Question DEGRADING Raptor lake CPUs

[Kocicak]

I noticed some reports about degrading i9 13900K and KF processors.

I experienced this problem myself, when I ran it at 6 GHz, light load (3 threads of Cinebench), at acceptable temperature and non extreme voltage. After only few minutes it crashed, and then it could not run even at stock setting without bumping the voltage a bit.

I was thinking about the cause for this and I believe the problem is, that people do not appreciate, how high these frequencies are and that the real comfortable frequency limit of these CPUs is probably at something like 5500 or 5600 MHz. These CPUs are made on a same process (possibly improved somehow) on which Alder lake CPUs were made. See the frequencies 12900KS runs at. The frequency improvement of the new process tweak may not be so high as some people presume.

Those 13900K CPUs are probably highly binned to be able to find those which contain some cores which can reliably run at 5800 MHz. Some of the 13900K probably have little/no OC reserve left and pushing them will cause them to degrade/break.

The conclusion for me is that the best you can do to your 13900K or 13900KF is to disable the 5800 MHz peak, which will allow you to offset the voltage lower, and then set all core maximal frequency to some comfortable level, I guess the maximum level could be 5600 MHz. With lowered voltage this frequency should be gentler to the processor than running it at original 5500 MHz at higher voltage. You can also run it at lower frequencies, allowing for even higher voltage drop, but then the CPU is slowly loosing its sense (unless you want some high efficiency CPU intended for heavy multithread loads).

Running it with some power consumption limit dependent on your cooling solution to keep the CPU at sensible temperature will help too for sure.

 

[IEC]

I would love to see the temps and if any degradation running the CPUs on PG or some other demanding app where the cores are at 100% load 24/7 for days.

It's the lightly loaded scenarios (1T, 2T) where they showed faster degradation. It's why 14900K-based Minecraft servers degraded at an abnormally high rate.

 

[Markfw]

It's the lightly loaded scenarios (1T, 2T) where they showed faster degradation. It's why 14900K-based Minecraft servers degraded at an abnormally high rate.

F@H only takes 1 core or less. If you do BOINC, you would set CPU numbers to 40% max, that would be light. One core is virtually idle.

 

[DrMrLordX]

I would love to see the temps and if any degradation running the CPUs on PG or some other demanding app where the cores are at 100% load 24/7 for days.

There's more risk to degradation from constant boost/idle behavior.

edit: @IEC beat me to it.

 

[Kocicak]

I upgraded my Z690 ACE system from a 12600K to a 14700K. Theoretically it should be safe from degradation due to being on the latest BIOS as of today's date prior to even installing the CPU.

The BIOS probably saves the CPU only from extreme voltage spikes and from very quick degradation. Intel never communicated any expected CPU life times before and after the final BIOS "fix".

It has a 280mm AIO with a LGA1700 contact frame which appears to be barely sufficient to keep it from thermal throttling under full load.

253W is a lot of power for such a small CPU, are you sure you do not want to run it with lower power limit? High temperature accelerates degradation.

What AIO is that? I encountered a few Arctic AIOs with very strongly tightened cooling plate screws, which led to strong bowing of the plate. The CPU mounted with contact frame tends to bulge out. So you have a sphere contacting a cylinder, which can result in very poor contact between the IHS and cooler base. The cooler base should be as flat as possible for Intel CPUs mounted with the frame.

The rig will be running 24/7 running Folding@Home on the GPU which is a lightly threaded workload for CPU purposes (usually) so I should be hitting max 1T boost with at least some regularity. I will monitor for any degradation or instability.

You will be hitting maximal frequency and pretty high voltages, what is your goal? Do you want to prevent the CPU from degrading (for that you should limit the frequency) or are you running an experiment and you want to see, how long it will take to degrade it?

 

[igor_kavinski]

253W is a lot of power for such a small CPU, are you sure you do not want to run it with lower power limit? High temperature accelerates degradation.

I have pushed my 12700K to 300W with high clocks using XTU. I have yet to experience instability.

 

[Kocicak]

I have pushed my 12700K to 300W with high clocks using XTU. I have yet to experience instability.

I would not publicly admit such crimes against silicon if I were you...

 

[Wolverine2349]

I have pushed my 12700K to 300W with high clocks using XTU. I have yet to experience instability.

Thats 12th Gen. 12th Gen is reliable 13th and 14th Gen even with microcode update are fragile.

The microcode update is a band aide I fear and not a long term solution for Raptor Lake.

But your 12700K is fine. Different stepping/die.

