Intel processors crashing Unreal engine games (and others)

[Saylick]

Quite the dilemma they’re in. I was thinking that last point, maybe discount an upgrade to ARL, but then customers would have to eat the cost of a new motherboard, unless they were to do some like half-off an upgrade for both CPU and mobo.

I bet they’re holding out on hope that the microcode update will be the fix they need so they can avoid these thorny issues.

I’m against the idea of a voucher or discount to upgrade precisely because of your point: to “solve” the problem the consumer would have to shell out more dineros. Why does the consumer have to pay additional money to fix a problem that was not their fault? It makes no sense unless the upgrade is for free, but the bean counters won’t accept that, so I reckon a no hassles RMA program is the true solution.

 

[jdubs03]

I’m against the idea of a voucher or discount to upgrade precisely because of your point: to “solve” the problem the consumer would have to shell out more dineros. Why does the consumer have to pay additional money to fix a problem that was not their fault? It makes no sense unless the upgrade is for free, but the bean counters won’t accept that, so I reckon a no hassles RMA program is the true solution.

I agree. That’s clearly the best option. Just get it done, take the hit and move on to ARL which needs to be solid so they can start building up their brand rep.

 

[branch_suggestion]

it’s times like this I wish x86 was open to more companies.

Always an option if alternatives get enough traction.

 

[marcUK2]

We don't know the cause of the apparent degradation, so we can't establish a "safe" profile. However, there are things you can do to make sure the CPU is operating within the limits recommended by Intel. Start by updating your BIOS and then check all the settings below to make sure motherboard is configured properly.

You can use the values in the "Performance" column for a 250W configuration, but you need to make sure the current limits are also enforced (ICCMAX). Also make sure that CEP, eTVB, TVB and TVB Voltage Optimizations are enabled. The last few settings are essential for making sure the CPU uses less voltage under light loads.

View attachment 103085

After some testing on the 14900 non-K, I set the ICCMAX to 275A, limited all my cores to 5300MHz, (Asus Multicore Enhancement Unlimited - 90C)
Running CB23 and HWMonitor, This limits me to 225W permanently and I can score 35000 in CB, which is about 3000 down from the best I can get.
These settings also limit the VCORE to 1.3V which I assume is safe for a 14900.

If I enforce limits or use Baseline, I score 22000 in CB, which is worse than both the 3950x and 12700 non-K I upgraded from. Should have kept it! Zen5 here I come. Intel Crap

 

[KompuKare]

After some testing on the 14900 non-K, I set the ICCMAX to 275A, limited all my cores to 5300MHz, (Asus Multicore Enhancement Unlimited - 90C)
Running CB23 and HWMonitor, This limits me to 225W permanently and I can score 35000 in CB, which is about 3000 down from the best I can get.
These settings also limit the VCORE to 1.3V which I assume is safe for a 14900.

If I enforce limits or use Baseline, I score 22000 in CB, which is worse than both the 3950x and 12700 non-K I upgraded from. Should have kept it! Zen5 here I come. Intel Crap

That no major review site has attempted to benchmark the new baseline is ridiculous.

I know baseline is a moving target due Intel being evasive, but so much for independent reviewers!

 

[Nothingness]

That no major review site has attempted to benchmark the new baseline is ridiculous.

I know baseline is a moving target due Intel being evasive, but so much for independent reviewers!

As @Philste wrote we can expect Computerbase to post new results. I hope they'll do that when publishing their Zen5 review.

I also expect Phoronix to do something, as they did in the past with new microcode updates.

 

[coercitiv]

I know baseline is a moving target due Intel being evasive, but so much for independent reviewers!

AFAIK Intel themselves have said K CPUs should use the Performance profile.


For the 13900K/14900K they recommend the Extreme profile:

 

[KompuKare]

As @Philste wrote we can expect Computerbase to post new results. I hope they'll do that when publishing their Zen5 review.

