Ryzen Owners Zen Garden

[R-T-B]

What year production do the stickers say

I don't know and they are too hard to get at now to justify going to look at them, but this was a very early kit. I bought it like when DDR4 was a hot new thing.

Probably wouldn't want to go much higher without paying attention to 4DIMM airflow too.

My whole case is basically a blower so I'm not really concerned too much about airflow.

But you are already at 1.45V

For whatever reason, the DIMMS fail to boot even at the bog standard 3200 XMP without this, though the stickers say 1.35v. I'm not really sure what is going on there.

For the record, the tertaiary and secondary timings are basically the old XMP profile at this point as I left the settings on auto there. Probably a lot more tuning to do.

It's shouldn't be badly binned B-die

To loop back on this, the fact it won't do 1T PERIOD is the only thing making me say this. That and that it seems volt hungry.

Is that very uncommon?

 

[freeagent]

The docp trc my mems set for 3200c14 is what I would use for 2K it’s crazy. Super loosey goosey.

Won’t do 1T?

Maybe they are a mixed set?

 

[tabascosauz]

For whatever reason, the DIMMS fail to boot even at the bog standard 3200 XMP without this, though the stickers say 1.35v. I'm not really sure what is going on there.

For the record, the tertaiary and secondary timings are basically the old XMP profile at this point as I left the settings on auto there. Probably a lot more tuning to do.

I have a feeling VSOC being where it is isn't helping at all. I know I said lower VSOC is always better (sorry, shouldn't have jumped the gun lol) but when working out memory settings at first it's always better to start at something safe like 1.1V and work your way down after you've tested everything else to be stable.

The minor voltages look really weird. Gigabyte (or all vendors, rather, at this point) should be following the same auto-rules as Asus, or at least on my GB boards and Asus board they seem to.

For a 1.1V VSOC, VDDG-IOD should be 1.05V, VDDG-CCD should be 1.0V, VDDP should be 0.9-0.95V. Lots of room for tweaking after that of course (I run 1.038V VSOC, 0.95V both VDDG, 0.9V VDDP), but it's a good place to start.

Also, most setups won't do 1T out of the box, command rate works a little differently on AMD. We have 1T, 1T Geardown ON, and 2T. Geardown and 2T are what most people run, depending on who you ask either of them can be faster depending on the benchmark. 1T Geardown OFF generally requires some CADBUS tweaking that isn't always guaranteed to work, and isn't much of a practical performance difference. 2T is fine.

As to volt-hungry...there's still an appreciable margin for binning regular B-die. You might get a kit that can only do 3600CL14 at 1.45V+, or a kit that'll breeze through 3800CL14 at barely 1.5V. Or you could get a kit like my old Viper Steels that's unfortunately on a A0 PCB, and can't do 3600CL14 at any voltage.

 

[WhitetailAni]

To loop back on this, the fact it won't do 1T PERIOD is the only thing making me say this. That and that it seems volt hungry.

Is that very uncommon?

I don't speak for all B-Die owners but my kit is doing 1T right now, not sure if GDM is on or off. Was bought in 2017, full specs(?) are 2x8GB 3200 C14-14-14-31. It is running with another 2x8GB kit of Vengeance LPX 3200 C16-18-18-36 with Nanya Tech dies.

 

[freeagent]

I can do like 3400ish, a little more with straight 1T, usually I run 1T gdm on, or 2T gdm off. But my limitation could be due to running a mixed set.

 

[tabascosauz]

I don't speak for all B-Die owners but my kit is doing 1T right now, not sure if GDM is on or off. Was bought in 2017, full specs(?) are 2x8GB 3200 C14-14-14-31. It is running with another 2x8GB kit of Vengeance LPX 3200 C16-18-18-36 with Nanya Tech dies.

Below 3600, 1T no GDM is easy with 0 extra voltage. I don't even think boards always even default to GDM on 3200CL16 XMP. Getting closer to 4000, just brute forcing 1T costs too much extra VDIMM without tweaks like ClkDrvStr @ 60 etc. Some of the 1T purists are super enthusiastic about how 60-20-24-24 CADBUS alone makes 1T a walk in the park, but has never worked for me on any kit on any board even at only 3600.

 

[WhitetailAni]

Below 3600, 1T no GDM is easy with 0 extra voltage. I don't even think boards always even default to GDM on 3200CL16 XMP. Getting closer to 4000, just brute forcing 1T costs too much extra VDIMM without tweaks like ClkDrvStr @ 60 etc. Some of the 1T purists are super enthusiastic about how 60-20-24-24 CADBUS alone makes 1T a walk in the park, but has never worked for me on any kit on any board even at only 3600.