 

[igor_kavinski]

Thats 12th Gen. 12th Gen is reliable 13th and 14th Gen even with microcode update are fragile.

Yeah. That's the problem I (and many others) have when people defend Raptor Lake. Alder Lake - no issues. Raptor Lake - issues. Clearly something went wrong. It's like saying please don't drive your latest Ferrari above the legal speed limit or it will catch fire. People have been hitting 5200+ MHz all core with Zen 4 and Zen 5 with no issues. No heat buildup causing those CPUs to degrade. Because the engineers did their homework.

 

[Kocicak]

People have been hitting 5200+ MHz all core with Zen 4 and Zen 5 with no issues. No heat buildup causing those CPUs to degrade. Because the engineers did their homework.

Those are made on two completely different processes of two different companies.

The Intel process itself may be perfectly fine, but it has its limits, as everything. 13th and 14 gen CPUs themselves may be perfectly fine, but they can have some peculiar hotspot, that strains silicon a little bit more than usual. The combination of the process, the CPU design and frequency and power limits Intel chose leads to failure. Another combinations may be perfectly fine.

E.g. 14900K with 3.6 GHz and 100W limits may last long years running heavy workloads 24/7.

5GHz and 160W limits may give you few years with reasonable consumer workloads.

 

[igor_kavinski]

E.g. 14900K with 3.6 GHz and 100W limits may last long years running heavy workloads 24/7.

5GHz and 160W limits may give you few years with reasonable consumer workloads.

Useless for most power users at these speeds.

But yeah, you make a good case. Let's take a bank loan, buy up ALL inventory of RPL and its Refresh and start on a worldwide project to deploy these CPUs at your recommended speeds in the education and business sectors. Elon won't know what hit him!

 

[Wolverine2349]

Those are made on two completely different processes of two different companies.

The Intel process itself may be perfectly fine, but it has its limits, as everything. 13th and 14 gen CPUs themselves may be perfectly fine, but they can have some peculiar hotspot, that strains silicon a little bit more than usual. The combination of the process, the CPU design and frequency and power limits Intel chose leads to failure. Another combinations may be perfectly fine.

E.g. 14900K with 3.6 GHz and 100W limits may last long years running heavy workloads 24/7.

5GHz and 160W limits may give you few years with reasonable consumer workloads.

It may be perfectly fine but it was a rushed design in case Meteor Lake was not ready. So I am not so sure.

I would not trust a 13t or 14th Gen in my NAS with TrueNAS. But I am running a 12th Gen 2 core Celeron that has the 6 + 0 die with low power and peak power like 25 watts or less underclocked and fanless and super stable for over 2 months.

Though Raptor Lake 8 + 16 die had no such low end parts.

There was only 2 Alder Lake dies 8 +8 and 6 + 0. And one Raptor Lake die which is 8 + 16.

Though intel lies and rebrands Alder Lake dies as 13th and 14th Gen Raptor Lake even though the IPC of those P cores are 5-6% lower as they in reality are Alder Lake just with 13XXX and 14XXX numbers for the lower end non K SKUs in m many cases. Cause Raptor Lake has only one such die which is 8 + 16.

 

[Kocicak]

Useless for most power users at these speeds.

Not sure it is useless. 5 GHz (which could be fine for normal consumer workloads) is only about 10% slower than 5600 or 5700 MHz that 14900K uses. You can also get a lot of performance with 160W.

And people running 24/7 wokloads probably also care about energy efficiency, which will be very good at lower frequencies. And such people tend to use different CPUs anyway - dedicated for workstations or servers.

Let's take a bank loan, buy up ALL inventory of RPL and its Refresh and start on a worldwide project to deploy these CPUs at your recommended speeds in the education and business sectors. Elon won't know what hit him!

I am confused by this (and also what has the one who should not be named to do with this?), but forcing the CPUs to be used in a way that will not destroy them and in an energy efficient manner would be of course good.

 

[Kocicak]

It may be perfectly fine but it was a rushed design in case Meteor Lake was not ready. So I am not so sure.

I would not trust a 13t or 14th Gen in my NAS ...

I am not aware of any error in the design, the CPUs run just fine. You just do not run 5 GHz CPUs at 6 GHz.
That is all that is to it.

It does not matter anymore what caused the CPUs to be able to run this "slow", because they are not to be respinned or optimised again. The same about the production process, nobody will invest any significant efforts in improving it.

The CPUs are what they are, and if you respect it, they will serve you well.

 

[Dave3000]

I am not aware of any error in the design, the CPUs run just fine. You just do not run 5 GHz CPUs at 6 GHz.
That is all that is to it.

