iMac 2019 - upgraded w 2 extra ram and the speed dropped to 2400 MHz?

[hampus]

Hi everybody.
I bought the iMac 2019 27" base model with 2x 4GB ram in it. Today I upgraded it with 2x 16 GB 2666 MHz DDR4 Corsair Vengeance. At first the computer didn't boot, but after inserting the new Corsair ram into the slots the original ram was, and placing the original ram in the empty slots, it managed to boot. But upon checking that there was registered 40 GB ram, I noticed that the speed of the ram had dropped to 2400 MHz from the original 2666 MHz. Why is this?

The ram is now placed like this (from top to bottom): 4-16-4-16

I have attached a screenshot of the memory submenu in the system information.

I appreciate your help!

 

[wardie]

AFAIK all the RAM modules will drop back to the speed of the slowest as the controller needs to run at a single frequency for it all (could be wrong regarding dual channel pairs though)

 

[DeltaMac]

Did you try moving the RAM sticks so the memory is installed in identical pairs per channel?

 

[macduke]

Should have gone with the Crucial. It's rock solid and inexpensive.

 

[sxl1681]

As DeltaMac mentioned, try pairing the capacities together, 4-4-16-16 or 16-16-4-4 to see if that helps. If it doesn't, I suspect that the Corsair Vengeance is geared for PC's with XMP support. These types of modules usually have conservative speeds programmed into the SPD chip so that the PC will boot/pass BIOS memory test, usually 2133 or 2400 speed. Once the PC fully boots up the BIOS, it loads the XMP profile to boost memory speed and then loads the OS. Macs do not support XMP and will only read what the base value is programmed on the SPD chip. Best to return the memory and buy something labeled as Mac specific or get non-XMP memory.

Note that you can use XMP memory in a Mac since it just ignores the XMP data, but you will need to look up exact part numbers on a manufacturer's site to find what the SPD settings are and get some that is programmed to 2666.

 

[hampus]

Did you try moving the RAM sticks so the memory is installed in identical pairs per channel?

Well, they are - the dual channel is A-B-A-B in the slots.

Funny thing is that the RAM are 2666 MHz but I suspect they are only if overclocked which isn't possible on a Mac

 

[DeltaMac]

Well, they are - the dual channel is A-B-A-B in the slots.

Yes, that works, of course.
I am still suggesting that, rather than A-B-A-B, try plugging the ram into A-A-B-B (So, pair of 16GB in channel A, and other pair of 4GB in Channel B slots.
(Humor me, just try it)
Do an NVRAM reset after moving the sticks around.

 

[smbu2000]

Have you tried only running the Corsair RAM? Also try the NVRAM reset like was suggested earlier.

I run the exact same 2x16GB Corsair RAM kit (CMSX32GX4M2A2666C18) in my 2018 Mac Mini and it has been working fine at 2667MHz since I installed it at the end of last year. I can't run it with the original 8GB kit that was included as there are only 2 memory slots on the Mini.

 

[Kevbasscat]

Because the 2x4Gb ram that came with your iMac is "single rank (single sided)." Anything less than 8GB as general rule is single rank. You added 32Gb of dual rank (ram chips on both sides of board) ram, and they simply don't match. Run a benchmark with 40Gb in then remove the original ram and run a another benchmark with just the 32GB ram . It will be quicker, verifying my point.

 

[hampus]

Because the 2x4Gb ram that came with your iMac is "single rank (single sided)." Anything less than 8GB as general rule is single rank. You added 32Gb of dual rank (ram chips on both sides of board) ram, and they simply don't match. Run a benchmark with 40Gb in then remove the original ram and run a another benchmark with just the 32GB ram . It will be quicker, verifying my point.

Should I keep the old ram and stay at 40 gig or would it be better to take the old out completely and go with with only 32 gig? Is the Mhz difference that meaningfull or is the volume og ram more important?

 

[Kevbasscat]

Should I keep the old ram and stay at 40 gig or would it be better to take the old out completely and go with with only 32 gig? Is the Mhz difference that meaningfull or is the volume og ram more important?

It would be better to remove and KEEP the Apple ram, in case you need to have it serviced. 32Gb DDR4 is faster than 40 GB mixed DDR4.

 

[SaSaSushi]

Because the 2x4Gb ram that came with your iMac is "single rank (single sided)." Anything less than 8GB as general rule is single rank. You added 32Gb of dual rank (ram chips on both sides of board) ram, and they simply don't match. Run a benchmark with 40Gb in then remove the original ram and run a another benchmark with just the 32GB ram . It will be quicker, verifying my point.

