Intel very misleading RAM speed support.

[C1REX]

I bit of a bombshell:

In this video
https://youtu.be/ZK6jAc8LY48?si=3kGVrZhMQphhFM5M

At 14:18 Wendell is saying that the official supported memory for recent Intel CPUs is only 4400MT/s
The bigger number of 5600MT/s is allegedly only for 2 DIMMS motherboard. It means motherboards that only have two RAM slots. Not 4 with 2 slots occupied.
Most people don't know that having empty RAM slots is more difficult for CPU memory controllers than not having extra empty slots at all.

That's why premium ASUS Apex series has only two RAM slots. However, such 2dimms motherboards are very rare.

Like this one: https://www.newegg.com/p/N82E16813119659

[r.linder]

C1REX a écrit :

Illusion of Progress a écrit :

I think we may simply disagree on whether it's a big deal or not.

Wendel looks surprised and disappointed about the supported RAM speed on Intel. According to him, AMD may actually have better specs for 2 RAM sticks on 4 DIMMs motherboards. He talks about this at around 24:40. He says that AMD is honest about their RAM speeds when Intel small print is not. Maybe this information is as obvious as water being wet for you, but for me, learning that Intel supported RAM speed is 4000-4400 MT/s is wild.

Sure, almost everyone uses EXPO/XMP anyway, but that's not the point. 14900K based gaming servers weren't stable with 5600 MT/s RAM, but they were stable at 4400 MT/s. This seems to be unrelated to the degradation issue.

Which can't be 100% proven until the microcode bugs are fixed, which is a more pressing issue regardless because processors are failing regardless of DRAM.

[Illusion of Progress]

C1REX a écrit :

Illusion of Progress a écrit :

I think we may simply disagree on whether it's a big deal or not.

Wendel looks surprised and disappointed about the supported RAM speed on Intel. According to him, AMD may actually have better specs for 2 RAM sticks on 4 DIMMs motherboards. He talks about this at around 24:40. He says that AMD is honest about their RAM speeds when Intel small print is not. Maybe this information is as obvious as water being wet for you, but for me, learning that Intel supported RAM speed is 4000-4400 MT/s is wild.

I'm not on a DDR5 platform, let alone one of both AMD and Intel, and let alone more than one of each in order to have enough experience on many boards/RAM to talk about who has better support.

But I thought it was often considered that Intel often attained better speeds in that area?

My overall stance to this is not I know for certain that Intel isn't being less honest (they at least are about other things...), because I don't know if they are or aren't, but simply to say that what I'm reading, "supported RAM speed goes down in certain scenarios", those scenarios basically amounting to "any configuration that is heavier may need lower frequencies to stabilize", has unfortunately been the "standard operating procedure" on the RAM side of things for a while now. It's definitely worse with DDR5, but that's not news either.

I asked where Intel's contradiction is. They say "up to 5,600 MHz+" and yeah that's pretty vague as it may as well be saying nothing at all in my mind. "The "up to" means it might not get that high, and the "+" means it might get higher, so then... where is it? So yeah, AMD being firm on a supported speed with the "+" is a bit better, but that means they have to be more conservative as a result, because these are speeds they definitely try and guarantee. So as far as I'd say, that's unable to contradict with much because it's already claiming it can be both lower or higher.

Now if this was stealth changed, like other things Intel's apparently been stealth changing lately (maybe the "up to" didn't exist originally or in some spots), or if Intel currently has it elsewhere that it is "5,600 MHz+" without the "up to" and has 4,400 MHz firmly stated as the maximum guaranteed elsewhere (where?)... then yeah, that would be a contradiction. Is that the case though? I'm asking, not denying. Where's Intel's contradictory supported claims?

C1REX a écrit :

Illusion of Progress a écrit :

Sure, almost everyone uses EXPO/XMP anyway, but that's not the point. 14900K based gaming servers weren't stable with 5600 MT/s RAM, but they were stable at 4400 MT/s. This seems to be unrelated to the degradation issue.

