Ryzen 3900x fails prime 95 with stock settings

commissioneranthony

[H]ard|Gawd
Joined
Jul 2, 2008
Messages
1,274
Hi. I recently purchased a used 3900x for a new PC build. I think there is something wrong with it.

Setup:
-ryzen 3900x
-asrock b550i (brand new, 1week old)
-(2) 16gb corsair 3200mhz LPX memory (brand new, 1week old)
-evga 550gm PSU (brand new, 1week old)
-intel NVME
-noctua u9s
-arctic mx-2
-quadro p400

I installed windows and did a quick prime95 test on the CPU at stock settings. No overclocking was performed on any component. My goal was a quick CPU validation test. The system failed prime95 dramatically. Many errors resulted, mostly rounding errors. Is this normal with Ryzen CPUs? Somehow I don't believe a CPU is supposed to fail prime95 at stock settings. I used a noctua u9s with two 92mm fans in a push-pull configuration. New Arctic mx-2 was applied.
Another oddity was the temperature reporting in HWinfo. At idle, the temperatures would fluctuate from 134f to 152f. This is the machine sitting idling with no programs running sitting at the desktop. I have screwed the noctua down to the cooler properly and there is definitely good mounting pressure as the screws have fully seated.

In my many years of PC building, I never had a cpu fail prime95 on me. No PII, PIII or P4 ever errored at stock on prime95. Neither did any athlon64 or phenom.

These are some official errors reported:

"FATAL ERROR: Rounding was 0.4885018486, expected less than 0.4
Hardware failure detected, consult stress.txt file.
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.
[Mon Jul 20 13:08:59 2020]"


During prime 95 HWinfo was reporting reporting temperatures from 150-170f

I find it rather strange. The seller also convienantly deleted the content of the original reddit for sale post and commented on two other 3900x build for sale posts shortly after the sale, looking for a CPU. I smell a scammer. I plan to ask for a full refund. What would you do? Thanks.
 
Last edited:
It shouldn't fail at stock settings regardless so there may be something to a defective or in your case a degraded chip. I would run the PBO maxed out test with Prime. See what the voltage averages under PBO. My guess is that the voltages will be very very low. The silicon FIT controller will raise or lower voltage to a safe point and if the chip is degraded it won't be able to handle stock voltages. A healthy chip will average 1.3v give or take.

Your temps are ok at load, 170f = 80c. I recommend you reference Celsius as that standard, saves ppl time having to convert.

Also, as mentioned above it could be memory... as Corsair is not the best ram for AMD Ryzen. Set the memory to a base speed of 2133mhz, not docp/xmp to remove it as a possible issue.
 
I have the XMP profile in bios enabled. The motherboard is on stock bios, I have not updated it since I received it in the mail.
I am going to try disabling xmp and see if that changes the results.
What is a "PBO maxed out test" ?
I have this exact ram:
https://www.newegg.com/corsair-32gb...air ddr4 memory 2x16gb-_-20-233-894-_-Product
It is unfortunately nanya dram which I find rather humorous. I expected more from corsair.
Thanks.
 
I have the XMP profile in bios enabled. The motherboard is on stock bios, I have not updated it since I received it in the mail.
I am going to try disabling xmp and see if that changes the results.
What is a "PBO maxed out test" ?
I have this exact ram:
https://www.newegg.com/corsair-32gb-288-pin-ddr4-sdram/p/N82E16820233894?Description=3200mhz corsair ddr4 memory 2x16gb&cm_re=3200mhz_corsair ddr4 memory 2x16gb-_-20-233-894-_-Product
It is unfortunately nanya dram which I find rather humorous. I expected more from corsair.
Thanks.
disable XMP profile and run the memory at default. Sometimes even to this day Ryzen CPUs can be fickle with memory.
 
There is a BIOS update V1.10, but yours may have shipped with it.
If XMP disabled works fine, then you need to spend some time tuning your RAM.
 
Make a bootable memtest flash drive and start diagnosing from there. If it fails then I would disable xmp. If it still fails I would test one stick \ slot at a time to see if you have a bad stick or slot on the mobo.
 
I suspect memory. Buy something like G.Skill if you can send the Corsair back.

I don't see why the seller would send you a defective CPU when he could obviously return it to AMD for replacement. It's definitely still under warranty.
 
