Intel processors crashing Unreal engine games (and others)

[moinmoin]

keeping the hate train going

So you don't want these Intel processor issues to be fixed and affected customers to have a product working as intended? Or what exactly is your issue with this thread you yourself keep posting in?

 

[coercitiv]

It seems AMD copied them a bit too well. So well that Intel decided to change to not resemble AMD

Friendly reminder that (some) mobo makers also managed to screw things up for AMD and customers. AMD should also have known better than to let them play with the red buttons. Always have limits.

 

[KompuKare]

Friendly reminder that (some) mobo makers also managed to screw things up for AMD and customers. AMD should also have known better than to let them play with the red buttons. Always have limits.

I guess the biggest difference is that without these auto-overclocking motherboards Intel would have won even fewer benchmarks.

And Intel marketing (and engineering) did not know that is very very unlikely. That motherboard manufacturers play loose with the specs? Well Asus for one have been doing that for decades so that in reviews they are at least 1% ahead.

 

[coercitiv]

Buildzoid published a new video, and someone on reddit was kind enough to write down the TL;DR. Here it is, emphasis mine:

1. Many LGA1700 boards disable settings like CEP, TVB, eTVB (against Intel recommendation), which means the CPU may try and boost to a higher clock value, even if voltage or temperature parameters aren't suitable for the CPU to try and hit those clocks
2. Many LGA1700 boards apply terrible AC/DC loadline values, and board vendors are sometimes change these values with different BIOS updates, which in conjunction with the above, means many 13th/14th Gen CPUs aren't stable at stock (this could also explain why we started seeing these issues post-14th Gen release, maybe board vendors started messing with loadline values to push clocks as much as possible)
3. Intel assumes board vendors and OEMs will use the tools they are supplied with, to verify their board and VR designs are applying correct and consistent voltages, but many board vendors are seemingly not doing enough testing here to ensure stability
4. Buildzoid says there's no reason why any Z790 motherboards he's aware of should be defaulting to the 125W/253W 'Performance' profile, even $200 Z790 boards are capable of pushing the 253W/253W 'Extreme' profile and 125W will limit an i9-13900K/i9-14900K in some games
5. Buildzoid doesn't think 13/14th Gen stability issues have much, if anything to do with PL1, PL2, Tau or Current values and is more likely caused by points 1 and 2.

 

[DrMrLordX]

What about the Subzero EVO here?

I haven't seen it tested, but it seems to have the same integrated block/pump setup seen on so many AiOs and that probably means: weak pump.

 

[igor_kavinski]

Still waiting for Intel to publish an end-user guide on setting values properly for their CPUs...

 

[Thibsie]

Only someone ignorant would use that with the lid on. At stock, it will keep throttling and never realize its full potential, unless you are fine with running only games and no serious multicore workloads but then it will perform almost similar to 14900K so what's the point of spending extra money?

It is sold with a lid. That's it.
Intel could have sold it without. They didn't.

 

[igor_kavinski]

It is sold with a lid. That's it.

No denying that. But if reviewers compare stock 14900KS to ARL Core Ultra 9, it would be unfair treatment of 14900KS or Intel specifically told them in their review guide to do that. Though I wouldn't mind if both are compared delidded. That would be fair too.

 

[Ranulf]

Still waiting for Intel to publish an end-user guide on setting values properly for their CPUs...

That is apparently asking too much. Its all on the mobo makers...so good luck everyone!

 

[igor_kavinski]

Intel CPUs crashing and Intel White Knight worrying about Turin/Granite Ridge power consumption

Man, I love the approach some people take to defend Intel's honor

It's a good thing Lisa Su is a sensible human being. Otherwise, they would be slinging dirt on her, "Intel CPUs crashing??? BUT BUT BUT Lisa was seen passed out at some event not too long ago. Things not looking good for AMD!".

 

[Thunder 57]

So AT released a review of the 14900KS today and apparantly ignored the memo (edit, not menu, derp) to keep things "consistent".

This does pose questions when it comes to testing and reviewing Intel's 14th and 13th Gen processors. We have been considering our standpoint on this, as we will typically test at the default motherboard settings with memory set to JEDEC specifications of the specific processor we're testing. For this review, we will be testing how we usually test, as this fits within the realm of keeping things consistent.

Ah yes to keep things consistent. Consistently crashing maybe, but consistent.

They say they are going to re-run the benchmarks with Intel's guidelines. We'll see about that. And if so, they should have held off on publishing this until it was complete. As someone mentioned in the comments, how can you have a conclusion when you don't even have all of the data? The article would have gotten a lot more attention if they used the new guidelines for their benchmarks. They would've been among the first to review the 14900k with the "correct settings".

Also, their results look ass backwards at times. How is the 7900 vanillia beating the 7950X/7950X3D so much in some of those benchmarks?

 

[DrMrLordX]

Also, their results look ass backwards at times. How is the 7900 vanillia beating the 7950X/7950X3D so much in some of those benchmarks?

It's possible that they have chipset drivers set up wrong (or not at all) when testing the 7950X3D, leading to threads migrating to the wrong CCD.

 

[igor_kavinski]

From the 14900KS review:

We are using DDR5 memory on the Core i9-14900KS, as well as the other Intel 14th Gen Core series processors, including the Core i9-14900K, the Core i7-14700K, Core i5-14600K, and Intel's 13th Gen at the relative JEDEC settings.

Where does Intel specify the JEDEC RAM settings for its CPUs? A link would've been nice. Ditto for AMD.

On my Z790 mobo, 12700K defaults to DDR5-3600 if default JEDEC profile is chosen. So it's the profile of the RAM, not the CPU.

 

[DAPUNISHER]

From the 14900KS review:



Where does Intel specify the JEDEC RAM settings for its CPUs? A link would've been nice. Ditto for AMD.