It does not matter anymore what caused the CPUs to be able to run this "slow", because they are not to be respinned or optimised again. The same about the production process, nobody will invest any significant efforts in improving it.

The CPUs are what they are, and if you respect it, they will serve you well.

Does this CPU degradation of 13th and 14th gen still happen when setting to the stock Intel power limits in the BIOS, if not set as a BIOS default, or keeping it at the Intel stock power limits if it's already set to the Intel stock power limits by default? Should a CPU not degrade if disabling the power limits in the BIOS as long as you have good enough cooling for the CPU and if it degrades because of this, is it considered a defect in the CPU?

 

[Hulk]

Does this CPU degradation of 13th and 14th gen still happen when setting to the stock Intel power limits in the BIOS, if not set as a BIOS default, or keeping it at the Intel stock power limits if it's already set to the Intel stock power limits by default? Should a CPU not degrade if disabling the power limits in the BIOS as long as you have good enough cooling for the CPU and if it degrades because of this, is it considered a defect in the CPU?

There are too many factors involved to answer this with a yes or no. How good is the silicon of the particular CPU? How good is the cooling? What is the use case for the CPU? Good silicon, with good cooling, and someone not running 1 or 2 cores at 6GHz around the clock will probably be okay.

In my opinion having had 5 of these Raptor CPU's, what kills them is the high voltage required for the very high 1 or 2 core boost frequencies. For example, 5.5GHz under load generally requires under 1.2 volts, perhaps 1.3 or a bit more while idling or under light load. But 6GHz requires nearly 1.5V for many of these parts.

So now consider the case where you are lightly loading 1 or 2 cores. Now you are jamming 1.5V into all 8 cores for really no reason since the increased performance is negligible. Heat won't be a problem since the core loading might not be all that high but that is a lot of high energy electrons running through those structures in the CPU.

Okay here's a real world example. One of my apps that loves frequency but doesn't really saturate the cores is Presonus Studio One. When rendering out a project with my 14900K all 8 cores will hit 5.7GHz (stock). Performance was nice, temps were good, but voltage was for me uncomfortably high. I don't like seeing anything over 1.3V, definitely not over 1.4V.

As has been written many times here, Intel over rated Raptor Lake and now they are paying the price. People will say Alder Lake doesn't have the problem. Of course it doesn't, it doesn't run at the high frequencies Raptor does. Run Raptor at 5.2GHz max and I bet you won't have issues. Intel just had to put "6.0GHz" on the box for Raptor Lake.

In hindsight Intel should have had much better binning on the higher end Raptor Lake parts and the ability to control Vcore per core. If they has 2 cores on a die that could run at 6GHz at 1.3V, then sure let 'em rip. But more often than not they just pushed enough voltage to make the spec, temporarily at least. I've seen a lot of data on these parts over at Overclockers.net over the last few years and only about the top perhaps 10% could reliably make the spec on the box in my opinion. Perhaps the 14900KS parts could reliably make the 14900K spec.

It's a PITA situation for Intel. Most of the chips won't outright fail, but they will start to need more volts to hit frequencies or have to be clocked down as the silicon degrades.

But again it's a situation of Intel's own making so they will have to suffer through it. Or it will be the one of the cuts that eventually brings them down.

 

[Markfw]

There are too many factors involved to answer this with a yes or no. How good is the silicon of the particular CPU? How good is the cooling? What is the use case for the CPU? Good silicon, with good cooling, and someone not running 1 or 2 cores at 6GHz around the clock will probably be okay.

In my opinion having had 5 of these Raptor CPU's, what kills them is the high voltage required for the very high 1 or 2 core boost frequencies. For example, 5.5GHz under load generally requires under 1.2 volts, perhaps 1.3 or a bit more while idling or under light load. But 6GHz requires nearly 1.5V for many of these parts.

So now consider the case where you are lightly loading 1 or 2 cores. Now you are jamming 1.5V into all 8 cores for really no reason since the increased performance is negligible. Heat won't be a problem since the core loading might not be all that high but that is a lot of high energy electrons running through those structures in the CPU.

Okay here's a real world example. One of my apps that loves frequency but doesn't really saturate the cores is Presonus Studio One. When rendering out a project with my 14900K all 8 cores will hit 5.7GHz (stock). Performance was nice, temps were good, but voltage was for me uncomfortably high. I don't like seeing anything over 1.3V, definitely not over 1.4V.

