News YouTuber resuscitates unstable Ryzen 7 5800X for $30 — a 300 MHZ underclock saved the day

[Admin]

A YouTuber fixes a degrading Ryzen 7 5800X by downclocking the CPU by 300 MHz in the BIOS.

YouTuber resuscitates unstable Ryzen 7 5800X for $30 — a 300 MHZ underclock saved the day : Read more

 

[punkncat]

It is premature to call that CPU "good" as is well documented in their own video. It discusses that he will still need to go forward with more testing to find out which core is unstable.

I really enjoy watching Bry and see his channel evolve with the times.

 

[umeng2002_2]

Is this article just a Chat GPT summary of some YouTube video?

 

[bit_user]

A CPU should never degrade within several years of use, let alone 12 months or less.

Overclocking can kill a CPU even faster than that. We don't know what kind of abuse it suffered, at the hands of its original owner.

 

[rluker5]

Overclocking can kill a CPU even faster than that. We don't know what kind of abuse it suffered, at the hands of its original owner.

Too many volts are bad. It seems like tech media is caught by surprise by this every time it happens.

 

[pug_s]

Dang, I brought this CPU about 8 months ago, and have no problems when gaming while cpu is pegged 100%. I hope my cpu won't degrade like this guy. Then again, now AMD released a Ryzen 5800XT with even higher clocks than this one and don't know if they fixed the quality of silicon issue.

 

[Notton]

My 5800X3D is perfectly happy with whatever PBO gives it. I've been using it as is for over a year with zero issues.
That 5800X is either a heavily abused CPU, or the owner has terrible luck with CPU lottery.

I could have sworn there was a way to disable a specific core in BIOS on AM4 mobos, even cheap ones.

 

[Alvar "Miles" Udell]

Is this article just a Chat GPT summary of some YouTube video?

The ironic thing is that if you read this article you don't have to watch the YouTube video. If it's monetized then you've denied him money, the same thing AI summaries from Gemini and Copilot are being accused of doing.

Ryzen CPUs don't have a reputation for reliability problems

Ryzen 1000 series owners, me being one of them, will disagree with you.

 

[NightForce]

Well, I've run a 5800X almost three years and never had an issue. But I run at stock clock and voltage. The Ryzen auto-overboost wasn't able to give me any extra. Guess I lost the bin lottery. But the proc works great. It is ridiculously fast at FLAC encoding, the main thing I do for work.

All that said, I have watched Tech Yes City for years as well, so much I know his ad reads by heart. He's a good dude with a super knack for finding deals especially on distressed hardware. Have not yet seen his 5800X video but I have no doubt it is legit.

 

[SkyBill40]

Well, I've run a 5800X almost three years and never had an issue. But I run at stock clock and voltage. The Ryzen auto-overboost wasn't able to give me any extra. Guess I lost the bin lottery. But the proc works great. It is ridiculously fast at FLAC encoding, the main thing I do for work.

All that said, I have watched Tech Yes City for years as well, so much I know his ad reads by heart. He's a good dude with a super knack for finding deals especially on distressed hardware. Have not yet seen his 5800X video but I have no doubt it is legit.

I, too, run a 5800X and have had great success with it for several years now and no issues. While I've not tried to OC it with PBO or Ryzen Master, I'm perfectly fine with it at stock. It chews through anything I throw at it and it was a massive upgrade over my old CPU.

 

[tamalero]

Overclocking can kill a CPU even faster than that. We don't know what kind of abuse it suffered, at the hands of its original owner.

This.. and with more and more motherboards pushing way farther than ever just to be "faster" than the competition..

The ironic thing is that if you read this article you don't have to watch the YouTube video. If it's monetized then you've denied him money, the same thing AI summaries from Gemini and Copilot are being accused of doing.



Ryzen 1000 series owners, me being one of them, will disagree with you.

This.. funnily.. in my case both with Ryzen and threadrippers..
Almost all the issues were memory related (unless you had those golden Samsung B Dies)
And some motherboard bioses being dog poop.

I still remember my threadripper having to HAVE the floppy option in the bios enabled to make it stable.
Why? absolutely no idea!
Turned that thing off? nope.. would crash with memory issues.. this was an ASUS mobo btw.

