sure how it'll overclock but it should be fun to play around with as the voltage is low. Will provide more details in the coming days.
 
  • Rep+
Reactions: bscool
#36 · (Edited)
@bscool
Tried 0003, 0707, and 0812 so far. Still can't seem to boot at any frequency 4,200+ MHz even with extremely loose primaries; doesn't matter if it's in Gear 1 or 2. I definitely think that there is something else that I'm missing.
Did you need to tweak any specific non-RAM-related BIOS settings in order to break the 4,200 MHz barrier? I do have a prebinned 12900k, so I don't think the IMC is the issue here. I think this issue probably applies for previous generations as well, since 4,200+ MHz has always been a sort of wall in general.
 
#37 ·
@Ichirou Try this cmo file


Should not need to change anything to run 4266 other than mem clocks. Maybe dram voltage. if it doesnt work let me know and Ill make a new one that works for me at 4266 and link it.

txt file to get an idea of setting I used. 4133c15bios707_setting.txt

Best practice is only to use cmo files from different bios version but so far the seem to work on Strix except bios 1001 I still dont know what to think about it.
 
#40 · (Edited)
@Ichirou Try this cmo file


Should not need to change anything to run 4266 other than mem clocks. Maybe dram voltage. if it doesnt work let me know and Ill make a new one that works for me at 4266 and link it.

txt file to get an idea of setting I used. 4133c15bios707_setting.txt

Best practice is only to use cmo files from different bios version but so far the seem to work on Strix except bios 1001 I still dont know what to think about it.
Very well, will test your 0707 BIOS CMO later.

That's what I imagined as well; I shouldn't be hard walling at 4,200 MHz as that's an easy to achieve frequency with a kit like this, especially loosened. Definitely seems like something is wrong. Up to 4,174 MHz is fine though, but that's not the goal here.
@Ichirou Some other ideas is to test each stick individually. then each slot individually to find weakest. a2 and b2.

Say 1 slot can do 4200 and the other 4300. then it tells you something.
So the motherboard itself might be a dud, you're saying? I'll try swapping the sticks around, but I don't think it's going to matter much; same batch and all.
This has been the easiest platform by far to overclock. If the setting I am using for you dont work, something is wrong. Issues somewhere with hardware would be my guess.

Edit Not sure if you know about vddq. Depending on bios version you might have to bump that up. and io also.

Again 707 and 901 are the best bios in my testing. 901 allows lower io/vddq.

But who knows maybe the sticks you got like another bios? I know many liked 707 for both SR and DR,

also i prefer to stick to 133 ratio so 4133, 4266 etc. i have never tried 4200 or 4300 so no idea how they work.

ROG STRIX Z690-A GAMING WIFI D4 maybe there is something there that will help you.
So far, I've tested VCCSA at 1.35-1.45V, and VDDQ at 1.35V. I will test 133 ratios in case there are frequency gaps.
VCCSA at 1.45V felt like it had ever-so-slightly more potential than at 1.35V.
It seems that your HWBOT thread link has a 1001 BIOS version, so I'll try that as well.

On a side note, I've attached the Thaiphoon readout to the main post.
 
  • Rep+
Reactions: bscool
#39 · (Edited)
This has been the easiest platform by far to overclock. If the setting I am using for you dont work, something is wrong. Issues somewhere with hardware would be my guess.

Edit Not sure if you know about vddq. Depending on bios version you might have to bump that up. and io also.

Again 707 and 901 are the best bios in my testing. 901 allows lower io/vddq.

But who knows maybe the sticks you got like another bios? I know many liked 707 for both SR and DR,

also i prefer to stick to 133 ratio so 4133, 4266 etc. i have never tried 4200 or 4300 so no idea how they work.

ROG STRIX Z690-A GAMING WIFI D4 maybe there is something there that will help you.
 
#41 ·
@Ichirou Havent you heard of binning sticks? Test 1 stick at a time.

Same bin the slots(a2 and b2 or slot 2 and 4). You will find one that clocks higher. Been like that every gen. Some vary more than others.

On z590 I had CPU/IMCs that one channel clocked 200_mhz higher on one channel. Either IMC or MB variance. Usually farthest b2 will clock higher.

Because we know your CPU/IMC ran 4300 on another system we need to find the weak link in your setup now. that is my thought process of testing 1 channel and stick at a time.
 
#42 ·
Ah, okay. I see what you mean. I'll do that then.
I've already tested out your 0707 CMO config by the way; didn't really get anywhere with it. Still stuck on 4,174 MHz max.

I'll let you know what happens once I try individual DIMMs in A2 and B2. If the motherboard really is at fault, that would be unfortunate. Wouldn't know what to do besides an RMA, since I am past the return period (not that I could return it anyway, due to the store's return conditions), and am not really interested in binning motherboards as it is almost guaranteed a loss with each board I have to resale.

Do you suggest swapping to an MSI Edge instead, or would that not really change much? I do intend to test quad DIMM setups after this, which would theoretically do better on T-Topology than Daisy Chain.
 
#45 ·
Oh hey something else I remember when testing bioses, Some versions tCL 15 wont even boot. Have to use 14 or 16. cant remember the bios version though. I dont think it was 707 or 901. I think one of the 8xx that did that.

You could try a1 and b1 but I dont think it will work. But what have you got to lose?

About swappping boards totally up to you. If it is worth possibly running say 4266 vs 4133? up to you. It would be interesting if you did it. I would like to see the results :)

What bios version are you on now?
 
