AMD Ryzen 3000 Builders Thread

Page 73 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Shmee

Memory & Storage, Graphics Cards Mod Elite Member
Super Moderator
Sep 13, 2008
7,382
2,418
146
This does sound likely to be a CPU issue then. That said, I have seen a few times faulty memory that passed the UEFI memtest86 4 passes...At this point I would attempt to RMA or return the CPU.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
So it can't be because of the stock cooler is not up to this task?
 

Shmee

Memory & Storage, Graphics Cards Mod Elite Member
Super Moderator
Sep 13, 2008
7,382
2,418
146
What were the temps like? If they were reasonable, then the CPU is likely just unstable under AVX load.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
The error appears almost right away (under a minute). Almost always worker 3 or 4 fails first.

According to Ryzen Master, the CPU temperatures when first worker has that rounding error is between 75-83 Celsius or so (not exactly the same each time).

If I let it run with the remaining workers it gets to about 90 Celsius. At which point I stop the test.

As you can see, the CPU temperatures are high. This is why I think it might not be the CPU, but the inadequate stock cooler for running Prime95.

PC case is Fractal Design Define R5, with 2 front 140mm fans and 1 back 140mm fan.

At idle, the CPU stays at about 34-37 Celsius.

In Cinebench R20 (during the multithread test) the CPU temperature gets to about 82 Celsius.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I've tested with Cinebench R20, Blender Benchmark, Windows 10 Memory Diagnostic, memtest86 v.8.2 (Passmark UEFI version), 3DMark (demo version from Steam), Unigine Heaven and Valley benchmarks. All OK (meaning no errors, nor crashes).

First thing I would recommend is to test with default RAM settings until you get your temp problem diagnosed. Shoot for XMP later.

But in Prime95 (Version 29.8 build 6) I get rounding errors almost right away. Usually for Worker 4 and 3.

That is not good. At stock with the stock cooler, the CPU should be able to select an appropriate voltage and clockspeed setting to get Prime95 stable out-of-the-box. My 3900x (for example) downclocks and undervolts pretty aggressively in this workload, even with massive cooling. My current "default" settings (I tweaked LLC and other stuff) runs Prime95 SmallFFTs at 1.104-1.128v vcore with a temp of 58C. It sits at around 3900 MHz running this bench. That's only 100 MHz higher than the advertised "base" clock. And under older AGESA versions with different LLC settings, it wouldn't even run that well. AMD has improved the Prime95 performance of this chip considerably with microcode updates.

If I disable AVX2 and AVX, I do not get this error. But I only let it run for about 10 minutes at a time, because the CPU temperature gets to about 90 degrees in this case.

90C with no AVX?!?!?

Disabling the memory XMP profile does not help with Prime95.

Glad you tried that. For now, i would recommend leaving XMP disabled anyway. One thing at a time.

I tried then the second available BIOS 7C02v31 (v.3.1, with 1.0.0.3AB). No change in Prime95.

I would stick with the ABBA UEFI variant, despite the problems the mobo OEM had in properly updating their product's UEFI. Bleh.

The Prime95 rounding error does not seem to be caused by my RAM, but I don't have any other DDR4 to test with (I only have some DDR3 modules).

I would highly recommend sticking to the SmallFFTs test (if you are not already) since it does not stress RAM.

Could it be because the stock cooler is not adequate for this Prime95 task?

Possibly. Might be a bad mount.

Or I have a bad CPU?

Also possible, but less probable than a bad mount or the stock cooler being crap.

Hard to say for sure, but you could run memtest86 to help eliminate RAM errors. If you have 4 passes no errors, it is likely the CPU.

Only Blend and Large FFTs even test RAM. Blend also produces much lower temps than SmallFFTs, so I suspect that he's running SmallFFTs anyway.

The error appears almost right away (under a minute). Almost always worker 3 or 4 fails first.

What it means is you're hitting an unstable clockspeed/voltage combo right off the bat. I have a plan to deal with that. In fact, let's try it now.