I also expect Phoronix to do something, as they did in the past with new microcode updates.

Don't mention ComputerBase.

Intel Baseline Profile: BIOS-Updates gegen instabile K-CPUs

Intel hält sich weiterhin zu instabilen K-CPUs bedeckt. Derweil rollen Mainboard-Hersteller ein „Baseline Profile“ per BIOS-Update aus.

www.computerbase.de

"Benchmarks folgen" they wrote back on the 26th of April. Months later, nada.

 

[DAPUNISHER]

I don't think I've ever proposed or endorsed conspiracy theories on the forum, but all the behavioral signs from Intel are telling me they're in cover up mode, more exactly in delay and damage control mode. Minimize RMAs. The official press release was meh and I was ready to let it go, but the reddit post and the 3rd party video endorsed by Intel stink from a mile away.

I saved you a seat on the couch

They released the first MC patch to fix what has been comically dubbed on another forum - terminal velocity boost on the top SKUs.

Now Intel presents us with MC patch 2 electric boogaloo. But before we can have the dance off to save the raptorverse, they have to put on their ba-ba-boogie shoes. And that'll take weeks. Which they conclude with, if in the meantime, your CPU stops being a clever girl, hit us up and we'll do you a solid.

This whole thing is a joke, no point in hitting the brakes now.

EDIT: Aussie Steve just concluded his DDR5 speed testing by saying HUB is not recommending 13&14 until at least after the patch next month. I don't know how it is possible to verify it solves the problem in that kind of timeframe. Realistically it'll take months to evaluate. I should hope he won't just bench it to confirm it doesn't nerf performance then give it a thumbs up.

 

[igor_kavinski]

EDIT: Aussie Steve just concluded his DDR5 speed testing by saying HUB is not recommending 13&14 until at least after the patch next month. I don't know how it is possible to verify it solves the problem in that kind of timeframe. Realistically it'll take months to evaluate. I should hope he won't just bench it to confirm it doesn't nerf performance then give it a thumbs up.

If any reviewer includes the 13900K(S)/14900K(S) benchmarks in Zen 5 review without an asterisk pointing out that there is no final Intel solution yet for stability issues, I would automatically consider them disingenuous.

If I were a reviewer for Zen 5, I would put a huge disclaimer at the beginning of the benchmark results:

Intel 13th/14th gen CPU stability/reliability/longevity not guaranteed by Intel as of the time of this publication.

 

[csbin]

4th attempt

https://twitter.com/x/status/1815735665201443020

 

[Joe NYC]

The way I see it, it reminds me of the Narcissist’s Prayer:
- Intel CPUs don’t have a problem.
- And if it did, it didn’t affect a lot of people.
- Even if it affected a lot of people, the solution doesn’t reduce performance.
- Even if it does reduce performance, it’s not my fault. It was the mobo maker’s fault.
- And if it was my fault, we didn’t mean it.
- And if we meant it, you the consumer deserved it because all you care about is who has the longest bar or biggest number in the benchmark.

I’m on Xitter and even I start to see the usual Intel defenders go down this Prayer, one step at a time. First it was denial, then it was minimization, etc etc. At some point, I wouldn’t be surprised to see them say RPL is in the past and we shouldn’t care anymore, so let’s just focus on ARL, ya? Please? Pretty please??

All this negativity.

Can you, just for a second, imagine what can be, unburdened by what has been?

 

[TheELF]

4th attempt

https://twitter.com/x/status/1815735665201443020

This is the first attempt!
"Intel default settings" was never introduced as a fix by intel, but rather as a list of settings to look at and if your mobo goes above them you should know that you are running outside of specs, if the media you follow portrayed it as a fix, then take a good long look at that media and maybe reconsider at least the way you consume it.
eTVB bug was never introduced as a fix by intel, look above. They just fund a bug while investigating and issued a bugfix.