#46 · (Edited)
Well, if I know for a fact that A2 is a dud, even if I do test the rest and they are good, the A2 slot would still limit the overclocking headroom for more than one DIMM.

I'll test A1 + B1 instead of A2 + B2. Not getting much luck. Seems to be pretty average.

BIOS 0707 with your CMO as a baseline to test from
 
#59 ·
@Ichirou So why does MSI Edge not run DR b die like Strix can if they have better slot to slot variance? But they cheaper MSI A can do DR 4000+ with tight subs?

The people on the MSI Edge and Tomahawks are the ones with the most issues running DR b die at higher frequencies if you read the z690 ddr4 form.

Not saying you are wrong but it doesnt add up with what I am seeing from user feedback and from people that have owned both MSI Edge and Strix.
 
#60 ·
It's possible that the Strix is better for DR while the Edge is better for SR, in comparison with each other. But that's an unsubstantiated guess.
Seeing as I have primarily SR kits, it wouldn't hurt to try out the Edge. I'll let you know how that turns out.
For now, I'll still try to play around with this Strix.
 
  • Rep+
Reactions: bscool
#62 · (Edited)
@bscool
Swapping to the beta 0003 BIOS, I managed to boot into Windows at 1.74V and stay on the desktop for a decent amount of time. Below are some readouts.

We're getting places. Let's go boys. Time to TM5.

Update: In TM5 1usmus config, getting lots of Error 6's, implying incorrect IMC voltage. Will have to tweak them.
 

Attachments

  • Rep+
Reactions: bscool
#63 ·
Maxmem Maxmem info?

I think beyond a certain voltage you have to use this.

Also a couple of my SR for reference so you have something to compare against. My timings are looser but my sticks are not that good.
 

Attachments

#64 · (Edited)
@bscool
Until I've stabilized the RAM and reworked the timings (they were done on Auto), things will probably be inconsistent in terms of benchmarks.
Also, the cache clock is a total mess since it's only on Auto with a Noctua. That's another significant factor. Until it's under water, it's gonna be bad.

What is this "Maxmem" used for?

At the moment, I'm getting a huge mix of Error 0 and 6 in TM5 1usmus, which implies that I still need to fine tune the voltages a bit more. Maybe the skews too.
Error 0 = not enough VDIMM or voltage in general, and Error 6 = unstable IMC voltage, usually too much

With VDIMM left at 1.74V, a VCCSA of 1.44-1.45V gives lots of 6's, while 1.40V gives lots of 0's. At 1.43V, I get them both equally.
Fine tuning will take a while, but I'll figure it out. Might have to play around with VDDQ since I'm at this middle point with VCCSA.

Are there any other voltages in the BIOS that is relevant specifically to the DIMM OR the IMC but not both?
 
  • Rep+
Reactions: bscool
#65 · (Edited)
@Ichirou Those were run with the cpu at default/auto/MCE or using AI oc. So no special setting for CPU or cache all auto more or less.

Cache should go to 4700 on on default cpu setting or using Ai OC. Opportunistic i think shamino called it. Clocks up when e core below certain clock I think is how it works.

Only setting sync all core turn off cache ramping up. Per core it will still clock to 4700 if left on auto/defaults.

I hope you didnt take it like I am saying something is wrong with yours. I thought it might be helpful for a comparison point.

Maxmem is for booting into windows at higher voltages. I think it is around 1.7v it needs to be used. I thought you were having issues getting into Windows before with high dram.

You voltages are too in my opinion for SR.

I have never used that high and wouldnt unless you are on Watercooling dims and CPU. I bet heat is part of your problems.

I dont adjust any skews, odts or anything. You shouldnt need to do any of that. Guys running high clocks and tight timings dont mess with that on z690 that I know of. That is z390/z490 stuff.

I mean mess with it and see. I never did and from what i know most clocking high clocks and tight timings dont.
 
#66 ·
@Ichirou Those were run with the cpu at default/auto/MCE or using AI oc. So no special setting for CPU or cache all auto more or less.

Cache should go to 4700 on on default cpu setting or using Ai OC. Opportunistic i think shamino called it. Clocks up when e core below certain clock I think is how it works.

Only setting sync all core turn off cache ramping up. Per core it will still clock to 4700 if left on auto/defaults.

I hope you didnt take it like I am saying something is wrong with yours. I thought it might be helpful for a comparison point.

Maxmem is for booting into windows at higher voltages. I think it is around 1.7v it needs to be used. I thought you were having issues getting into Windows before with high dram.
Ah, I see. Well for my PC, it's kinda clocking low in general, so I assume that it's just because of the CPU thermal throttling and pulling everything down.
Within expectation, of course. It's a Noctua with an incorrect bracket on a 12900k, lol.

Actually, in my case the issue is just with POSTing altogether; not exactly with getting into Windows. So that's entirely a BIOS training issue.
 
#70 · (Edited)
You are running memtest/tm5 with 1.7v? Dram?

you said your benching but then talk about memtest errors?

I know I am nosey huh ;)

Another thing if you get to where nothing seems to work. Turn off PSU and hold clear cmos button for 30 seconds or so can help.

Make sure you have your profile saved as the setting will be to default on next boot.
 
#71 ·
You are running memtest/tm5 with 1.7v? Dram?

you said your benching but then talk about memtest errors?

I know I am nosey huh ;)
Hey, that's part of the process :p
It's set at 1.74V VDIMM right now. Still tweaking voltages.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top