This is incorrect. I have been running the 8GB (2x4GB Micron) of single rank Apple stock RAM that came with my 2017 iMac along with 32GB (2x16GB) of dual rank Crucial Ballistix for a total of 40GB with no issues and now slowdowns for the past two years now.

So long as the specs match, single and dual rank RAM should be able to be used together with no problems.

It is more likely a situation where the Corsair RAM is actually overclocked to run at 2666 Mhz. Kingston HyperX caused similar slowdowns in the 2017 iMac when used in combination with the stock RAM.

 

[Kevbasscat]

This is incorrect. I have been running the 8GB (2x4GB Micron) of single rank Apple stock RAM that came with my 2017 iMac along with 32GB (2x16GB) of dual rank Crucial Ballistix for a total of 40GB with no issues and now slowdowns for the past two years now.

So long as the specs match, single and dual rank RAM should be able to be used together with no problems.

It is more likely a situation where the Corsair RAM is actually overclocked to run at 2666 Mhz. Kingston HyperX caused similar slowdowns in the 2017 iMac when used in combination with the stock RAM.

I would kindly challenge you run a benchmark with the Apple ram, then without the Apple ram. Although our computers run fine with the single rank ram, it is not optimal. That is why many who upgrade their 20019 iMacs 2666MHz ram find it only runs at 2400MHz with the 8Gb of Apple ram still included. Try it and post the results. Have a look at the threads belowm they all refernce 2019 iMacs running 2400MHz instead of 2666MHz.

 

[SaSaSushi]

I would kindly challenge you run a benchmark with the Apple ram, then without the Apple ram. Although our computers run fine with the single rank ram, it is not optimal. That is why many who upgrade their 20019 iMacs 2666MHz ram find it only runs at 2400MHz with the 8Gb of Apple ram still included. Try it and post the results. Have a look at the threads belowm they all refernce 2019 iMacs running 2400MHz instead of 2666MHz.

I agree that the Corsair likely runs at full speed on its own. It was a similar situation with the Kingston HyperX in the 2017 iMac. The problems came when it was used in combination with the non-overlocked stock RAM.

I was just disagreeing that the problem was single rank vs dual rank. They can be mixed without issues if the specs are similar.

Some RAM marketed as 2666MHz is actually 2400MHz RAM overlocked to run at 2666MHz. This RAM can work fine on its own at 2666MHz but not when installed in a system together with actual, non-overlocked modules.

Crucial's 2666MHz (including dual rank) RAM works fine installed together with stock RAM at 2666MHz.

 

[Kevbasscat]

I agree that the Corsair likely runs at full speed on its own. It was a similar situation with the Kingston HyperX in the 2017 iMac. The problems came when it was used in combination with the non-overlocked stock RAM.

I was just disagreeing that the problem was single rank vs dual rank. They can be mixed without issues if the specs are similar.

Some RAM marketed as 2666MHz is actually 2400MHz RAM overlocked to run at 2666MHz. This RAM can work fine on its own at 2666MHz but not when installed in a system together with actual, non-overlocked modules.

Crucial's 2666MHz (including dual rank) RAM works fine installed together with stock RAM at 2666MHz.

That's what I purchased. 32Gb of dual rank ram from Crucial, that was manufactured specifically for my iMac model. But if you look at threads below, you'll notice everyone has the same issue, Their iMac is supposed to run at 2666MHz but it only runs at 2500MHz due to the single/double rank mix. Test it yourself and see. You're never going to believe me until you do.

 

[SaSaSushi]

That's what I purchased. 32Gb of dual rank ram from Crucial, that was manufactured specifically for my iMac model. But if you look at threads below, you'll notice everyone has the same issue, Their iMac is supposed to run at 2666MHz but it only runs at 2500MHz due to the single/double rank mix. Test it yourself and see. You're never going to believe me until you do.

I believe you. Unfortunately I don't own the 2019 iMac, only a 2017 iMac which runs 2400MHz by default.

I've seen several reports from users upgrading with Crucial RAM in this thread who've reported no problems running it together with their stock RAM at 2666MHz.

Are you sure you got the correct Crucial modules?

 

[mikehalloran]

Are you sure you got the correct Crucial modules?

Since Micron is one of the OE RAM suppliers, that should be looked at.

Crucial is the retail arm of Micron.

https://www.crucial.com/usa/en/comp...ple/imac-(retina-5k,-27-inch,-2019)-imac-19,1

 

[Kevbasscat]

Since Micron is one of the OE RAM suppliers, that should be looked at.

