Ryzen Owners Zen Garden

[Mussels]

doesnt sound like anything i've ran into either

i've had BSOD issues on my ryzen 1400, since it doesnt like ram above 2400 (and i had 2667 and 3200 ram on it that was notoriously bad for 1st gen ryzen support)
It would always pass memory tests and had no issues, but i'd get the infamous black screen crashes until i lowered it

the repeated corruption makes me think it may well be the SSD itself - but the page fault errors scream memory/IMC related

 

[tabascosauz]

doesnt sound like anything i've ran into either

i've had BSOD issues on my ryzen 1400, since it doesnt like ram above 2400 (and i had 2667 and 3200 ram on it that was notoriously bad for 1st gen ryzen support)
It would always pass memory tests and had no issues, but i'd get the infamous black screen crashes until i lowered it

the repeated corruption makes me think it may well be the SSD itself - but the page fault errors scream memory/IMC related

So you think it might still be a RAM stability issue, that just doesn't show in benches? Interesting. On that note, I backed it off to the XMP timings and subtimings. 17-19-19-39, even back down to the 2T command rate. I lost about 1.5GB/s read and 1.4ns in AIDA, but I don't think losing my files on account of it grenading Windows again is a fair trade.

I can't even run, say 3200, at tighter timings. This board's latest BIOS is incapable of setting any DRAM voltage other than what JEDEC (1.2V) or the XMP profile (1.35V) tells it. If you enter anything else instead of Auto, it'll just treat it as 1.2V and get stuck on POST. A real testament to Gigabyte's commitment to providing for their non-X570 boards.

If the corruption continues to take place, this drive is probably out on its ear. It'll be the first and the last time I've ever trusted a company other than Intel, Crucial, Samsung or WD for my boot SSD.

 

[Mussels]

well yeah, the benches are at load, the crashes can be at idle/low load. stress tests dont show up every type of fault.

 

[Toothless]

Alright, my turn to join.

So my system specs are current for the Ryzen build. I did some tweaking and managed to get 4.5ghz all core with a peak voltage of 1.4v.

AMD Ryzen 7 3700X @ 4498.95 MHz - CPU-Z VALIDATOR

[f71p4u] Validated Dump by Veraltet (2020-05-12 05:41:47) - MB: ASRock X470 Taichi Ultimate - RAM: 32768 MB

valid.x86.fr

82.3 C being how hot it gets under full load, which I wanna say ain't too bad with that core clock.

I'm thinking of moving to memory timings but I might just leave that alone for now and enjoy my cpu. I'm actually excited it's getting this far.

 

[tabascosauz]

@Toothless did you just set static 1.4V? Probably a good idea to let it go back to doing its own boost thing after you're done grabbing all the tasty benchmark scores. Impressive achievement on a 3700X, nonetheless.

 

[Toothless]

@Toothless did you just set static 1.4V? Probably a good idea to let it go back to doing its own boost thing after you're done grabbing all the tasty benchmark scores. Impressive achievement on a 3700X, nonetheless.

It'll move between 1.3 and 1.4. Full load it'll sit at 1.35 and when using fewer than 8 threads it'll go back to 1.4 peaks. It'll only do 4.150ghz on its own boosting so I figured to tweak a little.

 

[tabascosauz]

It'll move between 1.3 and 1.4. Full load it'll sit at 1.35 and when using fewer than 8 threads it'll go back to 1.4 peaks. It'll only do 4.150ghz on its own boosting so I figured to tweak a little.

All core boost at around 41-41.5x is part of the design spec for 3700X; it's there because the Vcore at those speeds (~1.27-1.34V) is safe. When idle, basically none of the monitoring programs (including HWInfo) can read its actual idle Vcore, but when it's on Auto settings, the SVI2 TFN Vcore becomes accurate when the CPU is under significant load. By default they idle at between 0.2V-0.9V if you're not doing anything with the computer, it just isn't transparent to monitoring software.

I don't want to tell you what to do, but there's a lot of documentation on degradation for these 7nm silicon, and an increasing number of angry owners with mildly to significantly crippled chips after a few months of fixed frequency and fixed voltage settings. It's perfectly fine to play around with the settings and get some scores, but probably for the best to go back to default settings after. From the HWInfo window, it looks like manual OC in Ryzen Master is exactly what it sounds like, fixed 4.5GHz @ fixed 1.4V. The fluctuation just comes from you being on the default LLC setting, which allows a bit of droop.