Go into the UEFI. Make sure LLC settings are OFF, or at whatever setting corresponds to OFF (LLC 0). In my UEFI it is LLC "Standard". LLC "Auto" does . . . odd things, and I do not recommend it. Again, make sure LLC is off! Then set a negative voltage offset of -.1v in the UEFI. Do not set a static offset. Leave everything else stock. Then start rerunning your applications. Watch temps and clockspeeds, and compare to what you got with everything bone stock. I think you'll be pleased, but I could be wrong. Also be sure to compare benchmark results between runs! For example, compare your CBR20 score before and after the LLC/offset tweak. If the score gets lower with the tweak, then we may be seeing clock stretching, which means the offset is too aggressive.

In Cinebench R20 (during the multithread test) the CPU temperature gets to about 82 Celsius.

It shouldn't get that hot. Though I've heard of the stock cooler letting the CPU get up around 80C while running Blender for awhile, so . . . anyway, try the tweaks I recommended, see if that helps.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
Yeah, I did not mention this. In Prime95 I tried Small FFTs and also Blend. Same issue in both cases.

Don't forget that the Ryzen 5 3600's Wraith Stealth stock cooler is quite small. Not much bigger than the Intel stock cooler from my former i5-4440. Also it does not have a copper core.

Because at idle I get normal temperatures (about 35 Celsius / 96 Fahrenheit) I assume that my cooler is installed correctly.

as1SLlR.jpg


Because of the major issues with the only ABBA BIOS available for now for my motherboard (only a newer, but still Beta BIOS also has ABBA), I do not want to use that.

I disabled XMP.

For Load Line Calibration, I don't have an "OFF" setting. Only "Auto" and "Mode 1", "Mode 2", ... , "Mode 8". No explanation what each mode means.

1vWXpJ3.png


TW8uXwZ.png


For a voltage offset of -0.1V, would this setting be correct?

4kBl9Sv.png
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
Well, I tried with that -0.1V voltage offset setting (as in the above screenshot) and with LLC set at Mode 4.

Started Prime 95 Small FFTs. Both workers 3 and 4 gave that rounding error instantly.

Then I reverted the BIOS settings to Auto and I disabled Cores 2 and 6 (the weakest cores from each CCX), using Ryzen Master.

It asked for a Windows restart.

I tried Prime95 Small FFTs again. It did not gave that rounding error, but the CPU temperature raise quickly in 2 minutes, up to about 92 Celsius, so I stopped the test because of that.

So I am thinking I have a bad CPU. More exactly a sub-par Core 02 from it.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
@Seba

It appears you want Mode 1 LLC. That would be the lowest setting. Also, start tracking your vcore during benchmarks with CPU-z and let us know what it says. It is possible that your CPU is bad, though. Even with the tiny cooler, it shouldn't fail on . . . anything?

(btw, idle temps aren't necessarily an indication of anything, but since you're using the stock cooler I think we can dismiss the possibility that you applied TIM incorrectly since it's pre-applied)
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
In the meantime I took another approach at testing.

I set the CPU Ratio to 20 (in BIOS). So all cores locked at 2GHz.
Everything else is at default, except that I also re-enabled the memory A-XMP profile.

I ran Prime95 Small FFTs for 20 minutes. CPU temperature stabilized at 58 Celsius. No issues.

Now I raised the CPU ratio to 30, so all cores are locked at 3GHz. The CPU temperature seems to have stabilized to 69 Celsius now (I have 10 minutes from the start of this test at 3GHz). No errors so far.

Edit

I have over 40 minutes at 3GHz. CPU temperature still about 69 Celsius. No errors.
At 3GHz, Ryzen Master shows Peak Core Voltage 1.09999V and Average Core voltage the same.
Closed Ryzen Master and started CPU-Z. This shows a Core Voltage of 1.104V.

With the CPU Ratio at auto, from what I remember, the Core Voltage was about 1.3 something.

I am going to test at 3.6GHz now.

Edit 2

During this run at 3.6GHz, the CPU temperature oscilated between 82 Celsius and 90 Celsius (it increased, then it decreased, then it increased again and so on; when I stopped the test, the temperature was just under 83 Celsius).
Ryzen Master shows a Core Voltage of 1.1V.
CPU-Z shows 1.2V.
No errors after 30 minutes.
I stopped the test because of the high temperatures.
 
