32GB of B-die, 3200 C14, overlocking guidance?

[Audioboxer]

Received my B-die today, it's a 3200 C14 kit so I'm a bit in the dark running it at 3800.

Just fired some Ryzen DRAM calculator values in to do a 1 hour stability test with TestMem5 due to it being a second hand kit. I'm confident now an hour has passed the memory isn't a dud or anything.

Now it's the dark arts of memory timing tweaking, but I guess what I'm asking is if anyone else has some 3200 C14 and can cut a few corners for me with ballpark figures to aim for? I won't cut corners on stability checking, it's just the initial numbers game to try and plug in! Came from Micron E-die, so I'm not used to numbers this low.

Cheers

 

[BallistixOnZ490]

Samsung B die is typically better for secondary timings, Micron E die for clocks.

If its for Ryzen then you dont need anything more than whatever infinity fabric needs.

The magic point is if you can get 3800 CL14 out of it.

Try take it up to 1.45v to see if you can get lower timings at 3800 if it isnt already, 3800 CL16 is quite average for current DDR4 tbh.

 

[Audioboxer]

Samsung B die is typically better for secondary timings, Micron E die for clocks.

If its for Ryzen then you dont need anything more than whatever infinity fabric needs.

The magic point is if you can get 3800 CL14 out of it.

Try take it up to 1.45v to see if you can get lower timings at 3800 if it isnt already, 3800 CL16 is quite average for current DDR4 tbh.

Thanks, I'll have a look at bringing down the main timings first. See if CL14 can be done at 3800.

Funnily enough my Corsair Vengeance RGB Pro Micron E-die set I'm replacing this with just about squeezed out 16-19-16-16 at 3733. That was with a 3900xt that would not boot 1900 FCLK. This 5950x boots 1900 fine, and when I briefly tried 3800 on with Micron E-die with the same settings stable at 3733, not stable.

So already I'm doing way better. But it's likely my CL18 Micron E-die set was simply a very average bin.

What voltage is OK for daily stability with this B-die? I've read some people go up to 1.5v, but that seems a bit toasty lol.

 

[Audioboxer]

Ooft, CL14 looks like a tall order at 3800 and/or it's going to require some serious time put into every single setting to try and get it stable. I can post at CL14 at 1.5v but instability/crashes the second I boot into Windows.

15-15-15-15 might be a more reasonable trade off, but that will require trying to stabilise with GearDownMode off.

The joys of memory, looks like I have my work cut out for me!

 

[BallistixOnZ490]

So already I'm doing way better. But it's likely my CL18 Micron E-die set was simply a very average bin.

What voltage is OK for daily stability with this B-die? I've read some people go up to 1.5v, but that seems a bit toasty lol.

Depends. I've had my Micron E Die at 1.54v for over a year now, even longer at 1.5v on the previous system. Checked its temps with an IR thermometer under memtest and I can't remember how much it was but it wasn't excessive.

I wouldn't advise on that much Voltage for B die as I haven't tried it, but 1.45v is generally considered safe for DDR4. Also depends on how much it cost, I don't mind sacrificing my £150 kit that's only rated for 3200 CL16 but also does 4200 CL16 at said voltage. If it did fail though I wont find any comparable ram now without paying a kidney.

 

[BallistixOnZ490]

Ooft, CL14 looks like a tall order at 3800 and/or it's going to require some serious time put into every single setting to try and get it stable. I can post at CL14 at 1.5v but instability/crashes the second I boot into Windows.

15-15-15-15 might be a more reasonable trade off, but that will require trying to stabilise with GearDownMode off.

The joys of memory, looks like I have my work cut out for me!

Well my crucial can't even do 3700 CL14, but manages 4200 CL16, but that doesn't help for Ryzen. Ryzen wants 3733-3800 max I believe.

Samsung B die is meant to have a better chance of higher CL14 clocks, but if you can't get that then 15 is fine if it works. Or it could just be you have an average kit if it can only do CL16.

It doesn't even make that much difference anyway, just the usual silicon lottery and luck.

 

[EsaT]

What voltage is OK for daily stability with this B-die? I've read some people go up to 1.5v, but that seems a bit toasty lol.