On my Z790 mobo, 12700K defaults to DDR5-3600 if default JEDEC profile is chosen. So it's the profile of the RAM, not the CPU.

Intel only list up to 5600MT/s for the 14900KS so I presume that's what Gavin is using?

https://www.intel.com/content/www/u...-36m-cache-up-to-6-20-ghz/specifications.html

Everything past that is XMP/overclocking I suppose.

 

[igor_kavinski]

Intel only list up to 5600MT/s for the 14900KS so I presume that's what Gavin is using?

But there is no mention of CAS latency or the other subtimings there. So how did Gavin settle on what values to use for those?

 

[DAPUNISHER]

But there is no mention of CAS latency or the other subtimings there. So how did Gavin settle on what values to use for those?

Ask him.

 

[zir_blazer]

Where does Intel specify the JEDEC RAM settings for its CPUs? A link would've been nice. Ditto for AMD.

On my Z790 mobo, 12700K defaults to DDR5-3600 if default JEDEC profile is chosen. So it's the profile of the RAM, not the CPU.

Here is Raptor Lake-S/Raptor Lake Refresh-S DDR5 speeds support list: https://edc.intel.com/content/www/u...lume-1-of-2/010/processor-sku-support-matrix/

Note than the DPC definition is wrong, it should be SPC (Slots per Channel), not populated DPC (DIMMs per Channel). Basically, if you want out-of-the-box specification compliant 5600 MHz (Or 4800 MHz in Alder Lake-S), you can ONLY get that on 2 slots boards, whereas the vast majority of boards is 4 slots. 4 slots, even on the best scenario (2 SPC with 1 DPC populated that is Single Rank) , are capped to 4400 MHz (Same as Alder Lake-S). So if you get 3600 I assume that you're using 4 modules that are Dual Rank, which is the worst case scenario. Those are Intel defaults, and anything above that is overclocking.

 

[maddogmcgee]

I'm not sure what is more entertaining. Following this debacle or cities skylines 2.

 

[igor_kavinski]

So if you get 3600 I assume that you're using 4 modules that are Dual Rank, which is the worst case scenario. Those are Intel defaults, and anything above that is overclocking.

I'm using only 2 but they are EXPO modules with a few JEDEC profiles in there. No XMP.

Funny thing is, I can't even seem to get the JEDEC profile timings working which the ASROCK UEFI is reading from the SPD registers. Must be bad RAM or even worse, crap IMC. It was in 2022 though so things might have improved with the latest BIOS update. Gotta fire that PC up soon and check.

 

[igor_kavinski]

That's from a 12900KS. It's interesting because 12900K, 13900K, 13900KS, 14900K and 14900KS all show something like this:



Maybe the 12900KS thermal limits are there for a reason? As in, they noticed degradation on CPUs running at higher temps. For everything else, they decided to be reckless.

We will probably never know the truth. Any ex-Intel engineers wherever you are, now is the time for you to speak anonymously

 

[zir_blazer]

Funny thing is, I can't even seem to get the JEDEC profile timings working which the ASROCK UEFI is reading from the SPD registers. Must be bad RAM or even worse, crap IMC. It was in 2022 though so things might have improved with the latest BIOS update. Gotta fire that PC up soon and check.

Modules have a SINGLE JEDEC profile with a single supported clock speed, not multiple. There is also a sort of checkbox field for multiple possible CAS Latencies that have to be individually checked. The BIOS is supposed to be able to calculate with a formula lower clock speeds based on JEDEC clock speeds and possible CAS Latencies. Check this, you have "CAS Latencies Supported", and a list of possible JEDEC settings.

Which modules you have that default to 3600 MHz? If you have only two, it should be 4400.

 

[lakedude]

AT explains that they presume an "average user" isn't going to get into BIOS settings at all so they use whichever profile requires zero user intervention.

XMP/EXPO require fiddling around in the BIOS whereas JEDEC should just come up by default.

I take this to mean they are not fiddling with any of settings like CAS, just using stock/JEDEC.

 

[zir_blazer]

XMP/EXPO require fiddling around in the BIOS whereas JEDEC should just come up by default.

I take this to mean they are not fiddling with any of setting like CAS, just using stock/JEDEC.

If you don't interact with the BIOS, it will pick the highest JEDEC profile that is also compatible with Memory Controller electrical load. LGA 1700 4 DDR5 slots boards = 4400 MHz max.

And don't even get me started on DDR4 where the vast majority of DIY modules have 3200 MHz as an XMP profile and uses a 2400-2666 for JEDEC standard for reasons that I can't understand. They could use 3200 / 22-22-22 / 1.2V standard JEDEC and another 3200 1.35V XMP with tighter Timmings if they wanted.

 

[coercitiv]

And don't even get me started on DDR4 where the vast majority of DIY modules have 3200 MHz as an XMP profile and uses a 2400-2666 for JEDEC standard for reasons that I can't understand. They could use 3200 / 22-22-22 / 1.2V standard JEDEC and another 3200 1.35V XMP with tighter Timmings if they wanted.

For context on what @zir_blazer is annoyed about: there are 3200 / 22-22-22 / 1.2V modules on the market, and they most likely lack an XMP profile. So the customer can pick one or the other, but never both.

 

[lakedude]

For context on what @zir_blazer is annoyed about: there are 3200 / 22-22-22 / 1.2V modules on the market, and they most likely lack an XMP profile. So the customer can pick one or the other, but never both.

Sorry partner, not following. You mentioned one thing and then talk about one or the other not both?

Your example talks about lacking an XMP profile but zir_blazer is talking about modules that do have an XMP 3200 profile.

I know you were just trying to help but now I'm confused...