Last edited:

Seba

Golden Member
Sep 17, 2000
1,480
134
106
I set back the CPU Ratio to Auto.

Until now I always had Precision Boost Overdrive to default, which is "Auto". No idea what "Auto" means in this case. The options are Auto, Disabled, Enabled and a few others:

TBV2RJj.png


Just in case, I set PBO to Disabled. Same errors as with PBO on Auto.

Then I also changed CPU LLC to Mode 1 and this does make a difference.

I could run Prime95 Small FFTs for 28 minutes until I got a rounding error (it was worker 3).

The CPU temperature was high (90 Celsius).

The Core voltage is indeed lower now (Ryzen Master shows about 1.18V; CPU-Z shows 1.192V/1.2V - during this Prime95 test).
All cores boost to around 3690MHz (lower than with LLC at Auto).

So what to make of this?

Should I keep this CPU or should I return it?

Now it is still easier to return it, but if I get another one it might be the same (or could be even worse).
On the other hand, if the CPU is sub-par, but still (mostly) working, later will be harder or even impossible to return it.
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
It sound to me like the best course of action is to get an aftermarket CPU cooler, and see how it does. Being without a CPU for weeks or more is not easy. You could use the lower temps anyway.
 
  • Like
Reactions: Drazick

Seba

Golden Member
Sep 17, 2000
1,480
134
106
Since I don't intend to overclock the CPU, my plan was to stick with the stock cooler, at least for a while.

I might get an aftermarket cooler if there is a good price during Black Friday (which is not that far away).

I am looking at "tower" air coolers with one 120mm fan and which do not extend above the memory slots. And without LEDs, if possible.

Such as (sorted by local price, from low to high):
  • Deepcool GAMMAXX 400 (right now this is quite cheaper than the others from this list)
  • ARCTIC AC Freezer 34 eSports (1.48 times the price of GAMMAXX 400)
  • Cooler Master Master Hyper 212 Black Edition (1.91 times the price of GAMMAXX 400)
  • be quiet! Dark Rock Slim (2.17 times the price of GAMMAXX 400)
  • Noctua NH-U12S (2.87 times the price of GAMMAXX 400)
The last one is only an option if it gets a big discount.
 
Last edited:

Seba

Golden Member
Sep 17, 2000
1,480
134
106
I returned my Ryzen 5 3600 CPU.

Not sure if I should also return the MSI B450 Tomahawk MAX motherboard or not. I have a few more days in which I can return it.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
I returned my Ryzen 5 3600 CPU.

Not sure if I should also return the MSI B450 Tomahawk MAX motherboard or not. I have a few more days in which I can return it.

Probably the best course of action. It is evident that the chip can stabilize in AVX2 workloads, but for whatever reason, with the stock cooler, it won't do so using stock settings, which is highly abnormal. The boost map is supposed to pick a low enough clockspeed to maintain stability no matter what the workload.

That being said, your CPU temps were easily 5-10C hotter than many of the R5 3600 reviews I've read using the stock cooler. What's the net airflow through your case? Also, if you get another 3600, I do recommend a better HSF. At least the Gammaxx 400.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
With the other benchmarks I used (such as CineBench R20 or Blender Benchmark) the CPU temperature did get high, but not as high as with Prime95.

I have a Fractal Design Define R5 case.

It has two 140mm fans at the front, as air intake and one 140mm fan at the back, as exhaust.

The power supply (Seasonic S12G-650) is installed with the fan down, at the bottom of the case (so it draws air from outside of the case).

The graphics card is GTX 1060 6GB (Asus Strix, 3 axial fans) - not a major source of heat inside the case (at least not during Prime95 tests).

All drive cages are still in place inside the case (all are removable).

I have two 2.5" SSDs and two 3.5" HDDs.
 

DrMrLordX

Lifer
Apr 27, 2000
21,582
10,785
136
It has two 140mm fans at the front, as air intake and one 140mm fan at the back, as exhaust.