 

[wwenze1]

I had a Athlon 1.4GHz that was only stable at 1050MHz. But we didn't have YouTube back then.

Today I have a 6400 that is only stable with all C states and variable frequency turned off.

 

[gggplaya]

I bought a 5800x when it was first released. It was never stable and overbinned from the factory. All defaults settings, no overclock. I had to manually clock it to 4.4ghz to prevent it from boosting too high, otherwise it would bluescreen with whea uncorrectable errors. After a year of trying different things like ram and motherboard, I RMA'd it back to AMD and they sent me a replacement that was stable.

People are assuming this chip was overclocked heavily and degraded, it likely was overbinned from the start.

 

[JRStern]

Jeepers. Sounds like a ton of this stuff has been going on. But these kinds of problems may only get worse if circuit geometries keep shrinking. I've wondered for a long time how modern chips, for the most part, have been pretty durable. And motherboards. And power supplies. Even the fans! Stuff that 30 years ago all tended to have limited lifetimes. But especially circuit degradation over time, at ludicrously small nanometers, and now with these micro-connects between chips. One cosmic ray and blammo, you lose a core, or at least tolerance for highest frequency?

If it's going to spread may have to build in these kinds of self-tests and dropping a core or trying reduced clock in user-friendly bios or OS or super-tuner-app or something.

 

[bit_user]

People are assuming this chip was overclocked heavily and degraded, it likely was overbinned from the start.

Well, that depends a lot on when it was made.

Ryzen 5xxx B2 revision / stepping - post here your...

As title sais, I am interested in documenting the release dates of this B2 revision, so if you got a Ryzen B2, please post here your manufacturing date. For the ones who don't know how to spot it, it's the first four numbers fromt he second row, directly on the CPU.

www.overclock.net

I bought a 5800X in January 2023 and it's B2 rev. It has no trouble clocking up to 4.85 GHz and exhibits no stability problems. I did some stress testing on it, but the only setting I changed was putting PPT at 142 W (the motherboard BIOS defaulted to 105 W).

In addition to wondering whether the CPU in this article was abused, I'm also pretty curious which chip revision it is.

 

[bit_user]

these kinds of problems may only get worse if circuit geometries keep shrinking. I've wondered for a long time how modern chips, for the most part, have been pretty durable.

Yup, same. I heard that Intel used to have some kind of durability policy that their process nodes had to meet. Something like their processors having to withstand 10 years of continuous operation, or something like that. Obviously, with Raptor Lake, we see that even if that policy still applied to the design of the process node, the follow-through was lacking to ensure the finished CPUs would hold to that standard.

I'm having trouble locating a good source on that 10-year thing, but you can see it alluded to in this plot of EM (Electron Migration) rates for different interconnect materials:

Source: https://www.semiconductor-digest.co...ried-and-tested-copper-with-cobalt-liner-cap/
​

@thestryker , do you have any source on this "10 year" standard, for their process nodes?

And motherboards.

AFAIK, the main thing that would shorten the life of motherboards is the capacitors failing. Higher-end motherboards tend to use better caps.

And power supplies.

Same. Also, there's been a mini arms race between PSU manufacturers on their warranty periods, for higher-end models. I've seen some with warranties as long as 12 years!

Even the fans!

I have Noctua fans running probably 50% of the time, for over a decade, and no failures. But fans are no big deal to replace, if one dies on you.

now with these micro-connects between chips. One cosmic ray and blammo, you lose a core, or at least tolerance for highest frequency?

I'm not sure they're that sensitive. Anyway, cosmic rays are quite rare, at least at sea level.

Plus, the amount of metal it has to get through is non-trivial. I'm no expert on the subject, but if that were a frequent occurrence, it should make it difficult for manufacturers to offer a decent warranty on CPUs. Speaking of which, I wonder what the warranty is like on Xeons and EPYCs, these days.

If it's going to spread may have to build in these kinds of self-tests and dropping a core or trying reduced clock in user-friendly bios or OS or super-tuner-app or something.

Funny you should mention that, because Intel has recently added a feature called "In-Field Scan" to its server CPUs. That could be more of a necessity with high core-count CPUs, than anything else.

