News Puget says its Intel chips failures are lower than Ryzen failures — retailer releases failure rate data, cites conservative power settings

[bit_user]

I am currently limiting my reviews to CPU coolers, SSD heatsinks, thermal pastes, and cases. While I do give my thoughts on a product, my reviews generally aim to give people the information they need so that folks can come to their own conclusions on whether the product fits their needs.

One of the reasons I've focused on these items is that I'm less likely to deal with toxic fanboyism. Or at least, they were until this weekend.

I have to admit, my reviews are biased....

They're biased towards strong thermal performance and low noise levels

I haven't seen/heard anything about the current "controversy" other than a couple posts in this thread, but I just wanted to take this opportunity to thank you for making yourself accessible in these forums. I appreciate your thoughtful reviews and certainly don't take it for granted that you monitor the threads of your review articles and remain responsive to questions and feedback we pose.

Hopefully, this current nonsense will blow over soon enough and things can return to some form of normalcy. I just wanted to pay credit I feel is due and offer my support.

Thanks!
: )

 

[jkhoward]

Why is it that whenever AMD has “bad” press or good reviews you leave it as the featured topic for weeks. But when Intel has bad news you change it the next day. How much is Intel paying you?

 

[Albert.Thomas]

Why is it that whenever AMD has “bad” press or good reviews you leave it as the featured topic for weeks. But when Intel has bad news you change it the next day. How much is Intel paying you?

I'm genuinely curious. When has this happened? Please provide examples I can look up through the internet archive to verify.

 

[NedSmelly]

It looks though with the downclocking of the shipped CPUs with the conservative profile lower than the "intel performance" recommended profile, they likely don't do undervolting to ensure stability

Yes I don’t think Puget undervolts, they just spec/ship according to box specs (and no XMP).

I’ve followed their blog for a while as a guide for configuring my own media workstations, and AFAIK they are just as irritated as everyone else with Intel being vague about what the box power specs actually are.

 

[russell_john]

My office had Puget CAD/CAE workstations with a consumer level Intel 13900k chip and two rtx a4000s. Their settings perform significantly below just slapping a 13900k on an Asus z790 in multithreaded workloads. In fact, they don’t even stay at PL2 indefinitely.

That's just because stability is more important in professional applications than raw performance. The same is true for the A4000s, same die as the 3070 Ti yet a 3070 Ti will smoke it ..... However you can't stack 3070 Tis and they aren't nearly as efficient as an A4000 or as stable. the A4000 is so efficient you can run two of them and still use less electricity than a 3070 Ti

 

[YSCCC]

Yes I don’t think Puget undervolts, they just spec/ship according to box specs (and no XMP).

I’ve followed their blog for a while as a guide for configuring my own media workstations, and AFAIK they are just as irritated as everyone else with Intel being vague about what the box power specs actually are.

Yea and for whatever reason, I've tried the "newer" 0x125 microcode on my own rig, it ended up by default pumping 1.438V as the VID table of the CPU originally requests and with spikes of 1.441v... and with protection to stop me from undervolting which I've personally tested the 14900k being able to run stable at 1.398V max, with transient to single core capped 1.401V, which then I've disabled the intel profile and manually tuning in own undervolt profile... even if that is really safe, it is frustrated as hell as to having to dig into the bios and data, spending days to test and watch upon parameters just to hope it won't fail, a joke I saw on Buildzoid's ranting video was - "Intel TOTL final i9, shipped at 6Ghz, an experienced hardcore overclocker can make it run~~~~ 5.5ghz stable", If buildzoid was right that it is indeed the transient voltage call could go up to 1.6V+ when waking from idle low C state, lowering the PL and frequency would only slow the death, not fix it.

And pretty sure even Puget are in damage control mode and need to closely monitor any customer dissatisfaction to rise in the coming months.

 

[TheHerald]

Why is it that whenever AMD has “bad” press or good reviews you leave it as the featured topic for weeks. But when Intel has bad news you change it the next day. How much is Intel paying you?

Man you can't be serious. Just check a look at /r intel and /r amd front pages and youll see who's censoring and who isn't. Come on now.