High voltage is what almost everybody was saying from the beginning, maybe in different variations as in out of the box overclock and so on.
That's what the reply to the X says as well, ac/dc loadline out of whack.

EDIT: Aussie Steve just concluded his DDR5 speed testing by saying HUB is not recommending 13&14 until at least after the patch next month. I don't know how it is possible to verify it solves the problem in that kind of timeframe. Realistically it'll take months to evaluate. I should hope he won't just bench it to confirm it doesn't nerf performance then give it a thumbs up.

Look at voltages before the MC update during boost,
look at the voltages after the MC update during boost,
if after is below frying the CPU level and before is above
...
profit.

 

[Nothingness]

Don't mention ComputerBase.

Intel Baseline Profile: BIOS-Updates gegen instabile K-CPUs

Intel hält sich weiterhin zu instabilen K-CPUs bedeckt. Derweil rollen Mainboard-Hersteller ein „Baseline Profile“ per BIOS-Update aus.

www.computerbase.de

"Benchmarks folgen" they wrote back on the 26th of April. Months later, nada.

That's disappointing, I thought they were a serious review site

If they include results from Intel in their Zen5 review without redoing them after the various patches that'd be misleading indeed.

 

[Rigg]

All this negativity.

Can you, just for a second, imagine what can be, unburdened by what has been?

Perhaps Intel will come out with a Venn diagram that makes this all easier to understand.

 

[KompuKare]

That's disappointing, I thought they were a serious review site

If they include results from Intel in their Zen5 review without redoing them after the various patches that'd be misleading indeed.

I once posted on their forum on a news item (about older AM4 board not getting Zen3 support), and the impression I got is that they over value their contacts - in the case of my post more of "we will not sour our relationship with motherboard OEMs or AMD by asking awkward questions" vibe.

 

[Joe NYC]

Perhaps Intel will come out with a Venn diagram that makes this all easier to understand.

It would be an interesting Venn diagram with 3 overlapping "solution" circles and a point labeled "root cause" outside of all of the circles.

 

[coercitiv]

Buildzoid 1 hour long rant rant on 14900K in SuperMicro machines running Minecraft servers. I'll start listening now so can't offer any summary, but even the first 3minutes are worth it.

 

[igor_kavinski]

4th attempt

https://twitter.com/x/status/1815735665201443020

No, actually this is very good example that should be shown in schools:

"If at first you don't succeed, be like Intel and keep trying, despite having billions of dollars and thousands of employees because failure can bite ANYONE!".

 

[blckgrffn]

Buildzoid 1 hour long rant rant on 14900K in SuperMicro machines running Minecraft servers. I'll start listening now so can't offer any summary, but even the first 3minutes are worth it.

I actually turned this on and listened to the opening and can confirm. You can tell from the delivery just how impressed he is.

30% failure rate in a Supermicro board in 30-90 days running at Intel baseline with slow ECC DDR5.

The hosting company has discontinued offering these blades now.

 

[igor_kavinski]

AI Summary:

Spoiler: tldrthis summarization

The video discusses the issues with Intel's 14th generation 14900K CPUs, particularly in the context of running Minecraft servers. The key points are:

1. Minecraft servers are very single-threaded and memory-bound, which puts a lot of stress on the 14900K CPUs. This has led to a very high failure rate of these CPUs, with 30% of them failing within 1-2 months when running at 5.7 GHz.

2. To mitigate the issue, the data center running these Minecraft servers has started disabling the Thermal Velocity Boost feature, which has reduced the failure rate to under 5% over longer periods of time (3-4 months).

3. The data center has stopped offering new contracts for 14900K-based servers due to the high maintenance costs associated with frequently replacing failed CPUs.

4. The video also examines the power and current settings on a specific Super Micro server board running the 14900K, which appears to be following Intel's specifications.