 

[pug_s]

I bought a 5800x when it was first released. It was never stable and overbinned from the factory. All defaults settings, no overclock. I had to manually clock it to 4.4ghz to prevent it from boosting too high, otherwise it would bluescreen with whea uncorrectable errors. After a year of trying different things like ram and motherboard, I RMA'd it back to AMD and they sent me a replacement that was stable.

People are assuming this chip was overclocked heavily and degraded, it likely was overbinned from the start.

My guess is that AMD released 5800x first and a 5700x a few months down the road when your cpu should be binned as a 5700x or even less. I am glad that they released a B2 version of this and since I brought it new less than a year ago, hopefully I got this one.

 

[bit_user]

My guess is that AMD released 5800x first and a 5700x a few months down the road when your cpu should be binned as a 5700x or even less.

The 5700X came at least a year later. I forget whether or not it was after B2, but that revision might've been what enabled the 5700X.

I almost bought a 5700X, but then I saw the 5800X on sale and figured I could always dial back the max clock speed and PPT to match a 5700X, if I found it to be too power-hungry.

I am glad that they released a B2 version of this and since I brought it new less than a year ago, hopefully I got this one.

There are probably programs you can use to find out which you have.

 

[bit_user]

BTW, it seems the warranty on retail-boxed Ryzens is 3 years.

https://www.amd.com/en/resources/support-articles/warranty/RMA-03.html
​

So, I wouldn't worry too much about a premature failure, however I do think using PBO violates the warranty terms. Maybe that's why the previous owner didn't just RMA it?

 

[thestryker]

@thestryker , do you have any source on this "10 year" standard, for their process nodes?

I'm pretty sure most are expected to last longer than that, but I believe the decade mark is what Intel has used for their accelerated wear testing.

This is the only somewhat recent article I can think of which talks about durations, but doesn't get into specifics: https://www.anandtech.com/show/15839/electromigration-amd-ryzen-current-boosting-wont-kill-your-cpu

Roman talked to one of Intel's engineers who also touched on the subject: YT: Current CPUs are Overheating?

Anecdotally speaking my old SNB server box ran for over a decade 24/7 without issue (stock the entire time with ECC memory).

 

[bit_user]

Anecdotally speaking my old SNB server box ran for over a decade 24/7 without issue (stock the entire time with ECC memory).

Yeah, among the hundreds of Intel machines we've had at my job, I'm not personally aware of any CPUs definitely failing. However, sometimes there'll be a server that's throwing ECC memory errors on a certain DIMM, even after you replace the DIMM, and I never went further to track down whether it was the CPU or the motherboard that went bad, because the machine was already too old to bother with.

In another instance, I remember one CPU of a 2-CPU server throwing an error of some kind, but I never tried swapping the CPUs into different sockets to see if the error followed the CPU or not. Again, the machine was already about a decade old, so I didn't really care enough to follow it up.

So, those are two possible examples of CPU-level errors, in machines that ran continuously for 6 and 11 years, respectively. The 6-year-old one ran some pretty heavy compute workloads. The 11-year one ran medium/light-duty compute, most likely.

Before the whole Raptor Lake debacle, I always had the sense that Intel CPUs were better engineered to withstand the rigors of long-term usage. Nowadays, and especially with these new nodes and advanced packaging technologies, I'm just not sure.

I will say that I had no interest in a X3D CPU for my server, since the extra die just seemed to me like one more potential point of failure. However, since the 5800X3D seem to be holding up pretty well, that gives me hope that my 5800X will be reliable. I also like to avoid running it much above 80 C, just to be on the safe side. I guess I should also back off the peak clocks by a couple hundred 100 MHz and see if that avoids it seeing the highest voltages.

 

[thestryker]

Before the whole Raptor Lake debacle, I always had the sense that Intel CPUs were better engineered to withstand the rigors of long-term usage. Nowadays, and especially with these new nodes and advanced packaging technologies, I'm just not sure.

I think RPL is a great example of a company playing fast and loose to plug holes and shoving something out the door without doing enough testing. There is a lot more that can go wrong these days though given as you say advanced packaging and just the new things they're doing to keep processes shrinking.