Just a heads up - knowing how long you've held off on this upgrade, it'd be nice to have the new chip last as long as the old one. I've only been a few months earlier than you, upgrading from a 4790K as well. A lot of trial and error, ongoing learning, and frustrations past.

 

[Toothless]

All core boost at around 41-41.5x is part of the design spec for 3700X; it's there because the Vcore at those speeds (~1.27-1.34V) is safe. When idle, basically none of the monitoring programs (including HWInfo) can read its actual idle Vcore, but when it's on Auto settings, the SVI2 TFN Vcore becomes accurate when the CPU is under significant load. By default they idle at between 0.2V-0.9V if you're not doing anything with the computer, it just isn't transparent to monitoring software.

I don't want to tell you what to do, but there's a lot of documentation on degradation for these 7nm silicon, and an increasing number of angry owners with mildly to significantly crippled chips after a few months of fixed frequency and fixed voltage settings. Just a heads up - knowing how long you've held off on this upgrade, it'd be nice to have the new chip last as long as the old one. I've only been a few months earlier than you, upgrading from a 4790K as well.

What's interesting is when I leave the chip to do it's own thing, it'll go to 1.488v while staying at 4.150ghz. I'm thinking if I can get the voltage down just a bit more I should be okay for the long run.

EDIT: Peak 1.35v while staying at 4.5ghz. Either I got a really good chip or something ain't right, but the CPU-z bench keeps telling me good chip.

I should probably look in the bios to see what's going on, but when AMD's own software is letting me set these things I'm questioning if the voltage truly is 1.35v and I'm actually keeping clocks.

 

[tabascosauz]

What's interesting is when I leave the chip to do it's own thing, it'll go to 1.488v while staying at 4.150ghz. I'm thinking if I can get the voltage down just a bit more I should be okay for the long run.

VID for the individual cores is not like VID on Intel chips; it doesn't tell you anything. That's not Vcore. The SVI2 TFN Vcore field further down in the second block of CPU sensors tells you the Vcore, but only accurately during load. I don't think there's a single board that applies 1.488V to reach 41.5x all core.

Like I said, you'll regularly see idle voltages up to there in software, but it's not accurate. An oscilloscope will show you the voltage actually going on. The fact that HWInfo is open also throws another wrench into monitoring idle voltage, because every time the software polls, the voltage jumps up from deep idle @ 0.3-0.9V to 1.4V (which, by default, is safe because there's no actual load and current at that point) as the CPU reacts.

You can try P95 Smallest or OCCT on the more demanding settings if you want to make sure it's actually stable. I've never had a crash on CPU-Z before, it's not very intensive. For me, 4.15GHz @ 1.3V is the limit for my 3700X.

 

[Toothless]

VID for the individual cores is not like VID on Intel chips; it doesn't tell you anything. That's not Vcore. The SVI2 TFN Vcore field further down in the second block of CPU sensors tells you the Vcore, but only accurately during load. I don't think there's a single board that applies 1.488V to reach 41.5x all core.

Like I said, you'll regularly see idle voltages up to there in software, but it's not accurate. An oscilloscope will show you the voltage actually going on. The fact that HWInfo is open also throws another wrench into monitoring idle voltage, because every time the software polls, the voltage jumps up from deep idle @ 0.3-0.9V to 1.4V (which, by default, is safe because there's no actual load and current at that point) as the CPU reacts.

You can try P95 Smallest or OCCT on the more demanding settings if you want to make sure it's actually stable. I've never had a crash on CPU-Z before, it's not very intensive.

Would you say setting the voltage in the bios would be a sure way to tell?

 

[tabascosauz]

Would you say setting the voltage in the bios would be a sure way to tell?

Yeah. RM allows for per-CCX OCing, but that's about it. Any serious tweaks you want to be in BIOS for that. People used to report better OCs through RM, but it's not permanent, and that was also before people knew that their chips could degrade...

 

[Toothless]

Yeah. RM allows for per-CCX OCing, but that's about it. Any serious tweaks you want to be in BIOS for that.