The two seem likely to be very related to me, no?

He's referencing 4,400 MHz as the supported speed Intel also claims (again, where is this claimed, because I'm only finding that "up to 5,600+ MHz" claim, although I admit I'm not exactly hunting for it like it's treasure).

Was this claimed, or is this what he thinks it "should be" due to the fact that reducing memory speeds was found to be one of many things (key part?) that helped stabilize some (key part?) problematic systems, and in a scenario where chip degradation was a possibility (very, very key part?). You have to isolate those things first, no?

[A&A]

Illusion of Progress a écrit :

He's referencing 4,400 MHz as the supported speed Intel also claims (again, where is this claimed, because I'm only finding that "up to 5,600+ MHz" claim, although I admit I'm not exactly hunting for it like it's treasure).

https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/002/processor-sku-support-matrix/

[Illusion of Progress]

A&A a écrit :

Illusion of Progress a écrit :

He's referencing 4,400 MHz as the supported speed Intel also claims (again, where is this claimed, because I'm only finding that "up to 5,600+ MHz" claim, although I admit I'm not exactly hunting for it like it's treasure).

https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/002/processor-sku-support-matrix/

Thanks.

So from that, I don't really see a contradiction in what Intel themselves states.

But I'm guessing what this is the actual part of importance...

"1DPC refer to system with one DIMM slot routed per 64-bit channel, 2DPC refer to system with two DIMM slots routed per 64-bit channel."

That is admittedly a little strange to me, because the way it's worded implies the DIMM slots simply existing counts, whereas I would have thought this typically referred to DIMM slots that were populated. At least, that's how I've always interpreted that. Now I'm wondering if this is perhaps how they all define this though.

Now that I see this distinction, I sort of get what Wendell was stating. I thought the fuss was that maximum supported frequencies dropped with more DIMMs, or with higher rank DIMMs, populated into the configuration, and my reaction was "yeah, that's how it always works, and also, water is wet" but that definition is... interesting.

If they've all do/have been defining it this way though, then it's back to what I've been saying; it's not exclusive to Intel. We've known for a long time that officially supported speeds are often lower than profile speeds, and that profile speeds are often "considered overclocking".

At the end of the day, more DIMMs per channel/more ranks per channel means harder to stabilize, which means lower supported speeds. This still all comes down to being described by that.

And when the speeds most people get in practice is higher, this won't matter to most. I mean, two DIMMs of single rank? Regardless of whether Intel says 4,400 MHz or 5,600 MHz here, most people are going to get way higher in practice anyway (not that more clarity should be the norm). Well, ignoring all the issues the chips have going on...

[r.linder]

AMD has the same thing, the CPU and motherboard impact what memory specifications are actually guaranteed to work.

[[-iD-]]

not all ram is created equal.

[[-iD-]]

oh and mobos are supposed to be the ones that make correct timings and frequencies for ram.
its most likely mobo engineering issue. They supply the QVL. its amazing how lazy some mobo manufacturers are getting these days. we only had a few when i started that had issues. and asus was the best.... now asus has gone very downhill at least for mobos. their monitors are amazing.

[C1REX]

Illusion of Progress a écrit :

A&A a écrit :

https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/002/processor-sku-support-matrix/

Thanks.

Now that I see this distinction, I sort of get what Wendell was stating. I thought the fuss was that maximum supported frequencies dropped with more DIMMs, or with higher rank DIMMs, populated into the configuration, and my reaction was "yeah, that's how it always works, and also, water is wet" but that definition is... interesting.

If they've all do/have been defining it this way though, then it's back to what I've been saying; it's not exclusive to Intel. We've known for a long time that officially supported speeds are often lower than profile speeds, and that profile speeds are often "considered overclocking".

It was a bombshell for Wendell, and it’s mind blowing for me. I knew that 2 DIMM motherboards are better, but I didn’t expect that the mere existence of two extra unpopulated slots would drop the supported speed from 5200 MT/s to 4000 MT/s. I may be wrong here, but AMD supported speed in the same scenario is 4400 MT/s. Both drop to just 3600 MT/s with all 4 slots populated.