B-die scales quite well with (inside safe) voltage, but you have to keep temperature under control.
B-die is very temperature sensitive and around 50C is max before stability goes to hell.

So run also stress tests for CPU and especially GPU to make sure RAM won't flip over the moment you fire up game and case gets hotter.
Without load on CPU cores and InfinityFabric you also can't be sure that memory controller can remain stable when exposed to heat from those.

 

[Audioboxer]

Depends. I've had my Micron E Die at 1.54v for over a year now, even longer at 1.5v on the previous system. Checked its temps with an IR thermometer under memtest and I can't remember how much it was but it wasn't excessive.

I wouldn't advise on that much Voltage for B die as I haven't tried it, but 1.45v is generally considered safe for DDR4. Also depends on how much it cost, I don't mind sacrificing my £150 kit that's only rated for 3200 CL16 but also does 4200 CL16 at said voltage. If it did fail though I wont find any comparable ram now without paying a kidney.

That's interesting! When I was dabbling with my Micron E-die set I was advised its voltage scaling is pretty poor over 1.4v. Does it pick up again at 1.5v?

I paid £160 for this B-die set and will make most of that back selling my Cosair Vengeance Pro. Happy to get 3800 stable at decent settings. Doesn't have to be the best of the best.

Well my crucial can't even do 3700 CL14, but manages 4200 CL16, but that doesn't help for Ryzen. Ryzen wants 3733-3800 max I believe.

Samsung B die is meant to have a better chance of higher CL14 clocks, but if you can't get that then 15 is fine if it works. Or it could just be you have an average kit if it can only do CL16.

It doesn't even make that much difference anyway, just the usual silicon lottery and luck.

Running this for an hour in TestMem5 just to see if 15 is fine on some of the timings, 13 minutes in and OK so far. tCL 15 will require GearDownMode off which I believe can be a nightmare to get stable. If I turn GDM off with the above settings I get a boot loop.

Looks like I just have an average set, but I'll push it as much as I can.

B-die scales quite well with (inside safe) voltage, but you have to keep temperature under control.
B-die is very temperature sensitive and around 50C is max before stability goes to hell.

So run also stress tests for CPU and especially GPU to make sure RAM won't flip over the moment you fire up game and case gets hotter.
Without load on CPU cores and InfinityFabric you also can't be sure that memory controller can remain stable when exposed to heat from those.

I'll watch temps, running at 1.45v just now.

 

[chroniclard]

Quite a good guide I used.

MemTestHelper/DDR4 OC Guide.md at master · integralfx/MemTestHelper · GitHub

 

[BallistixOnZ490]

Right I think my E die is fine into the 60c temp range, I don't think Samsung can manage that.

 

[Audioboxer]

Quite a good guide I used.

MemTestHelper/DDR4 OC Guide.md at master · integralfx/MemTestHelper · GitHub

Thanks, that looks fantastic. I'll begin reading now whilst stability testing is going on. I'm confident CL16 will likely be my safe point at 3800, while I can boot CL14 getting it stable looks like a marathon and/or having to run over 1.5v. I've seen some people do it at 1.52v. I don't feel comfortable running at that voltage and while my PC is on a watercooling loop and has some good airflow coming in, I have no active cooling on the RAM.

This might be a bit of a daft question for everyone, but would I be right in saying 3800 at CL16 with as tight timings as possible and an FCLK of 1900 should be faster than 3600 CL14 and 1800? Or even trying for CL14 at 3733? I'd have a guess it'll all be synthetic benchmark posturing at that point, bigger gains from optimising my 5950x curve and what have you.

 

[BallistixOnZ490]

This might be a bit of a daft question for everyone, but would I be right in saying 3800 at CL16 with as tight timings as possible and an FCLK of 1900 should be faster than 3600 CL14 and 1800? Or even trying for CL14 at 3733? I'd have a guess it'll all be synthetic benchmark posturing at that point, bigger gains from optimising my 5950x curve and what have you.

For Ryzen it should be because of infinity fabric, for Intel it wouldn't be.

3733 CL14 if possible should be faster than 3800 CL16 though.

Also the problem with memory kits now is that they are heavily binned, so if a kit can do 3800 CL14 its going to be sold at a mark up for that.