1. What is the main issue with running Intel 14900K CPUs in Minecraft servers?

The main issue with running Intel 14900K CPUs in Minecraft servers is the high failure rate, with 30% of the CPUs failing within 1-2 months when running at 5.7 GHz. This rapid degradation is attributed to the single-threaded and memory-bound nature of Minecraft servers, which puts significant stress on the CPUs.

2. How has the data center mitigated the high failure rate of the 14900K CPUs?

The data center has mitigated the high failure rate of the 14900K CPUs by disabling the Thermal Velocity Boost feature on all the CPUs running the Minecraft servers. This has reduced the failure rate to under 5% over longer periods of time (3-4 months).

3. Why has the data center stopped offering new contracts for 14900K-based servers?

Based on the context provided, the data center has stopped offering new contracts for 14900K-based servers for a few key reasons:

The 14900K CPUs were failing at an extremely high rate on the Minecraft servers - the failure rate was around 30% within 1-2 months. 2. The maintenance and replacement costs were becoming very high, as a technician had to physically go into the data center, pull out the blade, and replace the failed CPU every time.

The data center no longer has replacement 14900K CPUs available, as they have stopped offering these servers. The person leasing the 14900K-based servers now doesn't have a way to get replacement CPUs.4. The high failure rate and maintenance costs made the 14900K-based servers an unsustainable offering for the data center. They have decided to stop providing new contracts for these servers.

In summary, the combination of extremely high failure rates, high maintenance costs, and lack of replacement CPUs led the data center to discontinue offering new 14900K-based server contracts.

4. What power and current settings are being used on the Super Micro server board running the 14900K?

Based on the context provided, here are the key power and current settings being used on the Super Micro server board running the Intel 14900K CPU:

Power Limits: - PL1 and PL2 are both set to 253W, which is the Intel specification for the 14900K. - There is no mention of a higher PL4 power limit, which is sometimes seen on gaming motherboards.

Current Limits: - The motherboard has a 200A current limit, which is the maximum current the VRM can handle. - The ICC Max (CPU current limit) is set to the standard range of up to 512A.

Load Line Calibration: - The AC and DC Load Line settings are both set to 1.1 mOhms, which is the Intel specification. - This leads to very high voltages being applied to the CPU.4. Memory Configuration:
The servers are running 128GB of low-profile DDR5-3600 memory, which is the Intel specification for 13th/14th gen CPUs. - They are not running higher memory speeds like DDR5-4400 or DDR5-4000.

So in summary, the Super Micro server board appears to be following the Intel specifications for power, current, and memory settings with the 14900K CPU. The high failure rates seem to be more related to the CPU's inability to handle the sustained workload of the Minecraft servers, rather than any unusual motherboard settings.

 

[blckgrffn]

AI Summary:

Spoiler: tldrthis summarization

The video discusses the issues with Intel's 14th generation 14900K CPUs, particularly in the context of running Minecraft servers. The key points are:1. Minecraft servers are very single-threaded and memory-bound, which puts a lot of stress on the 14900K CPUs. This has led to a very high failure rate of these CPUs, with 30% of them failing within 1-2 months when running at 5.7 GHz.2. To mitigate the issue, the data center running these Minecraft servers has started disabling the Thermal Velocity Boost feature, which has reduced the failure rate to under 5% over longer periods of time (3-4 months).3. The data center has stopped offering new contracts for 14900K-based servers due to the high maintenance costs associated with frequently replacing failed CPUs.4. The video also examines the power and current settings on a specific Super Micro server board running the 14900K, which appears to be following Intel's specifications.

1. What is the main issue with running Intel 14900K CPUs in Minecraft servers?

The main issue with running Intel 14900K CPUs in Minecraft servers is the high failure rate, with 30% of the CPUs failing within 1-2 months when running at 5.7 GHz. This rapid degradation is attributed to the single-threaded and memory-bound nature of Minecraft servers, which puts significant stress on the CPUs.