Is this relevant to people who use XMP and EXPO profiles? I don’t know. Maybe it’s not. Or maybe it does matter if the profiles boost the clocks so much. Maybe it does affect silicon degradation when we are so far off supported speeds.

It also raises questions about the existence of 4 DIMM DDR5 gaming motherboards. Why even have 4 slots when the hit to performance due to the extra slots is so huge? Or the hit to stability if we push clocks higher?

[[-iD-]]

C1REX a écrit :

It was a bombshell for Wendell, and it’s mind-blowing for me.

you just described paranoia fear and doubt.

[C1REX]

-iD- a écrit :

C1REX a écrit :

It was a bombshell for Wendell, and it’s mind-blowing for me.

you just described paranoia fear and doubt.

So you knew that Intel supported DDR5 speeds on a most common configuration is 4000MT/s?

[[-iD-]]

C1REX a écrit :

-iD- a écrit :

you just described paranoia fear and doubt.

So you knew that Intel supported DDR5 speeds on a most common configuration is 4000MT/s?

no idea what you are talking about, a youtube video isn't proof of anything, maybe his main board does not support that memory.

[C1REX]

-iD- a écrit :

C1REX a écrit :

So you knew that Intel supported DDR5 speeds on a most common configuration is 4000MT/s?

no idea what you are talking about, a youtube video isn't proof of anything, maybe his main board does not support that memory.

This is Intel’s official info.
The small print Wendell was referring to:

https://edc.intel.com/content/www/us/en/design/products/platforms/details/raptor-lake-s/13th-generation-core-processors-datasheet-volume-1-of-2/002/processor-sku-support-matrix/

[A&A]

C1REX a écrit :

Is this relevant to people who use XMP and EXPO profiles? I don’t know. Maybe it’s not. Or maybe it does matter if the profiles boost the clocks so much. Maybe it does affect silicon degradation when we are so far off supported speeds.

It also raises questions about the existence of 4 DIMM DDR5 gaming motherboards. Why even have 4 slots when the hit to performance due to the extra slots is so huge? Or the hit to stability if we push clocks higher?

https://www.techpowerup.com/review/intel-core-i9-13900k/5.html

They tested an i9 13900K with 2 dimms on a motherboard with 4 RAM slots. Look at Aida64. The memory runs at 6000MHz, but it's important to note that the controller runs at 3000MHz (Gear 2, half the memory speed). The memory controller at standard voltage (1.1v for DDR5) should not be able to achieve the same clock as the RAM in this case and surely most PCs with such processors and RAM kits run like this.

[r.linder]

A&A a écrit :

C1REX a écrit :

Is this relevant to people who use XMP and EXPO profiles? I don’t know. Maybe it’s not. Or maybe it does matter if the profiles boost the clocks so much. Maybe it does affect silicon degradation when we are so far off supported speeds.

It also raises questions about the existence of 4 DIMM DDR5 gaming motherboards. Why even have 4 slots when the hit to performance due to the extra slots is so huge? Or the hit to stability if we push clocks higher?

https://www.techpowerup.com/review/intel-core-i9-13900k/5.html

They tested an i9 13900K with 2 dimms on a motherboard with 4 RAM slots. Look at Aida64. The memory runs at 6000MHz, but it's important to note that the controller runs at 3000MHz (Gear 2, half the memory speed). The memory controller at standard voltage (1.1v for DDR5) should not be able to achieve the same clock as the RAM in this case and surely most PCs with such processors and RAM kits run like this.

Pretty sure the controller clock is half of the RAM speed because DDR makes it an effective speed of 6000. AMD FCLK is the same way.

[_I_]

yea, it ticks on up and down, for ddr
wave cycle in hz = up+down
sd and earlier tick only when moving down, ddr ticks both state changes up and down