 

[Pierce2623]

That's just because stability is more important in professional applications than raw performance. The same is true for the A4000s, same die as the 3070 Ti yet a 3070 Ti will smoke it ..... However you can't stack 3070 Tis and they aren't nearly as efficient as an A4000 or as stable. the A4000 is so efficient you can run two of them and still use less electricity than a 3070 Ti

I’m fully aware of why thy they tune it that way. I was dimply trying to illustrate for others what it is that they actually do. (Which is mostly just reintroducing the “Tau” figure for a maximum time it will boost to the limits. While you say the 3070ti easily beats the rtx a4000, if I brought the 3070ti down to the rtx a4000’s TDP, the a4000’s tighter bin will have better performance. I also should’ve mentioned that they’re actually rtx a4000 Ada cards. They’re crazy good. 3080 performance in compute applications at 3060 power draw.

 

[thesyndrome]

I find it odd that no one has mentioned how the President of Puget systems is also on the Intel Board of Advisors?

This seems like pertinent information.

 

[jkhoward]

Man you can't be serious. Just check a look at /r intel and /r amd front pages and youll see who's censoring and who isn't. Come on now.

Hey look, now it’s like 8 new stories down. How weird!

 

[COLGeek]

Folks, there are no shills, of any variety, at Tom's Hardware. Zero.

Please knock off the "outrage" and focus on the actual topic.

Thank you.

 

[YSCCC]

I find it odd that no one has mentioned how the President of Puget systems is also on the Intel Board of Advisors?

This seems like pertinent information.

I do believe that's less conflict of interest in this front, but likely they will be biased since:

1) they sell and promote mostly intel
2) when he's board of advisors, he should have more access to internal recommendations of what to do and not to do during tuning, where consumers and other system builders won't ever know except some mysterious "tech god"

Either way, it will be interesting to see how this evolve, if it goes bad, time for me to find reason to upgrade and spending another $2000, though this time around intel won't get my money

 

[stuff and nonesense]

I do believe that's less conflict of interest in this front, but likely they will be biased since:

1) they sell and promote mostly intel
2) when he's board of advisors, he should have more access to internal recommendations of what to do and not to do during tuning, where consumers and other system builders won't ever know except some mysterious "tech god"

Either way, it will be interesting to see how this evolve, if it goes bad, time for me to find reason to upgrade and spending another $2000, though this time around intel won't get my money

Bias is where you find it, or go looking for it.
People look for confirmation bias, “I bought the right thing” or “I dodged a bullet there”.
Contradictory reviews can appear biased, those that run against the consensus. An example I feel this true for is “Tech Notice” on YouTube. In my opinion he skews towards Intel/nvidia. That might be true, it may also indicate my own biases however conscious or unconscious. It might also represent his use case…

Be rigorous, do your own research, aggregate the results, throw away the outliers. Don’t take anything on face value - whether any individual has an agenda or otherwise believe they do until evidence corroborates their findings. Think critically.

 

[YSCCC]

Bias is where you find it, or go looking for it.
People look for confirmation bias, “I bought the right thing” or “I dodged a bullet there”.
Contradictory reviews can appear biased, those that run against the consensus. An example I feel this true for is “Tech Notice” on YouTube. In my opinion he skews towards Intel/nvidia. That might be true, it may also indicate my own biases however conscious or unconscious. It might also represent his use case…

Be rigorous, do your own research, aggregate the results, throw away the outliers. Don’t take anything on face value - whether any individual has an agenda or otherwise believe they do until evidence corroborates their findings. Think critically.

Actually, to me, bias is in everyone's head, you can't be a human and without bias, those who claim they don't, is a liar or dillusional at best. And TBH, this who crisis for Intel IMO is the worst to find the root cause and since the consequence is so big to Intel, they likely don't want to admit unless something even worse will happen to them, they are the only likely personales to find out what is the real issue, not us, not GN, not Wendell, not Puget and not some holy tech savy, it can range from they just push the process too far to there is a hidden in a billion transistor error which slipped through QC screening and will be eventually triggered by something.

I said for the last post is that I don't think Puget is stupid enough to be a blind defender for Intel coz he's in the board of advisory, but rather, one have to say something as a big builder, and especialaly his failure rate is kind of ok at this point of time, he got to push it out especially given how he have lower the profiles compared to the default.

The other "bias" I say in point 2 was ment to say knowing his background, he likely get a lot more info on what the intel engineers thinks are most important for those workstations optimized for certain workload, which, will surely grant him much more stable platform than anyone else.