Magic used to happen when 2x16 kits only went up to 3200. So much unbinned goodness to be had back then.

 

[Audioboxer]

For Ryzen it should be because of infinity fabric, for Intel it wouldn't be.

3733 CL14 if possible should be faster than 3800 CL16 though.

Also the problem with memory kits now is that they are heavily binned, so if a kit can do 3800 CL14 its going to be sold at a mark up for that.

Magic used to happen when 2x16 kits only went up to 3200. So much unbinned goodness to be had back then.

My infinity fabric actually posts at 2000mhz, but I have no idea where to even begin testing that for stability lol. RAM would need to run at 4000, correct?

I knew in advance buying 2x16GB would potentially not overclock as well as 4x8GB, but this purchase was a quick and dirty "Oh there is a second hand b-die set and I can get it for £160". Didn't want to drop down to 16GB which I was eyeing up at retail costs (some expensive 3800 CL14 sets at 16GB ~ 2x8) for well binned RAM to replace my not so good Micron E-die Corsair Vengeance.

After having just bought a 5950x, my budgets for computer gear are a bit trim right now lol.

edit - WHEA error galore in event viewer at 2000 lol. Scratch even thinking about trying to stabilise that.

 

[chroniclard]

Thanks, that looks fantastic. I'll begin reading now whilst stability testing is going on. I'm confident CL16 will likely be my safe point at 3800, while I can boot CL14 getting it stable looks like a marathon and/or having to run over 1.5v. I've seen some people do it at 1.52v. I don't feel comfortable running at that voltage and while my PC is on a watercooling loop and has some good airflow coming in, I have no active cooling on the RAM.

This might be a bit of a daft question for everyone, but would I be right in saying 3800 at CL16 with as tight timings as possible and an FCLK of 1900 should be faster than 3600 CL14 and 1800? Or even trying for CL14 at 3733? I'd have a guess it'll all be synthetic benchmark posturing at that point, bigger gains from optimising my 5950x curve and what have you.

Using that guide I got my 3200 b die running at ok 3600 CL16 with ok timings, wont go higher(that I can get working) with a 3900X. 5950X should be better I expect.

I wouldnt worry about getting an ultimate overclock, while there are performance improvements to be had its not going to change your life. Stable and tight as possible.

 

[BallistixOnZ490]

My infinity fabric actually posts at 2000mhz, but I have no idea where to even begin testing that for stability lol. RAM would need to run at 4000, correct?

I knew in advance buying 2x16GB would potentially not overclock as well as 4x8GB, but this purchase was a quick and dirty "Oh there is a second hand b-die set and I can get it for £160". Didn't want to drop down to 16GB which I was eyeing up at retail costs (some expensive 3800 CL14 sets at 16GB ~ 2x8) for well binned RAM to replace my not so good Micron E-die Corsair Vengeance.

After having just bought a 5950x, my budgets for computer gear are a bit trim right now lol.

edit - WHEA error galore in event viewer at 2000 lol. Scratch even thinking about trying to stabilise that.

Oh no you're wrong there.

2x16 overclocks significantly and much better than 4x8. My 2x16 kit was the first I saw anywhere online that reached 4000+ for 16 Gb modules, and at the time the best I saw was 2x8 4400 CL15 also on unbinned Micron E die kits.

Overclocking ram is much more dependant on the motherboard and IMC. The best boards for it are the ones with only 2 ram slots in daisy chain, the worst are 4 slots with t topology.

Some manufacturers (gigabyte) make their 2 slot boards still with T topology and 2 extra 'ghost' ram slots available. I got one to replace my dead Z390 first and sent it back for a refund after noticing that.

I only got the 10900K upgrade because my Z390 died and I couldn't find a suitable Z390 replacement anymore as they were being phased out. I got better ram overclock too (4000 CL15 1.5v to 4200 CL16 1.54v was only possible from the Z390 > Z490 switch, not something I would have done if the Z390 hadn't failed).

 

[Audioboxer]

Using that guide I got my 3200 b die running at ok 3600 CL16 with ok timings, wont go higher(that I can get working) with a 3900X. 5950X should be better I expect.