Crucial is the retail arm of Micron.

https://www.crucial.com/usa/en/compatible-upgrade-for/Apple/imac-(retina-5k,-27-inch,-2019)-imac-19,1

I think the point here is assist the OP in resolving their issue of having the extra two ram modules purchased that run at 2400MHz not 2666MHz. This is due to an incompatibility between the single rank ram supplied by Apple and everyone else's dual rank ram for cards larger than 4-8Mb. I have no issues with my ram, it runs perfectly with the 8Gb of Apple removed, but if you look below at the threads you'll notice every post is a question about their 2019 iMac running at 2400MHz not 2666MHz as spec'd. I challenged people to run a benchmark with the Apple ram and without the Apple ram. So far, it's all talk, no action. I stand by my advice.
[doublepost=1562636950][/doublepost]

 

[SaSaSushi]

I think the point here is assist the OP in resolving their issue of having the extra two ram modules purchased that run at 2400MHz not 2666MHz. This is due to an incompatibility between the single rank ram supplied by Apple and everyone else's dual rank ram for cards larger than 4-8Mb. I have no issues with my ram, it runs perfectly with the 8Gb of Apple removed, but if you look below at the threads you'll notice every post is a question about their 2019 iMac running at 2400MHz not 2666MHz as spec'd. I challenged people to run a benchmark with the Apple ram and without the Apple ram. So far, it's all talk, no action. I stand by my advice.

I'm sorry to hear about your problems and I hope you get them resolved.

You are not helping the OP solve their problems, however, by suggesting that the problem is combining single and dual rank modules. There are users in the thread I linked to above using single and dual rank modules together in their 2019 iMacs at 2666MHz.
[doublepost=1562638070][/doublepost]

Since Micron is one of the OE RAM suppliers, that should be looked at.

Crucial is the retail arm of Micron.

Indeed, which is why I always highly recommend Crucial for iMac upgrades.

 

[kschendel]

The difference in total system performance between 2400 MT/s and 2666 MT/s (note, not MHz) is likely to be about 1-5% and most often on the low end of that range. If this were a Ryzen-based machine, the number might be a bit higher, but Intel CPU's aren't excessively memory speed sensitive.

(DDR4-2400 runs at 1200 Mhz and you get 2 transfers per clock, which is why it's Double Data Rate. It runs at 2400 million transfers per second, MT/s, not 2400 MHz. Yes, the manufacturers and memory vendors all think we're too stupid to understand proper units and they will say 2400 MHz, but it's still wrong regardless.)

 

[Kevbasscat]

The difference in total system performance between 2400 MT/s and 2666 MT/s (note, not MHz) is likely to be about 1-5% and most often on the low end of that range. If this were a Ryzen-based machine, the number might be a bit higher, but Intel CPU's aren't excessively memory speed sensitive.

(DDR4-2400 runs at 1200 Mhz and you get 2 transfers per clock, which is why it's Double Data Rate. It runs at 2400 million transfers per second, MT/s, not 2400 MHz. Yes, the manufacturers and memory vendors all think we're too stupid to understand proper units and they will say 2400 MHz, but it's still wrong regardless.)

Thank you for the clarification, quite helpful. I fear referring to ram ie. DDR4-2400 in MT/s would cause confusion to many since it's pretty much the world default? I have such issues anyone who uses an incorrect term because in their eyes the believe they're helping an ignorant public user base. We're all ignorant at some point so it's these little misrepresentations that further the misunderstanding and create confusion. As someone from an electronics testing and engineering background these clarifications are important for understanding and communicating it's proper function and ability. This "dumbing down" for the public's sake is what creates further confusion and therefore misguided conclusions.

One more question. Since the 8 gig of ram that is shipped in new iMacs is single rank, and any added ram is daul rank, does it mean one side of the board (single) only as opposed to both sides (dual), or does it refer to single rank as "Single Date Rate," also? I have seen benchmarks of Dual Data Rate, ram with Apple's single rank ram included, and the benchmarks are lower than without Apple's inclusion. This does not mean it's reported incorrectly, for example 2400MT/s as opposed to the default 2666MT/s, in system reports, it simply means the overall performance of the unit is slightly slower. This has been documented or reported in a number of places on MR, and for the life of me, I cannot seem to find one at the moment. I may have to dig out my original 2x4Gb of Apple ram and do the benchmarks myself, for everyone's consideration, as there are conflicting reports both ways. I don't have an agenda to prove anything, but as someone with my background, the absolute truth and a real understanding of the fundamentals is essential to a good "clear" working knowledge and application of our systems. At the moment I'm not physically able to do the ram benchmarks, as I'm healing from a leg issue related to my knee replacement. This is why I haven't posted mine. If anyone is able to test and report, this may go a long way to understanding the conflicts users are having with incompatibility issues or poorer performance. Any help would be greatly appreciated. Thank you for your clarification.