If I can find the GDM setting in my BIOS I'll turn it off and see if it's stable. If it is I'll turn it back on, then go for 3400+ and if I succeed to go for 3400 1T GDM off, though if I get 3400 1T GDM on I'll take it.
GDM is definitely on; I tried to do 3333 C15 and it went up to C16. C14 was just not happening.

 

[Nordic]

I purchased two more sticks of ram for a total of 64gb. Oddly enough I have an easier time running my memory at xmp speeds now.

 

[R-T-B]

Maybe they are a mixed set?

They came together, so I don't think so.

Now:

I have a feeling VSOC being where it is isn't helping at all. I know I said lower VSOC is always better

It is. I tried higher with nearly no benefit (it'll actually do 3800 at 0.95). Thing is the ram just gets really lose timings around the 3800 mark, for some weird reason. CAS 17/18 and the like. It's much tighter here, so we get a better overall picture as far as I am concerned.

I seem to either be being beat up by my 4-layer board, or I have some really weird ram lol.

The minor voltages look really weird. Gigabyte (or all vendors, rather, at this point) should be following the same auto-rules as Asus, or at least on my GB boards and Asus board they seem to.

This may be the origin. Maybe the elite is just stupid with minor voltages. Will try the manual ones you suggest.

 

[tabascosauz]

They came together, so I don't think so.

Now:

It is. I tried higher with nearly no benefit (it'll actually do 3800 at 0.95). Thing is the ram just gets really lose timings around the 3800 mark, for some weird reason. CAS 17/18 and the like. It's much tighter here, so we get a better overall picture as far as I am concerned.

I seem to either be being beat up by my 4-layer board, or I have some really weird ram lol.

It might boot at 0.95V but is IF actually stable at 0.95V? ie. random dropouts, reboots, funky behaviour, I had some of that when trying to run 1.025V VSOC.

I was more concerned about the minor voltages - CLDO_VDDP, VDDP-CCD and VDDP-IOD are all derivative rails that derive from VSOC, so they can never exceed VSOC. If they appear to be set higher than VSOC, then the board is most likely ignoring the settings and resorting to something unknown on its own. VDDP 0.9-0.95V, CCD=VSOC-0.05V, IOD=VSOC-0.1V, good defaults

Shouldn't be the board unless Gigabyte fucked something major with the mem topology, 2DPC encompasses both 4x8GB SR and 2x16GB DR and neither should be any harder to run than the other.

Honestly if you can do 3600, I'd just stay there and do flat CL16 and tighten up all the timings, do 1T Geardown, and maybe try for 145ns tRFC. Might end up equal or faster than CL15 at those v loose board defaults. I doubt you'll ever actually notice a performance difference to even 3800CL14.

Anecdotally, I much preferred working on 4200 16-16-16 on my A2 Viper Steels than I did 4200 17-18-18 on my A0 Vipers. Just...BIOS behaved very weird at times. Maybe they were right after all, Ryzen just doesn't have a great time with odd CL.

If I can find the GDM setting in my BIOS I'll turn it off and see if it's stable. If it is I'll turn it back on, then go for 3400+ and if I succeed to go for 3400 1T GDM off, though if I get 3400 1T GDM on I'll take it.
GDM is definitely on; I tried to do 3333 C15 and it went up to C16. C14 was just not happening.

Bet those Nanyas are really holding back the B-die huh still, really surprised the Nanyas actually do flat 16, I always thought they said Nanya was just trash

 

[R-T-B]

It might boot at 0.95V but is IF actually stable at 0.95V? ie. random dropouts, reboots, funky behaviour, I had some of that when trying to run 1.025V VSOC.

Think so... ran Prime95 large ffts overnight. Was fine.

 

[Space Lynx]

Think so... ran Prime95 large ffts overnight. Was fine.

I find 20 mins of Aida64 stability test is a great tool for Ryzen setups as well. If it can last 20 mins in that it pretty much is rock solid stable in my experience.

 

[R-T-B]

@tabascosauz, your new IMC voltatge parameters helped me eek out a lot more Trfc:

Behold:

I think I can probably do some more with geardown and stuff, but for right now, I'm going to be crazy and use my PC.

 

[WhitetailAni]

Bet those Nanyas are really holding back the B-die huh still, really surprised the Nanyas actually do flat 16, I always thought they said Nanya was just trash

Probably. Once I move on from DDR4 I'll take out the Nanya kit and see just how far this B-Die could go.
Based on the Internet Nanya does seem to be trash, maybe I lucked out? The kit I got at Micro Center was a Vengeance LPX 3200 C16 kit marked "Ryzen Ready", timings 16-18-18-36. I assumed the "Ryzen Ready" meant they didn't use notoriously Ryzen-hating Hynix dies. Guess I was right.
I've also got the sticks running in a weird config - one B-Die stick and one Nanya stick per channel. My reasoning at the time of install was "this way each channel will be the same and one won't be better than the other".
Here's what CPU-Z has for one of the sticks:

 

[freeagent]

I had some Nanya's many many moons ago, they weren't terrible.. they might have surprised me too that's probably why I remember them the way I do. I had a set in SDR and DDR2..