Okay. Doesn't look like you've got another major heat source, especially not during your Prime95 benching. And 2x140mm fans should be able to supply enough ambient air to the HSF to prevent it from re-uptaking its own exhaust to any considerable extent. But those CBR20 temps seemed . . . abnormally high even still. The last 3600 review I saw showed a Blender temp of 80C after a solid hour of testing, and Blender is a bit more intense than CBR20's MT test. 82C after what is basically less than a minute of CBR20 seems a bit much. Something isn't quite right there. Not sure if it's the board overvolting the CPU (which is why I recommended LLC1 + negative voltage offset) or a faulty CPU. But if you can't get Prime95 SmallFFTs stable even with -.1v offset and LLC 1 then I'm thinking the CPU may just be a dud. Might be cracked solder under the IHS for all I know.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
During Blender Benchmark (quick and full), the CPU temperature was 80 something Celsius. I think under 85 Celsius.

Edit:
Ah, and when I run Prime95 with LLC set as Mode 1, I did not also had that -0.1V offset anymore.

I previously reverted it to Auto (I restored the BIOS to defaults) and I wanted to first try only with LLC set at Mode 1.
I did not expect to last that long without errors. After that 28 minutes run it was too late to try another test with the voltage offset.
 
Last edited:

VirtualLarry

No Lifer
Aug 25, 2001
56,229
9,990
126
That being said, your CPU temps were easily 5-10C hotter than many of the R5 3600 reviews I've read using the stock cooler. What's the net airflow through your case? Also, if you get another 3600, I do recommend a better HSF. At least the Gammaxx 400.
With the other benchmarks I used (such as CineBench R20 or Blender Benchmark) the CPU temperature did get high, but not as high as with Prime95.
I tried to point out the nearly obscene temps of these CPUs, running PrimeGrid (or Prime95 with AVX/AVX2) workloads, they really push these CPUs to the edge and beyond, thermally, with "normal" heatsinks.

I went through three different air heatsinks, including a 125W AMD 4-heatpipe copper-cored heatsink, still had crazy 95C throttling temps. Mine didn't have computation errors, though.

Finally, moved my R5 3600 into my rig with a 240mm AIO WC kit, where it's mostly happy.

OP, get 240mm water for your R5 3600, or go home. Probably best that you returned that CPU though, sounds like a dud. These 3600 CPUs get the "worst" bins, pretty-much, as far as leakage and temps and whatnot.

Edit: I was also getting 56C idle temps on mine, before I put it under water, and before the 1.0.0.3 AB/ABB/ABBA BIOSes came out.
 
Last edited:

Seba

Golden Member
Sep 17, 2000
1,480
134
106
These 3600 CPUs get the "worst" bins, pretty-much, as far as leakage and temps and whatnot.
I am suspecting the same thing. I mean it's to be expected that they would put the worst bins on Ryzen 5 3600, but these might be not just marginally worse, but only borderline functional.

Maybe this is where Ryzen 5 3600X (although almost the same on paper as Ryzen 5 3600), has an edge?

But 3600X is 25% more expensive than 3600.

And reading the cooling sub-forum I found out that AMD cheapened on the Wraith Spire cooler for Zen 2, which now has a aluminium-only heatsink. But the name of the cooler is the same.

I went through three different air heatsinks
Can you list them?
 
Last edited:

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,483
14,434
136
I think this one was suggested before here and recommended https://www.amazon.com/ARCTIC-Freezer-eSports-DUO-Configuration/dp/B07MJGNJB3/ref=sr_1_2?crid=EGQLRD00MRXX&keywords=arctic+freezer+34&qid=1570725891&sprefix=arctic+freez,aps,197&sr=8-2

Edit @VirtualLarry This one looks good and is less expensive, but strangely, the AM4 is the only one missing from this list. Is AMW that different than AM2 or AM3 ? In the mounting area https://www.amazon.com/dp/B07TWQ2VLR?ref_=ams_ad_dp_ovrl&th=1

Edit2: This article suggests that they are identical https://linustechtips.com/main/topi...ting-plates-seems-to-be-backwards-compatible/

That article says only for the clip type. But I think thats what the one above is.
 