 

[kschendel]

...I fear referring to ram ie. DDR4-2400 in MT/s would cause confusion to many since it's pretty much the world default?...

That's the excuse that is made, but really, how confusing is MT/s anyway? compared to the confusion caused when someone looks at their actual memory frequency using a tool like e.g. CPU-Z and it says 1200 MHz? I don't really care if it's confusing -- and I claim it isn't -- it's simply wrong. Hz is defined as "cycles per second", not "random things you want to count per second". But my Standards side is showing, I guess.

As to the other thing, memory rank basically refers to the number of "chip select" lines that you need to address all of the memory on any given stick. You can't always tell the rank of a memory stick from its size. Most DDR4 8GB sticks are single rank, but if they are made from 4Gbit chips they would be dual rank. I think all current 16GB sticks are dual rank, and 32GB sticks would be quad rank, but I could be mistaken. Memory rank has various subtle effects on performance that are hard to predict without knowing the details of the memory controller, and I don't know that much about current Intel integrated memory controllers. (Nor Ryzen controllers, for that matter.) Ranks are additive across slots, so if you have a memory channel with two slots, one with a 1R stick and one with a 2R stick, that channel is going to see 3 memory ranks -- which may, or may not, cause it to decide to slow down.

A lot of memory speed and timing decisions are made by the firmware (UEFI, BIOS), and without knowing a lot more than I do about what goes on inside there, it would be hard to say what effects occur when memory is added to the computer.

 

[Kevbasscat]

That's the excuse that is made, but really, how confusing is MT/s anyway? compared to the confusion caused when someone looks at their actual memory frequency using a tool like e.g. CPU-Z and it says 1200 MHz? I don't really care if it's confusing -- and I claim it isn't -- it's simply wrong. Hz is defined as "cycles per second", not "random things you want to count per second". But my Standards side is showing, I guess.

As to the other thing, memory rank basically refers to the number of "chip select" lines that you need to address all of the memory on any given stick. You can't always tell the rank of a memory stick from its size. Most DDR4 8GB sticks are single rank, but if they are made from 4Gbit chips they would be dual rank. I think all current 16GB sticks are dual rank, and 32GB sticks would be quad rank, but I could be mistaken. Memory rank has various subtle effects on performance that are hard to predict without knowing the details of the memory controller, and I don't know that much about current Intel integrated memory controllers. (Nor Ryzen controllers, for that matter.) Ranks are additive across slots, so if you have a memory channel with two slots, one with a 1R stick and one with a 2R stick, that channel is going to see 3 memory ranks -- which may, or may not, cause it to decide to slow down.

A lot of memory speed and timing decisions are made by the firmware (UEFI, BIOS), and without knowing a lot more than I do about what goes on inside there, it would be hard to say what effects occur when memory is added to the computer.

From someone who's job it was to insure proper reference, and test it, these distinctions are important. There can be subtle ways in which things can be misinterpreted leading to incorrect conclusions, that's why. If you were writing code to the OS, something as simple as that could derail all your work. The difference is MHz is frequency as opposed to MT/s being a transfer speed. They have two very different waveforms, definitions and functions. There is too much ambiguity as to ram already and for those of us trying to understand and then implement changes we need absolute clarity not muddying up our understanding or misunderstanding. I was on the original USB design team. I wrote the test specs and qualification procedures/specs. If I had substituted MHz for MT/s, omg, Intel, Microsoft, you name them would have had a field day with me and my incompetence, thus undermining my credibility. Lol, that cable was soooo slow, we kept asking, "why again are you trying to build something Apple's 1394 (firewire 400) was already killing it at? It was about 10 MT/s literally. Like a moose with flippers.

I'm going to dig into how exactly ram works and specifically DDR4, as it applies to right now, so maybe we can answer some of these questions being raised with certainty. I'll report back afterward, hopefully with insight and pragmatic solutions. Until then, I hope other solutions are found.

 

[mikehalloran]

Hz is defined as "cycles per second",

Sorry. If you're going to get technical, Hertz is defined as completed cycles per second.

This is why everyone went to it in the 1970s. The old designation, cps, was confusing since it was used refer to either single (half) or completed (double/full) cycles.

You have to be of a certain age to remember when this was phased in. Guilty.

 

[kschendel]

Sorry. If you're going to get technical, Hertz is defined as completed cycles per second.

Ugh. Got me on that one. I knew I should have looked it up.

You have to be of a certain age to remember when this was phased in. Guilty.

Same here.

I recall that another reason for Hertz was that "cycles per second" was often just shortened to "cycles" or "megacycles" etc, losing the time element. Yeah, it was implied, but still.