As has been written many times here, Intel over rated Raptor Lake and now they are paying the price. People will say Alder Lake doesn't have the problem. Of course it doesn't, it doesn't run at the high frequencies Raptor does. Run Raptor at 5.2GHz max and I bet you won't have issues. Intel just had to put "6.0GHz" on the box for Raptor Lake.

In hindsight Intel should have had much better binning on the higher end Raptor Lake parts and the ability to control Vcore per core. If they has 2 cores on a die that could run at 6GHz at 1.3V, then sure let 'em rip. But more often than not they just pushed enough voltage to make the spec, temporarily at least. I've seen a lot of data on these parts over at Overclockers.net over the last few years and only about the top perhaps 10% could reliably make the spec on the box in my opinion. Perhaps the 14900KS parts could reliably make the 14900K spec.

It's a PITA situation for Intel. Most of the chips won't outright fail, but they will start to need more volts to hit frequencies or have to be clocked down as the silicon degrades.

But again it's a situation of Intel's own making so they will have to suffer through it. Or it will be the one of the cuts that eventually brings them down.

My take. You should be able to buy a CPU, use it as hard as you want 24/7 @100% and not be worrying about BIOS updates, settings or anything else. Even though it is easy to not get proper cooling for aftermarket systems, you should be be able to get a reasonable size cooler and just not worry. With 13th and 14 Gen Intel this is not possible, period. Their competition does not have this problem.

 

[Hulk]

My take. You should be able to buy a CPU, use it as hard as you want 24/7 @100% and not be worrying about BIOS updates, settings or anything else. Even though it is easy to not get proper cooling for aftermarket systems, you should be be able to get a reasonable size cooler and just not worry. With 13th and 14 Gen Intel this is not possible, period. Their competition does not have this problem.

Yes. Zen 5 is better and you are always there to make sure that point is driven home. But he was asking specifically about Raptor Lake degradation.

 

[Markfw]

Yes. Zen 5 is better and you are always there to make sure that point is driven home. But he was asking specifically about Raptor Lake degradation.

But I keep hearing the same rationalizations, adjust the bios, lower vcore, run it slower. You should not have to do anything like that to a CPU to do what you paid for. My point gets ignored, and thats why I reply.

 

[Hulk]

But I keep hearing the same rationalizations, adjust the bios, lower vcore, run it slower. You should not have to do anything like that to a CPU to do what you paid for. My point gets ignored, and thats why I reply.

My comment was in jest. As you know I agree with you and you are correct!

 

[Markfw]

My comment was in jest. As you know I agree with you and you are correct!

Sorry, I don't always pick up that good on sarcasm. But my comment does still stand. I am sick of rationalizations. We all know the chips are
1) designed badly.
2) Set up to make the high frequency's to win benchmarks
3) possible manufacturing problems.
4) and due to these degrade badly, and require bios changes and updates that many users are not even qualified to do, or don't know its required.

 

[Kocicak]

In my opinion having had 5 of these Raptor CPU's, what kills them is the high voltage required for the very high 1 or 2 core boost frequencies. For example, 5.5GHz under load generally requires under 1.2 volts, perhaps 1.3 or a bit more while idling or under light load. But 6GHz requires nearly 1.5V for many of these parts.

...

In hindsight Intel should have had much better binning on the higher end Raptor Lake parts and the ability to control Vcore per core.

The CPUs will still run pretty hot and inefficiently at 5.5 GHz.

The CPUs with 5.5 GHz + 253W and 5.0 Ghz + 160W limits seem like COMPLETELY DIFFERENT PRODUCTS.

I am not sure that tests of the silicon that last a very short time during binning in manufacture can reliably reveal how will the CPU endure the long term stress.

A CPU with very conservative settings and wide safety margins will be 100% reliable, but will have most of the time large unused performance potential.

CPUs with aggresive settings will fail sometimes, consumers should be aware what are they buying and understand the risk, but the failure rate should still be manageably low (in low single digit %).

A third disastrous scenario of cocky execs selling the products with settings beyond what the products can handle and failure rates in tens of percent should never happen and I believe that it is one of the main reasons Pat Gelsinger was forced to retire. Intel simply cannot financially afford to handle this major screw-up properly and admit that the products need to run slower and have lower performance. They need to pretend that everything is OK now with the new "final fix microcode" and silently exchange the failed CPUs while pretending that nothing bad is happening anymore.

14900K with 5.0 Ghz + 160W limits could be the conservative variant, with even lower limits recommended for 24/7 high stress operation

14900K with 5.5 GHz + 253W limits could be the aggresive variant.

Anything above that should have never happened.