I won't say he is trying to save intel's ass or whatever, and still interested to see after say, 3 more months, how the nos. per month failure at field for the Ryzen vs RPL

 

[Mattzun]

Puget’s failure rate are probably closer to a mass market failure rates ( Dell or HP ) than the failure rate on home builds
Dell hasn’t jacked up the cost of the two and three year warranties
That means that either the failures don’t have a major impact on them or intel is reimbursing them
This is likely due to lower power limits, no XMP etc

The failure data from game servers is equally unrepresentative of a typical homebuild running games. They are running loads that trigger the voltage spike 24/7.

None of the data we have is representative of the failure rates for home builds using motherboards with good power delivery and aggressive out of the box power settings for gaming.

Only intel has the data on how many people with Boxed cpus requested RMAs.
That might come out during the lawsuits, but it might not .

 

[Gururu]

Puget’s failure rate are probably closer to a mass market failure rates ( Dell or HP ) than the failure rate on home builds
Dell hasn’t jacked up the cost of the two and three year warranties
That means that either the failures don’t have a major impact on them or intel is reimbursing them
This is likely due to lower power limits, no XMP etc

The failure data from game servers is equally unrepresentative of a typical homebuild running games. They are running loads that trigger the voltage spike 24/7.

None of the data we have is representative of the failure rates for home builds using motherboards with good power delivery and aggressive out of the box power settings for gaming.

Only intel has the data on how many people with Boxed cpus requested RMAs.
That might come out during the lawsuits, but it might not .

Home build rates would never amount to any meaningful data. The number of variables involved is too great. Even if my neighbor had an identical setup and the same games, the fact that our own hands might have spread the thermal paste on the CPU is hard to ignore. The only data that is trustworthy is that generated from the mass market from manufacturers with sound QC and QA processes for each SKU.

 

[bit_user]

I do believe that's less conflict of interest in this front, but likely they will be biased since:

1) they sell and promote mostly intel
2) when he's board of advisors, he should have more access to internal recommendations of what to do and not to do during tuning, where consumers and other system builders won't ever know except some mysterious "tech god"

3) A sort of Stockholm syndrome -type effect. Include customers in some way, to make them feel like "part of the team" and make them feel heard. Suddenly, they go from criticizing you externally to redirecting that criticism via the Board of Advisors. Plus, you give them an inside scoop on all the "exciting new stuff" Intel is doing, and now they're probably one of the best brand ambassadors.

That's a hypothetical, to be clear, but I'm sure it happens. It might even be one of the reasons they haves something like a board of advisors.

 

[Guardians Bane]

If Puget's statistics are true, then where's the AMD customer posts and outrage then.

Since 1995, I've never had an AMD or Intel CPU fail on me. Had a few GPUs, hard drives and PSUs break though.

They buy in bulk and the CPUs are tray CPUs. Nothing different about them. But when oems and system builder buy in bulk there more apt to get bad CPUs.

What some people aren't taking into consideration is the actual low percentage of said CPUs. Those numbers are just par for the course from both Intel and AMD.

The Intel fanboys only see the graphs and say,see!!! AMD is worse! But what they ignore is the part where Puget Systems admitta that they're preparing for degrading and faulty CPUs to start incoming. Puget even states:

"The most concerning part of all of this to us here at Puget Systems is the rise in the number of failures in the field, which we haven't seen this high since 11th Gen. We're seeing ALL of these failures happen after 6 months, which means we do expect elevated failure rates to continue for the foreseeable future and possibly even after Intel issues the microcode patch.

Based on this information, we are definitely experiencing CPU failures higher than our historical average, especially"

So by Pugets own admission, Intel failure are on the rise. But Intel die hards just want to stick their heads in the sand and say, nuh uh!

 

[YSCCC]

They buy in bulk and the CPUs are tray CPUs. Nothing different about them. But when oems and system builder buy in bulk there more apt to get bad CPUs.

What some people aren't taking into consideration is the actual low percentage of said CPUs. Those numbers are just par for the course from both Intel and AMD.

