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

[_I_]

did something change, or are you just guessing?

2 dimms populated is less work on the cpu than 4, its been that way since the imc has been on the cpu die

[C1REX]

_I_님이 먼저 게시:

did something change, or are you just guessing?

2 dimms populated is less work on the cpu than 4, its been that way since the imc has been on the cpu die

You are right.
2 populated slots are easier than 4 populated slots but having only two slots is even easier due to how they are connected. The reason is slower daisy chain connection with 4 slots.

BuildZoid explains it better:

https://youtu.be/Q8XLvZTqyX8?si=LAkyVGtKADUQWkWq

[Illusion of Progress]

This isn't new. This isn't exclusive to Intel.

Memory stability has typically always been harder to achieve when these factors increase...

1. DIMM count.

2. Rank count.

3. Frequency.

4. Tighter timings.

The difference is it's been getting more pronounced, especially with DDR5.

Try and pair an AM5 platform with four, dual rank DIMMs (these would be 32+ GB capacity DIMMs right now) and see how likely you are to hit that sweet spot of 6,000+ MHz.

Most people on DDR5 platforms are probably going with a pair of single rank DIMMs (2x 16 GB, single rank). This is rather pedestrian insofar as how demanding it is. You won't see the impact there. Double the DIMMs and/or double the ranks, which you have no choice but to do if you want higher capacities, and it's now much harder to stabilize. You often need to drop frequency (and/or loosen timings) to get it stable.

Further reading, if you're interested in a deeper covering of it...

https://forum-en.msi.com/index.php?threads/ram-explained-why-two-modules-are-better-than-four-single-vs-dual-rank-stability-testing.363139/

This is by no means an Intel thing. Intel has a lot on their plate that is worth criticism, but this isn't one of them.

[C1REX]

Illusion of Progress님이 먼저 게시:

This isn't new. This isn't exclusive to Intel.

Most people don’t know that 2dimms only motherboards have faster connection and potentially better RAM speed. It’s not a common knowledge.

Now I wonder if AMD’s official 4800 RAM speed is for normal 4dimms motherboards or rare 2dimms ones.

[A&A]

"Memory Types

Intel® processors come in four different types: Single Channel, Dual Channel, Triple Channel, and Flex Mode. Maximum supported memory speed may be lower when populating multiple DIMMs per channel on products that support multiple memory channels."

They tell you, but no one reads...
Even if it runs at official speeds, it will probably run on Gear 2 and not many people will know, but who tells you what "Gear" is anyway.

[Illusion of Progress]

Look at the factors I mentioned that make a memory configuration harder to stabilize. One of them is more DIMMs. The two slot motherboards therefore lack the possibility of one of those factors being put into play to begin with.

And it's not common knowledge for two reasons.

1. Most people don't buy higher capacities of RAM, which is where the chances of there being four DIMMs instead of two, or dual rank instead of single rank, or perhaps both, are more likely. Those are the things that impact this. The other people familiar with it would be those pushing RAM (very high frequencies/timing tuning). Most people probably buy a pair of single rank DIMMs, set the profile speeds, and move on.

2. This wasn't as pronounced before (but it's definitely been an on again, off again thing, seen more often early in a RAM generation). DDR5 is especially limiting in this regard.

But it's neither exclusive to DDR5, nor Intel. If you look at some motherboard specification pages, you'll often see something like "#DPC #R xxxx+ MHz" ("DPC" is DIMMs per channel, and "R" is ranks) where the "#" is either 1 or 2, and the more DIMMs and ranks you add, the lower the memory frequency will be. An imaginary example might be something like this...

1DPC 1R 4800+ MHz
1DPC 2R 4400+ MHz
2DPC 1R 4000+ MHz
2DPC 2R 3600+ MHz

It might not necessarily scale linearly/that way, but it's an example that the ceiling keeps dropping. The first is two single rank DIMMs, the last is four dual rank DIMMs. The memory frequency you can stabilize keeps getting lower as DIMM count/ranks go up. It's physics, not a failing of Intel specifically.