So I went ahead and set 1.35v for vcore in the bios, and 4.5ghz core clock. It's still functioning just as before with no issues. Running P95 now to see how that goes.

 

[Mussels]

basically: we all love ryzen, but we dont want to see yours burn out and die an early death for 3FPS

 

[Toothless]

So it passed a five minute run of P95 at said settings above. If I need to drop the voltage down more so be it but I'm still curious on the lack of crashes. I should mention its 1.3v on load and 1.35v on low loads/idle.

 

[tabascosauz]

So it passed a five minute run of P95 at said settings above. If I need to drop the voltage down more so be it but I'm still curious on the lack of crashes.

The downside to P95 is that in order to find instability that isn't extremely obvious, it can take hours and hours. But if it continues to be stable, I'm starting to think that you might have a golden chip. 4.5GHz @ 1.4V at only 80C or so is pretty stellar. Maybe yields have improved exponentially. Did you happen to take a picture of the text on the IHS? The batch code sheds light on when yours was made.

If you can run CB R20 a couple times at those settings without any issues, that would be pretty stable.

 

[Toothless]

The downside to P95 is that in order to find instability that isn't extremely obvious, it can take hours and hours. But if it continues to be stable, I'm starting to think that you might have a golden chip. 4.5GHz @ 1.4V at only 80C or so is pretty stellar. Maybe yields have improved exponentially. Did you happen to take a picture of the text on the IHS? The batch code sheds light on when yours was made.

Between the OCCT, P95, CPU-z tests it's still peaking at 1.35v and load 1.3v. Peak temp says 83.8* and idle is around 35-50 with a video going on. I think I might be good!

 

[tabascosauz]

Between the OCCT, P95, CPU-z tests it's still peaking at 1.35v and load 1.3v. Peak temp says 83.8* and idle is around 35-50 with a video going on. I think I might be good!

So basically, taking into account droop, it's basically doing 4.5GHz all core @ 1.3V. That's the best 8-core specimen I've ever seen, and I'm surprised this isn't a 3800X. Might be worth taking the voltage down further to see where the limit is / what symptoms of instability look like. Can you run like 5 runs of CB R20 without crashing? Should be more intensive than all three of the above tests.

 

[Toothless]

So basically, taking into account droop, it's basically doing 4.5GHz all core @ 1.3V. That's the best 8-core specimen I've ever seen, and I'm surprised this isn't a 3800X. Might be worth taking the voltage down further to see where the limit is / what symptoms of instability look like.

After the last few days of issues and magical things popping up (found my AM4 mounting today in a box, yay!) I think it's time to see some games that were limited by my 4790k. I'll see how far it'll go at some point.

@tabascosauz I should mention that R20 won't make it through a run, but at 4.4 it'll run it all day.

 

[Chomiq]

Is anyone else dealing with Windows / drive corruption on their Ryzen builds, of any generation? I've run into data corruption issues a number of times now since moving to the 3700X platform I have now, back in August.

• The first handful of times I chalked it up to testing lower Vcore limits at 4-4.1GHz and unstable memory configurations, which resulted in a lot of BSODs, some of which may have rebooted abruptly. But a chkdsk and sfc scan seemed to rectify the problems. I reinstalled Windows, just in case, which was easy to do as this was all in the first month or two, when AMD was still figuring out firmware.
• Then, when I switched to my current memory kit, a botched install of G.skill's RGB control software and drivers also grenaded my Windows installation. So I reinstalled Windows again. Lesson learned, don't ever restart via the installation wizard's prompt.
• I spent some time at 16-18-18-32 1.35V under the illusion of stability until MW came along and proved that DJR needs to stay at 16-19-19. Some BSODs ensued, but I went back to stable settings and all was good. Chkdsk and sfc both found disk and Windows installation errors respectively, and fixed them.
• Booted up today and was met with an instant PAGE FAULT IN NONPAGED AREA bluescreen. On the next boot, I ran a chkdsk command, which found quite a few errors and spent a few minutes fixing them. Subsequently, an sfc command also found verification errors and fixed them. Both are clean now.

This memory kit is 100% stable through mt86, P95 large, IBT, membench, etc. So is the previous kit. These also aren't graphics-related (I know what that looks like because at the time of purchase, the WHQL drivers were unsatisfactory and creating some artifacts), so all I can think of are it being platform-related so a Ryzen firmware thing, or the SX8200 drive itself.