The Intel fanboys only see the graphs and say,see!!! AMD is worse! But what they ignore is the part where Puget Systems admitta that they're preparing for degrading and faulty CPUs to start incoming. Puget even states:

"The most concerning part of all of this to us here at Puget Systems is the rise in the number of failures in the field, which we haven't seen this high since 11th Gen. We're seeing ALL of these failures happen after 6 months, which means we do expect elevated failure rates to continue for the foreseeable future and possibly even after Intel issues the microcode patch.

Based on this information, we are definitely experiencing CPU failures higher than our historical average, especially"

So by Pugets own admission, Intel failure are on the rise. But Intel die hards just want to stick their heads in the sand and say, nuh uh!

That's also what I read from them, and even if the microcode really fixed the issue and no more further degradation, those which are already degraded partially will likely runs into issues much faster, so IMO they are also finger crossed and hope it won't be a pain in the a__.

I watched a podcast of Wendell last night and one funny point he said was that among his 50% failure rate, it was a vigorous week long stress test, some, he would say probably half of those failed 50%, only sees a few or even a single error popped up, not acceptable for his server uses, but if used in a desktop case, likely one won't even notice it, as say, one game crash to desktop in a week, most users probably think it's just some unlucky random glitch, only on server side and they read through the logs might find it's a certain core having issues.

 

[rapidwolve]

The numbers are there for all to see, they publish reviews with their settings.

Yes, the 13900k scores around 30-31k (10 minute test)with the settings they ship them with, but on the same note, the 7950x scores 33k. The difference is 10% in performance, but the zen 4 parts have twice the failure rate.

Also their workstations aren't really targeted at rendering workloads so these numbers are kinda useless. They are targeted at photoshop / premiere / lightroom etc, tasks that the 13900k (with the aforementioned settings) is faster than the 7950x anyways.

Here is their review

13th Gen Intel Core Processors Content Creation Review

Following AMDs recent release of their Ryzen 7000 Series desktop processors, Intel is fighting back with their own launch of the Core 13th Gen processors (code named "Raptor Lake"). Unlike AMD's launch which moved to a new socket and added support for DDR5, the 13th Gen CPUs are a drop-in...

www.pugetsystems.com

If they had twice the failure rate, where is all the screaming from elsewhere that AMD chips are failing at the same rate, besides Puget's numbers? Don't just go by their numbers.
See Puget has, and seems always will be, an Intel lover. Yes they will grumble and build AMD based units, but they push Intel. So naturally..defend Intel, make AMD big bad monster.

Besides that, why is Puget defending Intel and their now exposed shadiness? Intel knew it had started having issues back at end of 2022 and they weren't in the hundreds. Nada was mentioned by either Intel or Puget Systems, and those defective chips were not recalled. Then they also had reports in mid-late 2023 of the voltage issues, yet again, no microcode was re-written back then. And to add insult to injury, they flat out refused the vast unknown number of RMA's worldwide.

 

[thestryker]

If they had twice the failure rate, where is all the screaming from elsewhere that AMD chips are failing at the same rate, besides Puget's numbers? Don't just go by their numbers.
See Puget has, and seems always will be, an Intel lover. Yes they will grumble and build AMD based units, but they push Intel. So naturally..defend Intel, make AMD big bad monster.

Besides that, why is Puget defending Intel and their now exposed shadiness? Intel knew it had started having issues back at end of 2022 and they weren't in the hundreds. Nada was mentioned by either Intel or Puget Systems, and those defective chips were not recalled. Then they also had reports in mid-late 2023 of the voltage issues, yet again, no microcode was re-written back then. And to add insult to injury, they flat out refused the vast unknown number of RMA's worldwide.

You're not doing anyone any favors by playing the "let's make stuff up" game. Puget also isn't defending anyone, but rather giving out their failure information and showing a level of nuance nobody else has.

The failure rates Puget are seeing are quite believable, but you have to pay attention to all of the numbers. The AMD systems have a much lower field failure rate which means once they make it out they're largely good, but have a higher shop failure rate which means they have a higher failure rate from the factory. The other very important part is that their failures rates are very low period (it looks like they have a 4% failure rate from the factory for AMD and the field failure rate seems really small maybe around 0.2%).

As a regular consumer who buys singular parts that means the chances of you getting something bad are still small. For a business if there has to be a failure having it happen before getting to a customer is absolutely when you want it to occur. For individual buyers this is more annoying, but I'd say it's still definitely better than having some random failure some random period of time down the road.