[_I_]

intel and amd have always rated the imc way slower than it can go

they just guarantee it will work at that speed with 4 dimms and whatever timings the dimms support in their jdec profile

[C1REX]

_I_님이 먼저 게시:

intel and amd have always rated the imc way slower than it can go

they just guarantee it will work at that speed with 4 dimms and whatever timings the dimms support in their jdec profile

If understand Wendell right, and please correct me if I’m wrong, the base specs for 14900K for 2 sticks on a regular 4dimm motherboard is 4400MT/s. Way less than official 5600MT/s would suggest.

Even Tom is in disbelief.

[r.linder]

Why does it even matter when it's the least of people's concerns with Intel right now? They were doing fine running even as high as 7000+ so I don't see why the advertised memory specification being partially inaccurate is such a big deal. It's possible for any CPU, AMD or Intel, to struggle with any memory kit for any reason.

Doesn't hold a candle to the eTVB bug or VIA oxidation.

[_I_]

C1REX님이 먼저 게시:

_I_님이 먼저 게시:

intel and amd have always rated the imc way slower than it can go

they just guarantee it will work at that speed with 4 dimms and whatever timings the dimms support in their jdec profile

If understand Wendell right, and please correct me if I’m wrong, the base specs for 14900K for 2 sticks on a regular 4dimm motherboard is 4400MT/s. Way less than official 5600MT/s would suggest.

Even Tom is in disbelief.

because its up to the board mfg to make the ram paths to the ram be able to support their 'oc' speeds
with the worst design or implantation of it, it will still be able to hit intels spec with 4 dimms

https://ark.intel.com/content/www/us/en/ark/products/236773/intel-core-i9-processor-14900k-36m-cache-up-to-6-00-ghz.html

Memory Types
Up to DDR5 5600 MT/s
Up to DDR4 3200 MT/s

i dont see the problem

[C1REX]

r.linder님이 먼저 게시:

Doesn't hold a candle to the eTVB bug or VIA oxidation.

If Wendell is right and Intel truly doesn't guarantee stability even at 4800MT/s with 2 sticks of RAM on a regular 4 DIMM motherboard, and the specs are for 2-DIMM-only motherboards, then it just adds to the dishonesty of Intel and questions the quality of their memory controller.

But yes. Sounds like nothing comparing to degrading and failing CPUs.

[r.linder]

C1REX님이 먼저 게시:

r.linder님이 먼저 게시:

Doesn't hold a candle to the eTVB bug or VIA oxidation.

If Wendell is right and Intel truly doesn't guarantee stability even at 4800MT/s with 2 sticks of RAM on a regular 4 DIMM motherboard, and the specs are for 2-DIMM-only motherboards, then it just adds to the dishonesty of Intel and questions the quality of their memory controller.

But yes. Sounds like nothing comparing to degrading and failing CPUs.

The problem can also literally be linked to the microcode bug itself since the instability seems to be heavily related to the Ring bus, even running at 5 GHz was making things unstable when the Ring bus was also overclocked but seemed to not happen when it wasn't.

If the 12900K doesn't share the same issue with memory then it's probably related to the existing major issues.

[C1REX]

_I_님이 먼저 게시:

i dont see the problem

According to Wendel this specs is for rare 2 DIMMS only motherboard and is lower for regular 4dimms motherboard.

I've checked Intel's benchmarks notes and DDR5 data is for 2dimms only motherboards (Asus ROG Maximus Z790 Apex)
https://edc.intel.com/content/www/us/en/products/performance/benchmarks/desktop/

[_I_]

the intel h chipsets are limited to 2 dimms

is that what you are talking about?

[r.linder]

_I_님이 먼저 게시:

the intel h chipsets are limited to 2 dimms

is that what you are talking about?

There are two slot Z690 and Z790 motherboards (and other top end chipsets) that are made specifically with memory overclocking in mind

The presumption is that Intel used these boards as the baseline for some reason, I think it's just because of instability with the Ring bus