• I can't think of much I can do on the firmware front. I'm always on the latest BIOS, which is 1.0.0.4, and always kept up to date with chipset firmware.
• I bought the SX8200 because it's both a value and performance leader, but ADATA isn't the most reputable of manufacturers in my mind (despite having several of their impressive products over the years). But I can't find anyone who's had similar problems on this drive.

I've not had this kind of recurring corruption with any Intel platform I've built on over the years. My laptop is NVMe as well - the Windows installation on that has never faltered over the past 2 years, installed on the OEM PM981, which is a 3.0 x4 drive. I recently upgraded to a SN750 drive, also no issues.

A BR1500MS UPS protects my desktop, so it's not a power loss issue, either.

SX8200 Pro here since last August, no data corruption issues.

 

[lexluthermiester]

Between the OCCT, P95, CPU-z tests it's still peaking at 1.35v and load 1.3v. Peak temp says 83.8* and idle is around 35-50 with a video going on. I think I might be good!

Are you on the stock cooler?

 

[Toothless]

Are you on the stock cooler?

NH-D15 Chromax.

 

[Nordic]

Inspired by Toothless, I decided to see what my 3900x could do with a manual overclock. It was stable at 4300mhz at 1.35v. I was able to benchmark at 4400mhz and 1.375v but it wasn't stable. I am now back to running stock pbo off.

 

[27MaD]

Can you guys help here?: https://www.techpowerup.com/forums/threads/question-about-ryzen-2000-compatibility-with-a320.267017/

 

[heky]

Inspired by Toothless, I decided to see what my 3900x could do with a manual overclock. It was stable at 4300mhz at 1.35v. I was able to benchmark at 4400mhz and 1.375v but it wasn't stable. I am now back to running stock pbo off.

Have you tried running a per CCD overclock? I run CCD0 @ 4.4 and CCD1 @ 4.2 with peak voltage set to 1.325v in Ryzen Master. Am running it like that for about a month now, really like the performance and the temps are lower also.

 

[tabascosauz]

So apparently, this enigma of a Trident Z kit I have sitting around is the gift that keeps giving (?). It's too bad I don't have another DDR4 platform build to experiment with this, as it's a 16GB kit that was supplanted by this 32GB Trident RGB kit of DJR that I currently have. So it's just laying around.

Thaiphoon read the ICs as 4Gb D-die, which we all know is one of the oldest "legacy" ICs from Samsung and universally panned. You can't even find any information on this IC anymore, because Samsung replaced it with a much newer 8Gb D-die that clocks much higher in M/T but has poop timings. But the XMP is rated for 3200 16-16-16-36 @ 1.35V with tRFC as low as you'd usually see from B-die. It can't be B-die because the ICs are 4Gb. Such tight timings, rock solid at 1.35V, yet 4Gb D-die? You never see shit-tier ICs doing flat 16 timings on XMP, just look at the current 3200C16 Trident lineup, it's all a 16-18-18-38 shit-show.

This kit's contemporaries (these kits are all old, pre-Neo and pre-DJR days for G.skill) have a letter suffix that usually correctly denotes the IC revision regardless of manufacturer - so Rev.E in a kit would end in -E, CJR in -C, and B-die in -B. If it was indeed D-die, shouldn't the 042 code have been 04213X4810D, not 04213X4810E? And if the reddit guide is to be believed, two other owners of sticks with the exact same 04213X4810E code confirm that it's E-die as it should be.

So it's got to be an error, these ICs must be 4Gb E-die instead of D-die, then. That makes a lot of sense, given the unusually tight XMP; E-die is traditionally regarded as being second best behind B-die. But I've never seen Thaiphoon make a mistake before, I've only ever seen it fail to read the IC information. I guess that since literally nobody has this kit (F4-3200C16D-16GTZ), G.skill was never aware of the mistake, or it could have been a freak one-off programming mistake, since there are a few others verified with E-die on this kit?

Maybe I should get a B550 board for my 3700X to solve the C14S fitment issue, then put a 3300X in my current board so I can play with this old RAM kit some more on the side. And it visibly has the A2 PCB as well, which apparently the latest DRAM calc doesn't support despite it being the better PCB revision.