I don't think saying AMD has a higher failure rate is useful, despite it being technically true, because the nuance is the important part here. I think it's pretty clear that Puget included those numbers just to reassure their customers as to why they were still selling Intel parts.

I see these are the primary takeaways for these numbers:
Custom tweaked power profiles does make a difference in the failure rate. However for all their tweaking 13/14th generation parts are failing at a high rate compared to 12th once deployed and consistently continue to do so. AMD has massively improved their long term quality from Zen 3 to Zen 4 (though this could change over time). AMD also seems to have a slightly higher failure rate from the factory.

 

[TheHerald]

If they had twice the failure rate, where is all the screaming from elsewhere that AMD chips are failing at the same rate, besides Puget's numbers? Don't just go by their numbers.
See Puget has, and seems always will be, an Intel lover. Yes they will grumble and build AMD based units, but they push Intel. So naturally..defend Intel, make AMD big bad monster.

Let me turn your argument against you.

If intel had higher failure rate then why doesn't it show on pugets numbers? The other people that are screaming about Intel are and seems always will be amd lovers. They push amd. So naturally they defend amd and make intel seem like the big bad monster

I sound absolutely ridiculous don't I? That's what I thought, hope you got the point.

 

[TheHerald]

You're not doing anyone any favors by playing the "let's make stuff up" game. Puget also isn't defending anyone, but rather giving out their failure information and showing a level of nuance nobody else has.

The failure rates Puget are seeing are quite believable, but you have to pay attention to all of the numbers. The AMD systems have a much lower field failure rate which means once they make it out they're largely good, but have a higher shop failure rate which means they have a higher failure rate from the factory. The other very important part is that their failures rates are very low period (it looks like they have a 4% failure rate from the factory for AMD and the field failure rate seems really small maybe around 0.2%).

As a regular consumer who buys singular parts that means the chances of you getting something bad are still small. For a business if there has to be a failure having it happen before getting to a customer is absolutely when you want it to occur. For individual buyers this is more annoying, but I'd say it's still definitely better than having some random failure some random period of time down the road.

I don't think saying AMD has a higher failure rate is useful, despite it being technically true, because the nuance is the important part here. I think it's pretty clear that Puget included those numbers just to reassure their customers as to why they were still selling Intel parts.

I see these are the primary takeaways for these numbers:
Custom tweaked power profiles does make a difference in the failure rate. However for all their tweaking 13/14th generation parts are failing at a high rate compared to 12th once deployed and consistently continue to do so. AMD has massively improved their long term quality from Zen 3 to Zen 4 (though this could change over time). AMD also seems to have a slightly higher failure rate from the factory.

Only thing I disagree with is, zen 4 don't ship doa. They are not literally dead. They are just in such bad shape that they bsod or crash before they even make it out of the door. Zen 3 to zen 4 was a huge quality decrease the way I see it. Going from cpus that will degrade within a couple of years to cpus that will degrade and died within a couple of hours, wouldn't call that a good showing.

 

[stuff and nonesense]

Let me turn your argument against you.

If intel had higher failure rate then why doesn't it show on pugets numbers? The other people that are screaming about Intel are and seems always will be amd lovers. They push amd. So naturally they defend amd and make intel seem like the big bad monster

I sound absolutely ridiculous don't I? That's what I thought, hope you got the point.

There are lies, damn lies and statistics.
We can all cherrypick whatever supports our arguments.

You would need to aggregate Puget’s numbers in with Dell, HP, Lenovo etc. and the minority of PCs that are self built enthusiast builds. Then look at the bell curve.

 

[thestryker]

Only thing I disagree with is, zen 4 don't ship doa. They are not literally dead. They are just in such bad shape that they bsod or crash before they even make it out of the door. Zen 3 to zen 4 was a huge quality decrease the way I see it. Going from cpus that will degrade within a couple of years to cpus that will degrade and died within a couple of hours, wouldn't call that a good showing.

Shop versus field is just shipping we don't know what period of time they're failing over. Those Zen 3 parts could be failing a month out or years out we do not know. Now maybe over time Zen 4 will look similar to Zen 3 as manufacturing gets better and failures over time occur, but for a SI shop failures are significantly better than field even if they're higher.