I wouldnt worry about getting an ultimate overclock, while there are performance improvements to be had its not going to change your life. Stable and tight as possible.

Wise words! I'm more happy about my FPS increases in games with the 5950x than anything. AMD finally bringing single core/light load performance up to Intel standards.

Oh no you're wrong there.

2x16 overclocks significantly and much better than 4x8. My 2x16 kit was the first I saw anywhere online that reached 4000+ for 16 Gb modules, and at the time the best I say was 2x8 4400 CL15 also on unbinned Micron E die kits.

Overclocking ram is much more dependant on the motherboard and IMC. The best boards for it are the ones with only 2 ram slots in daisy chain, the worst are 4 slots with t topology.

Some manufacturers (gigabyte) make their 2 slot boards still with T topology and 2 extra 'ghost' ram slots available. I got one to replace my dead Z390 first and sent it back for a refund after noticing that.

I only got the 10900K upgrade because my Z390 died and I couldn't find a suitable Z390 replacement anymore as they were being phased out. I got better ram overclock too (4000 CL15 1.5v to 4200 CL16 1.54v was only possible from the Z390 > Z490 switch, not something I would have done if the Z390 hadn't failed).

My bad! I was certain I read because of the modules on both sides of the RAM when its 16GB each stick it can be more unstable. RAM has always been my weak point for understanding/overclocking. Most of them time I haven't bothered much and get a sore head when trying to learn/get down in the mud with all the reboots/tiny changes and repeating the cycle lol.

Get something decent enough and then test it for 24 hours has been my motto with RAM. I just sit back in awe where people have tweaked every last setting to +1 above unstable.

 

[Aegis]

Is there any tangible benefit (aside from a slight boost to infinity fabric) to running Audioboxer's B-die at 3600/16 vs the Hynix I'm using at 3333/14? Mine has slightly better latency = 8.4 vs Boxer's 8.8 (and is a £50 16GB kit of Klevv Bolt 3200/16).

Link to my timings:

https://www.dropbox.com/s/w73o5o2rtne8axo/ZenTimings_Screenshot.png?dl=0

 

[BallistixOnZ490]

Is there any tangible benefit (aside from a slight boost to infinity fabric) to running Audioboxer's B-die at 3600/16 vs the Hynix I'm using at 3333/14? Mine has slightly better latency = 8.4 vs Boxer's 8.8 (and is a £50 16GB kit of Klevv Bolt 3200/16).

For the price difference I doubt it.

I've been looking around for a £50 16 Gb kit, best I can find is a single 16 Gb bare module for £62.

 

[Aegis]

I've been looking around for a £50 16 Gb kit, best I can find is a single 16 Gb bare module for £62.

Yeah, prices have gone up since I bought mine back around February - 'bout £70 now (2x8GB)

 

[Audioboxer]

Is there any tangible benefit (aside from a slight boost to infinity fabric) to running Audioboxer's B-die at 3600/16 vs the Hynix I'm using at 3333/14? Mine has slightly better latency = 8.4 vs Boxer's 8.8 (and is a £50 16GB kit of Klevv Bolt 3200/16).

Link to my timings:

https://www.dropbox.com/s/w73o5o2rtne8axo/ZenTimings_Screenshot.png?dl=0

Doubt it and you'd need to make sure IF was stable on your CPU first.

I can make most of what I spent back selling my Corsair RGB Ram, so dumping £150+ into B-die if you were aiming for 32GB might be better spent elsewhere.

I believe latency is key for gaming as well.

This is currently where I have my timings.

GDM off looks like a hell of a fight to try and get stable for C15, C14 will either need a stupid amount of voltage/loose timings at 3800 or FCLK dropped down and RAM at 3600/3733. Speaking of FCLK I could try and stabilise 1933/1966, but I need to do more research into how to do that.

1900 is stable, no WHEA errors. 2000 boots Windows, but a lot of WHEA errors. 1966 slightly less, and 1933 the lowest number of WHEA. So I would guess I can probably stabilise 1933, or maybe even 1966. Then RAM could be put a bit faster and maybe at 16-16-16-16.