Failing Prime95 dramatically (and very quickly) is exactly how I diagnosed my defective 3700x. It's always smart to look for other problems first (memory, power supply, etc) of course but if Memtest is passing and Prime95 is consistently throwing errors on the same threads, it's very likely a faulty CPU.
 
Last month I had a friend have his 3700X fail prime95. My first thoughts were memory instability so I had him run multiple instances of hclmemtest until he had almost all ram under the test.

https://hcidesign.com/memtest/

Within a few minutes of starting HCL failed the test so I had him reset the speed to stock and the test passed. Then we had him reseat the ram and it passed. Then the XMP passed and finally Prime95 passed.
 
Hey guys. Thanks for all the input. I disabled xmp profile and just set it to auto in bios. Prime95 ran without error for 50minutes. I decided to stop the test there
I'm doing another clean win10 install with all the appropriate drivers and going to run the test for a few hours.
I'm also going to give the amd utility a try
Is there any other utility I can run that can verify that my CPU is not messed up? I used to leave prime95 run for a solid day when overclocking P4's of yesteryear :p
Thanks for the input guys :D
 
Hey guys. Thanks for all the input. I disabled xmp profile and just set it to auto in bios. Prime95 ran without error for 50minutes. I decided to stop the test there
I'm doing another clean win10 install with all the appropriate drivers and going to run the test for a few hours.
I'm also going to give the amd utility a try
Is there any other utility I can run that can verify that my CPU is not messed up? I used to leave prime95 run for a solid day when overclocking P4's of yesteryear :p
Thanks for the input guys :D
OCCT has a good stress test you can use too, but if you're able to run Prime95 for nearly an hour I think you're fine. To try to stabilize the RAM with XMP enabled, you could try bumping the voltage up to 1.4V, and also try switching VDDCR SOC Voltage to offset mode (something I had to do.)
 
Bump your SOC voltage slightly (like 1 step over stock). Might help you run XMP with your memory.
 
XMP can be tweaked fairly easily by just relaxing the timings and boosting voltage to numbers known to work for others on the board.
 
My RAM has NEVER been stable at its XMP setting with either my Ryzen 1400, 1700 or 3900X.
I just accept that its necessary to tweak RAM these days.
 
Probably bought one with a weak IMC, not defective but just picky with timings. My launch 3900x had this issue.
 
Update that bios. Even with my newer x570 type boards I initially had issues with XMP till i updated bios. Yes B550 is newer but still you should start with updating that bios as that may be all you need to fix your XMP issues from what I have found.
 
Also the fact that you can post and get into windows with XMP means you should be able to get it XMP stable with some tweaking as suggest above
 
Thanks for the input guys. Switched from XMP profile to "auto" in bios and the rounding errors in prime95 went away. I made a new win10 pro install with all the necessary drivers. Prime95 showed no errors for 5 hours.
Currently running memtest but it never shows the results. Ran memtest for 8 hours and the desktop stopped displaying video output. machine running but no display. I reran prime95 about an hour ago. same issue no video output I can't see the results. Tried hitting ESC + y (to save results as html file) but it never exits memtest.
This does not seem normal for memtest. I always remember seeing the results after a memtest run. Well, I'm guessing this ram is going back to newegg.
I'm updating my bios next and then giving OCCT a shot. \
I'm thinking of grabbing a pair of these as replacement ram, but I'm still looking.
https://www.newegg.com/g-skill-32gb-288-pin-ddr4-sdram/p/N82E16820232413

Thanks for all the help guys!

Edit: memtest backscreen likely caused by LCD being unplugged and displayport not syncing when monitor was turned back on. redoing memtest results with motnitor on.
 
Last edited:
What is the graphics card you are running? Sounds like the typical black screen crash.
 
I'm using a Quadro p400.
Just updated bios to v1.1. Bios picks up ram as ddr4-2133 on "auto". I manually set to ddr4-3200. XMP profile not enabled.
Windows randomly crashes when ram set manually to 3200mhz. Just opening win10 start menu crashed windows 10. A brief BSOD error message mentions it's a memory error.
I put ram back at auto ddr4-2133. Previous successful prime95 test done at auto ram settings of ddr4-2133.
I am currently running Windows Memory Diagnostic, mdsched.