 

[tabascosauz]

Probably. Once I move on from DDR4 I'll take out the Nanya kit and see just how far this B-Die could go.
Based on the Internet Nanya does seem to be trash, maybe I lucked out? The kit I got at Micro Center was a Vengeance LPX 3200 C16 kit marked "Ryzen Ready", timings 16-18-18-36. I assumed the "Ryzen Ready" meant they didn't use notoriously Ryzen-hating Hynix dies. Guess I was right.
I've also got the sticks running in a weird config - one B-Die stick and one Nanya stick per channel. My reasoning at the time of install was "this way each channel will be the same and one won't be better than the other".
Here's what CPU-Z has for one of the sticks:

I also thought Nanya was trash, but the memtesthelper guide was updated recently to include 8Gb Nanya B-die and it looks really good?? I don't really follow any Nanya stuff. Thaiphoon is such a shitshow of a guessing game, I don't know if there's any good way to tell aside from pulling off the heatspreaders.



Better than CJR and on par with 4Gb E-die and DJR is a very bold claim indeed. But according to the guide, Nanya doesn't scale tRCD, tRP or tRFC, so still v impressed you managed 16-16-16.

However, Nanya had some great DDR3 iirc ooops that was elpida lol, rip in pieces elpida swallowed by micron

@tabascosauz, your new IMC voltatge parameters helped me eek out a lot more Trfc:

I think I can probably do some more with geardown and stuff, but for right now, I'm going to be crazy and use my PC.

Have you done any memory testing with TM5 or HCI?
Memory Testing with TestMem5 TM5 with custom configs | Overclock.net
MemTest for Windows Download Page (hcidesign.com)

I had no idea CLDOs could even affect tRFC but 140ns is very healthy indeed if it's stable.

But even if 140ns throws errors in testing, imho the absolute limit (lowest tRFC value below which no POST regardless of VDIMM) and stable point for tRFC generally aren't too far apart. Maybe would only have to back off 3-5ns or so at most if you don't want to add VDIMM.

You should bench some AIDA to find out exactly what the difference between 252 and 630 tRFC is

 

[R-T-B]

Have you done any memory testing with TM5 or HCI?
Memory Testing with TestMem5 TM5 with custom configs | Overclock.net
MemTest for Windows Download Page (hcidesign.com)

Generally I just use Prime95 large ffts with AVX2/FMA instructions on. Are these a lot better?

It's already passed an hour and a half of large ffts.

 

[tabascosauz]

Generally I just use Prime95 large ffts with AVX2/FMA instructions on. Are these a lot better?

It's already passed an hour and a half of large ffts.

its a decent memory test but more for DIMM temp stress and memory controller stress

HCI and TM5 give you a bit more reference as to how stable/unstable your setup is

• 300% in HCI is mostly stable, 600% is probably good for a daily, I'd personally do 1000-2000% for a daily, generally needs to run overnight to reach that coverage
• 3 full cycles in TM5 with anta777 config is good (you can drop whatever config you like in the folder, it's just a .cfg file, but anta777 Extreme is generally liked), will be about 4 ish hours

I would usually then come back to P95 Large and run it for a hour or so to test IF, but it's just 3600 so you're probably fine. But I'm lazy now haven't used large FFT much, Ryzen 5000 is much less temperamental than Ryzen 3000, and on the Ryzen 4000 APUs the iGPU is the primary IF stress tester and even harder memtester than HCI or TM5 so

 

[WhitetailAni]

Better than CJR and on par with 4Gb E-die and DJR is a very bold claim indeed. But according to the guide, Nanya doesn't scale tRCD, tRP or tRFC, so still v impressed you managed 16-16-16.

I didn't even know the company existed until I got the kit, looked at the dies, was like "wait the dies are from who??", Googled them, and got an answer "We make RAM and other things."

I'm amazed I got the timings I did too; I first went for frequency and hit a wall at 3333, though I didn't mess with the IMC voltage at all (left at 1.1). After I landed on 3333 I just sent the timings down bit by but until I landed at 16-16-16 since I didn't think I should put the other timings lower than the CAS latency, though if it's fine I may shoot for lower after I do attempt for 3400.
I'm not touching CPU overclocking since PBO is doing its job.