2. How has the data center mitigated the high failure rate of the 14900K CPUs?

The data center has mitigated the high failure rate of the 14900K CPUs by disabling the Thermal Velocity Boost feature on all the CPUs running the Minecraft servers. This has reduced the failure rate to under 5% over longer periods of time (3-4 months).

3. Why has the data center stopped offering new contracts for 14900K-based servers?

Based on the context provided, the data center has stopped offering new contracts for 14900K-based servers for a few key reasons:

The 14900K CPUs were failing at an extremely high rate on the Minecraft servers - the failure rate was around 30% within 1-2 months. 2. The maintenance and replacement costs were becoming very high, as a technician had to physically go into the data center, pull out the blade, and replace the failed CPU every time.

The data center no longer has replacement 14900K CPUs available, as they have stopped offering these servers. The person leasing the 14900K-based servers now doesn't have a way to get replacement CPUs.4. The high failure rate and maintenance costs made the 14900K-based servers an unsustainable offering for the data center. They have decided to stop providing new contracts for these servers.

In summary, the combination of extremely high failure rates, high maintenance costs, and lack of replacement CPUs led the data center to discontinue offering new 14900K-based server contracts.

4. What power and current settings are being used on the Super Micro server board running the 14900K?

Based on the context provided, here are the key power and current settings being used on the Super Micro server board running the Intel 14900K CPU:

Power Limits: - PL1 and PL2 are both set to 253W, which is the Intel specification for the 14900K. - There is no mention of a higher PL4 power limit, which is sometimes seen on gaming motherboards.

Current Limits: - The motherboard has a 200A current limit, which is the maximum current the VRM can handle. - The ICC Max (CPU current limit) is set to the standard range of up to 512A.

Load Line Calibration: - The AC and DC Load Line settings are both set to 1.1 mOhms, which is the Intel specification. - This leads to very high voltages being applied to the CPU.4. Memory Configuration:
The servers are running 128GB of low-profile DDR5-3600 memory, which is the Intel specification for 13th/14th gen CPUs. - They are not running higher memory speeds like DDR5-4400 or DDR5-4000.

So in summary, the Super Micro server board appears to be following the Intel specifications for power, current, and memory settings with the 14900K CPU. The high failure rates seem to be more related to the CPU's inability to handle the sustained workload of the Minecraft servers, rather than any unusual motherboard settings.

Oh man, we've come full circle. I might get to know what YT'ers are saying now with walls of text. An hours worth of content in about 90 seconds.

 

[Saylick]

Oh man, we've come full circle. I might get to know what YT'ers are saying now with walls of text. An hours worth of content in about 90 seconds.

RedGamingTech is gonna be in shambles now.

Just run any one his videos through the summarizer, and then behold! A 10 minute long video gets converted to… a single sentence.

This is the first attempt!

It’s Intel’s 4th attempt at getting the 1st attempt right.

 

[KompuKare]

Oh man, we've come full circle. I might get to know what YT'ers are saying now with walls of text. An hours worth of content in about 90 seconds.

Yep, mostly ignore techtubers as you can't skim read so AI auto summary is great. YouTube media stars cannot be happy though!


So from the AI summary (saved me nearly an hour), even at very conservative Supermicro W680 rates, and after disabling TVB they are so falling at near to 5%. Certainly a lot better than 30% but for tier 1 silicon shouldn't we expect 0.05% or similar?

The $64 billion inflation-adjusted question is, at the <5% failure rate settings how would benchmarks look? And since what does running at something higher than DDR5 3600 do to that 5% failure rate?

Anyway, back to that video. First 5 minutes set an interesting pace? That's about my techtubers tolerance limit, so let me try that...!

 

[igor_kavinski]

Remember all those Intel users who used to brag that Alder/Raptor Lake really isn't that power hungry compared to Ryzen in lightly threaded workloads and then they would post the TPU single threaded power graph to prove their point?



LOL @ their beloved CPU failing in Minecraft servers even at this lower power workload