I have a feeling this ram is crap. Corsair should really stop investing resources in gaming chairs and put it back in memory, what they used to be known for :eek:
I am really hoping the CPU does not have a bad memory controller.
 
Last edited:
up your ram voltage to 1.4v. every corsair set ive ever used needed more juice to be fully stable.

pretty much this.. if it's 3200c16 memory it's likely hynix which for some odd reason needs 1.37-1.38v to actually run XMP. no clue why and like pendragon i've had to do it with every hynix m die memory kit whether corsair or gskill.
 
I'll give that a shot. This is nanya corsair ram. CPU-z mentioned it's nanya dies. This is the exact ram I purchased:
https://www.newegg.com/corsair-32gb-288-pin-ddr4-sdram/p/N82E16820233894?Item=N82E16820233894

if you look on the modules themselves there should be a version number and that'll tell you exactly what modules they're using since corsair uses memory from all the chip manufactures within the same sku so you can actually get completely different modules buying 2 different kits of the same model sku.
 
Update: Ran memtest86 on the corsair ram at non xmp profile (2133) and no errors resulted. I was going to try the 1.37-1.38v trick, but I have a day left on the return from newegg so I'm claiming the return.
I just snagged a kit of Ballistix Sport LT 32GB (2 x 16GB) DDR4-3200 PC4-25600 CL16 ram instead. I think I will have a lot more luck with that getting it closer to XMP profile.

Thanks everyone for all the input!
 
My RAM has NEVER been stable at its XMP setting with either my Ryzen 1400, 1700 or 3900X.
I just accept that its necessary to tweak RAM these days.
Same here. Expensive DDR4 4000 memory won't even boot at XMP. I didn't want to run at 4000, but even with the fabric at 3600, it still won't boot. The timings from the Ryzen memory tool will not work either. I'm running at 3600 with timings I got from someone with a gigabyte board and I'm using 1.40v.
 
Last edited:
I got a 3900XT and x570 Aorus Master last weekend, and ran into similar issues with my Corsair 2133 (Vengeance RGB PRO XMP3200 4x16GB). Running the XMP profile would result in random reboots and blue screens - didn't matter if the if the system was under heavy load or not.

I manually set the timings and frequency to the XMP SPD recommended @ 1.35V, and it's been rock solid so far. I did notice that if I had the XMP profile enabled, it was mismatching the proper tRC settings.
 
Update: Grabbed a kit of Ballistix Sport LT 32GB (2 x 16GB) DDR4-3200 PC4-25600 CL16 ram and it worked at XMP profile right off the bat. 8 hours of prime95 and no errors.
memtest up next.
Thanks again for all the help guys. It means a lot. :)

Nice. It's a good thing it was the shit ram instead of the cpu. Btw, you just went thru 8 hours of P95, memtest is not going to uncover anything new. Give your chip a breather.
 
is this true?...Ryzen CPU's are known to have issues with Corsair memory?

It's not the ram exactly as it's Corsair and the BS moves they do with respect to the integrity of the ICs used. They have a bad habit of switching to cheap ass ICs later in the product cycle, this then creates problems for users expecting a certain IC or quality. Basically avoid their shit like the plague.
 
It's not the ram exactly as it's Corsair and the BS moves they do with respect to the integrity of the ICs used. Basically they switch to cheap ass ICs later in the product cycle, this then creates problems for users expecting a certain IC or quality. Basically avoid their shit like the plague.

I've been using G.Skill for my last few Intel builds and I am planning on using them again for my upcoming Zen 2 or Zen 3 build...never had any issues with them
 
My RAM has NEVER been stable at its XMP setting with either my Ryzen 1400, 1700 or 3900X.
I just accept that its necessary to tweak RAM these days.
Built a 2600x G.Skill rig with over two different types of boards (faulty PSU killed them) and I just ran 3600 at 3200 XMP settings and never had a problem.
 
It's not the ram exactly as it's Corsair and the BS moves they do with respect to the integrity of the ICs used. They have a bad habit of switching to cheap ass ICs later in the product cycle, this then creates problems for users expecting a certain IC or quality. Basically avoid their shit like the plague.
This, my Corsair ram is Samsung b-die, others aren't. The fact that you won't know what you're getting until you buy/open is like playing Russian roulette with your memory. I got some 3000C15 and it came with b-die and runs xmp no issues with my 1600 on a b450. Others aren't so lucky.
 
Back
Top