Also, what's good software for really stressing memory? I used memtest86 to test for stability but I'm not sure if there's anything better.

 

[freeagent]

Anyone have to RMA their Zen 3 CPU recently? My 5900X will not pass TM5 even @ stock. Going to put my 5600X in shortly and see what happens.

So far, my observations..

CPU has been acting a little weird. From the start, I noticed memory speed had almost no impact on latency.. ok not the end of the world. Since I ordered a new board for a spare CPU, I pulled out a pair of my ram. Now running 2x8.. Was testing known speeds and timings, but was failing within minutes. I let it run @ full stock, and it still threw errors within minutes. Uh-oh. This morning woke up to errors @ stock. So I had my cooler off and CPU's cleaned and swapped in under 15 minutes.. not bad.. anyways I started testing and she was passing. My ram is fine, because I honestly was hoping it was just that. Ram is good, board is good, 5600x is good, 5900x is bad

I didn't take a screen of the failures from the other CPU. But errors started within 7 minutes. Bummer dude. Just running it was fine, but load up something like that and it would fail.

Edit:

I have initiated the RMA.

Edit again:

Ok I plugged it back in in the hopes that it was just a weird uefi bug.. it just passed the spot it errored at.. ffs lol..

Last edit!

IDKWTF to say, really. Its working just fine again. I would not run for more than 7 minutes, sometimes less. I honestly thought my CPU was done. Never had one go before.. so I don't know what its like when they are on the way out lol..

 

[Space Lynx]

Anyone have to RMA their Zen 3 CPU recently? My 5900X will not pass TM5 even @ stock. Going to put my 5600X in shortly and see what happens.

So far, my observations..

CPU has been acting a little weird. From the start, I noticed memory speed had almost no impact on latency.. ok not the end of the world. Since I ordered a new board for a spare CPU, I pulled out a pair of my ram. Now running 2x8.. Was testing known speeds and timings, but was failing within minutes. I let it run @ full stock, and it still threw errors within minutes. Uh-oh. This morning woke up to errors @ stock. So I had my cooler off and CPU's cleaned and swapped in under 15 minutes.. not bad.. anyways I started testing and she was passing. My ram is fine, because I honestly was hoping it was just that. Ram is good, board is good, 5600x is good, 5900x is bad

I didn't take a screen of the failures from the other CPU. But errors started within 7 minutes. Bummer dude. Just running it was fine, but load up something like that and it would fail.

View attachment 208684


Edit:

I have initiated the RMA.

Edit again:

Ok I plugged it back in in the hopes that it was just a weird uefi bug.. it just passed the spot it errored at.. ffs lol..

Last edit!

IDKWTF to say, really. Its working just fine again. I would not run for more than 7 minutes, sometimes less. I honestly thought my CPU was done. Never had one go before.. so I don't know what its like when they are on the way out lol..

View attachment 208717

I think ram is more finnicky then we give it credit for... and it has less to do with your CPU. so prob would be better off using prime 95 cpu only test to truly test overnight cpu issues. or aida64 system stability test

 

[freeagent]

It was really weird. dropping in the other CPU seemed to fix whatever was going on. But when it failed TM5 @ stock that was it for me I was frustrated. I do have it running properly now.. and I'm not going to fuck with it for awhile.. I am running a beta bios, maybe that had something to do with it.. I really don't know what to say and I do feel kind of silly. But I was working on it for 6 hours at that point and my brain was a waffle.

She's running good, this is the fastest single core score I have been able to pump out with this thing.. someone with a 5800X topped me so I will have to try harder

freeagent`s 3DMark CPU PROFILE 1 THREAD score: 1010 marks with a Ryzen 9 5900X (hwbot.org)

 

[Space Lynx]

does anyone know the equivalency of the ps5 zen cpu to its modern day cpu? speaking cpu only not the gpu built in...

is it comparble to a 5800x or 3700x?

 

[Mussels]

does anyone know the equivalency of the ps5 zen cpu to its modern day cpu? speaking cpu only not the gpu built in...

is it comparble to a 5800x or 3700x?

dont we know that from the weird APU kits?
This AMD mini PC kit is likely made out of b0rked PS5 chips | PC Gamer
4700s seems to tag behind the 3700x, close single threaded and worse multi

 

[Space Lynx]

dont we know that from the weird APU kits?
This AMD mini PC kit is likely made out of b0rked PS5 chips | PC Gamer
4700s seems to tag behind the 3700x, close single threaded and worse multi

I guess I just don't understand how some games can run at 4k 120hz on ps5 but a higher end 5600x rx 6800 pc like i used to own can't do 4k 120hz in similar games...