Last edited:
  • Like
Reactions: Drazick

krumme

Diamond Member
Oct 9, 2009
5,952
1,585
136
The 3600 runs super fine with stock cooler in not so well ventilated case in one of my kids gamer machines. Some cheap asus b350 board. This cpu is ultra lean.

If there is computational errors I would just return cpu and mb and get new. No reason to use to much time on it. It should work out the gate no probs at all. And absolutely with stock cooler. Easiest setup ever. Plug and play.

Remember to try standard bios settings first without doing any stuff, if you havnt tried that. Reset with newest bios.
 

B-Riz

Golden Member
Feb 15, 2011
1,482
612
136
Since I don't intend to overclock the CPU, my plan was to stick with the stock cooler, at least for a while.

I might get an aftermarket cooler if there is a good price during Black Friday (which is not that far away).

I am looking at "tower" air coolers with one 120mm fan and which do not extend above the memory slots. And without LEDs, if possible.

Such as (sorted by local price, from low to high):
  • Deepcool GAMMAXX 400 (right now this is quite cheaper than the others from this list)
  • ARCTIC AC Freezer 34 eSports (1.48 times the price of GAMMAXX 400)
  • Cooler Master Master Hyper 212 Black Edition (1.91 times the price of GAMMAXX 400)
  • be quiet! Dark Rock Slim (2.17 times the price of GAMMAXX 400)
  • Noctua NH-U12S (2.87 times the price of GAMMAXX 400)
The last one is only an option if it gets a big discount.

The Gammaxx 400 is your best option; I have run it on the 3900X and 3600 no issues, it covers the whole cpu, unlike the Freezer 34 and does a better job than the Hyper 212.

I would get a Gigabyte X570 board if you are thinking of getting a new board.

Here is my testing of the 3600 with Gammaxx 400 on a B350 board.

 

Venoom

Member
Nov 10, 2009
113
0
76
hi everyone,

I am running a Ryzen 3900X (motherboard: ASUS TUF Gaming X570-Plus) inside a be quiet! Dark Base 700 case.
Stock cooler and stock fans.

From the beginning, I noticed some weird things. Performance is fine (at least according to UserBenchmark).

But idle temps are 55° to 60°C (!!) and core voltage (according to the BIOS) 1.47V (!!!!!)

I tried googling and found some contradicting information.

to be safe, I offset the voltage by -0.15V. Temps in the BIOS went down to 45°-50° and performance wasn't really affected (at least according to UserBenchmark)

Should I be worried?
 

Hans Gruber

Platinum Member
Dec 23, 2006
2,092
1,065
136
I have a brand new in the box CoolerMaster Master Liquid Cooler 280mm. Is that compatible with the Ryzen 3000 chips? Cooler Master sells an adapter for AM4. Do I need one of those adapters or will it work out of the box? Next question. 3600 or 3700x or maybe even a 2600 or 2700X and wait out the silicon on Zen 2 until early next year. 16GB of ram or 32GB of ram? Run the ram @ 3200mhz or 3600mhz? I think I will go with the Asus Rog Strix X470-F motherboard.

I have a Ryzen test system with an 1200 with 240mm water cooler and B350 motherboard. I could go the route of a 2600 or 2700x route and pass the CPU onto my other system.
 

Seba

Golden Member
Sep 17, 2000
1,480
134
106
The 3600 runs super fine with stock cooler in not so well ventilated case in one of my kids gamer machines. Some cheap asus b350 board. This cpu is ultra lean.
Can you try the latest Prime95 Small FFTs (without disabling AVX/AVX2) and with 12 workers on that PC and report the results (if you get errors or not and what temperatures you have - temperatures as reported by Ryzen Master)?

I would get a Gigabyte X570 board if you are thinking of getting a new board.
X570 boards are out of the question. I won't pay more for the motherboard than for the CPU.

The Gammaxx 400 is your best option; I have run it on the 3900X and 3600 no issues, it covers the whole cpu, unlike the Freezer 34 and does a better job than the Hyper 212.
Interesting. No review of Artic Freezer 34 eSports I saw or read mentioned the small size of the base.
 
Last edited: