Deep dive: MSI shows off CAMM2 desktop PC memory

[seeprime]

And here I am with a small bag of Rambus sticks waiting for someone to bring me a socket 423 Pentium 4 to upgrade. It might be to give them to tje recycle guy to melt the pennies worth of gold of of them.

I welcome the chance to test new theoretically faster RAM. But, RDIMMs were also faster than DDR, but lost out due to being more expensive. Make the new kid cheap and it'll be a winner, regardless of what consumers want.

 

[hahahanoobs]

I welcome the chance to test new theoretically faster RAM. But, RDIMMs were also faster than DDR, but lost out due to being more expensive. Make the new kid cheap and it'll be a winner, regardless of what consumers want.

I remember actually falling for that hype on that one. Plus the C-RIMM module? It was a mess, but I was younger then, which is why I'm curious how so many people are falling for the CAMM2 hype when what has been tested has been a nothingburger so far. It's just specs on paper and claims made by the ones trying to sell you something.

It's just crazy how we go back and forth on ignoring hype and waiting for reviews one day, to taking companies' word as gospel the next.

 

[opckieran]

The failure of understanding is simple: Easy custom configuration and user choice. Most people WANT upgradeability that is additive, IE, they want to be able to upgrade their RAM by ADDING more memory WITHOUT removing the existing system memory. The CAMM standard takes away from that dynamic in a way that is completely unacceptable. It is, by design, wasteful and needlessly restrictive.

We consumers don't care. We want our ability of choice and easy customization to continue. The engineers need to go back to the drawing board and come up with solutions that do NOT change the current dynamic.

And clearly the "current dynamic" has reached its limits. And even if it hadn't? It's already well established adding more memory sticks to an existing subsystem tends to come with a bunch of other potential problems! Compatibility, stability, speed reduction can all be affected when adding sticks of RAM. In some cases severe reduction especially when going above 2 sticks and/or double rank! Intel and AMD both have recently come out with statements substantiating that as a matter of fact.

By turning the entire memory subsystem into a single, unified module, the CAMM standard takes away that goofiness while maintaining the ability to upgrade. Which, again, having that ability to upgrade even at an albeit diminished level is better than not having an upgrade path at all, because you know darn well that OEMs want nothing more than to move to onboard DRAM on laptops and will likely be forced to abandon DIMM on desktops too within a generation or two, given that DRAM speeds increase exponentially.

But if you hate the CAMM design philosophy so much, you are welcome to submit your own. It just has to be able to last beyond DDR6, unlike DIMM.

 

[noblody]

What are you basing that on?

On the fact that the only thing needed is the validation of these modules on some actual boards.

 

[noblody]

And here I am with a small bag of Rambus sticks waiting for someone to bring me a socket 423 Pentium 4 to upgrade. It might be to give them to tje recycle guy to melt the pennies worth of gold of of them.

I welcome the chance to test new theoretically faster RAM. But, RDIMMs were also faster than DDR, but lost out due to being more expensive. Make the new kid cheap and it'll be a winner, regardless of what consumers want.

RDRAM was a lot different. It required new memory modules, new interfaces, new memory controller. It provided more bandwidth compared to SDRAM at the expense of latency, power consumption, die size, price. DDR memory made RDRAM lose the bandwidth advantage while keeping power consumption and latency in check. This CAMM thingy is not a different type of memory, only a different memory board that connects to the same memory controller through a different connector LGA vs edge connector. It remains to be seen if the advantages of CAMM outweigh the hassle of redesigning the board layout and the price difference of the LGA socket. When intel first released the LGA socket CPU in 2008 sources mentioned the price of "LGA-775 socket at well under $1 per socket" so probably not a cost issue with the socket.

 

[seeprime]

RDRAM was a lot different. It required new memory modules, new interfaces, new memory controller. It provided more bandwidth compared to SDRAM at the expense of latency, power consumption, die size, price. DDR memory made RDRAM lose the bandwidth advantage while keeping power consumption and latency in check. This CAMM thingy is not a different type of memory, only a different memory board that connects to the same memory controller through a different connector LGA vs edge connector. It remains to be seen if the advantages of CAMM outweigh the hassle of redesigning the board layout and the price difference of the LGA socket. When intel first released the LGA socket CPU in 2008 sources mentioned the price of "LGA-775 socket at well under $1 per socket" so probably not a cost issue with the socket.

Pricing of CAMM memory will lead to its success or failure. The market will choose what's most cost effective.

 

[nitebird]

I think a future shrinking of the MB is inevitable as more stuff moves off the MB onto the CPU, and a shrinking memory profile helps that.

I dream we'll see separate CPU and GPU sockets on the MB with shared central memory. The memory will be the high bandwidth kind to keep the GPU fed and the CPU will have enough X3D L3 cache to keep up. I'm aware of the reasons why this won't happen, but it would be cheaper and so very cool.

 

[Biostud]

CPU support will follow. Shorter, less traces and better signal integrity will make possible faster memory, tighter timings, wider buses. It just won't happen today.

Also as far as I understand it, the reason why you can't run the high speeds on regular DIMMs is the long distances, while it would be easier to run 1:1 with CAMM2 modules. So just because IMC can't run 8000 M/T on regular DIMM doesn't mean it won't be able to run 8000 M/T on a CAMM2 module. Or at least that is the idea, which yet has to be tested IRL.

Hopefully each mboard maker will make at least one CAMM2 motherbaord for each modern socket CPU, which supports it.

 

[Theinsanegamer]

My question is, why not put it on the BACK of the mobo, if its so low profile? More room back there.

I think a future shrinking of the MB is inevitable as more stuff moves off the MB onto the CPU, and a shrinking memory profile helps that.

I dream we'll see separate CPU and GPU sockets on the MB with shared central memory. The memory will be the high bandwidth kind to keep the GPU fed and the CPU will have enough X3D L3 cache to keep up. I'm aware of the reasons why this won't happen, but it would be cheaper and so very cool.

That would require new motherboard standards, and if intel couldnt get BTX to catch on, IDK if a memory profile is gonna do that.

CAMM was a solution to the signal integrity issues faced by thousands of engineers (who are far more knowledgeable about this issue than anyone here) designing laptops. SODIMM is on its way out (there likely won’t even be any DDR6 SODIMMS), and DIMM will inevitably follow. Get used to it!

This attitude is how you end up with a boot up your arse.

 

[opckieran]

This attitude is how you end up with a boot up your arse.

What, because SODIMM is clearly dead? Technology evolves, standards come and go as they always have (CPU slots, AGP, etc). If not LPCAMM2, what exactly is the alternative to soldered DRAM? Because there sure is a lot of hate for LP/CAMM2 here… what, is there some other upgradable standard that can support higher RAM frequencies that I’m missing here?

 

[RaXelliX]

Makes no sense for desktop, space isn't an issue and the slots can easily be positioned near the cpu, eliminating a lot of the memory trace issues they had on laptops with sodimms, along with less issues for path impedance and drive strength as you don't have to worry as much about it being low power (also one module means you have to replace the whole thing rather than upgrading ram a bit at a time...)

Space is absolutely an issue even on desktop. Slots cant be easily positioned near the CPU due to heatsinks interfering with DIMM's.
Also how many people actually upgrade their ram by buying extra modules these days? The most common DDR5 kit size is 2x16GB. This would imply that 32GB wont be enough and aside from professional workloads (unless you count 1000 chrome tabs as pro) gamers dont need 32GB.

For laptops its probably gonna be great, but for PC... it would annoy me. I really really like how you just put the current ram sticks in. It's so easy and fast. No bolts and screws needed. I got the feeling... this will be like an SSD. Annoying to install. Oh well, at least installing stuff is a 1 time thing, especially for the ram.

How often do you swap your RAM? I would think this would annoy reviewers more than end users. I never take out my RAM. Even when im deep cleaning my PC. There's just no point.

Performance would have to be freaking amazing for me to put up with:

- Screws for installation
- More exposed pins
- Taking up more physical space on the motherboard
- Adding more RAM [could] involve removing existing modules
- Cost and compatibility
- Resale market

Screws could be replaced by latches in the future as has happened with M.2. This is just the first iteration.
Exposed pins how? Unless there's a second slot at the back of the board that's empty it's a nonissue. LGA sockets have for more fragile pins.
Takes about as much space as current RAM, but with much better CPU cooler compatibility.
Adding more RAM today already could involves removing existing modules unless you buy the same (lower) speed.
One module vs 2 or 4. I dont see a big cost increase.
Resale market will happen once it becomes a standard. Same as happened with M.2 instead of SATA.

What are you basing that on?

DDR4 started from 2133 and by the end of the generation there were 5500+ modules. DDR5 started at 4800. Currently we are at 8400 and very likely it will reach 11000 by the time DDR6 launches.

The failure of understanding is simple: Easy custom configuration and user choice. Most people WANT upgradeability that is additive, IE, they want to be able to upgrade their RAM by ADDING more memory WITHOUT removing the existing system memory. The CAMM standard takes away from that dynamic in a way that is completely unacceptable. It is, by design, wasteful and needlessly restrictive.

We consumers don't care. We want our ability of choice and easy customization to continue. The engineers need to go back to the drawing board and come up with solutions that do NOT change the current dynamic.

Most people dont upgrade their RAM, ever. They plug in what they bought and it stays there. Also it is not if a second slot could not be placed at the back of the board for this exact purpose. Already DDR5 has problems running four sticks and some people are dreaming of higher speeds and triple channel ie 6 slots. Like I said earlier - common DDR5 kit size is 32GB. By the time 32GB is not enough for most people DDR6 will launch. You also conveniently ignore the fact that adding more RAM currently means buying the same speed as you're running because speed will be determined by lowest common denominator. Or buying higher speed, downclocking and and relaxing timings to match the old sticks. A mess either way. I did this once with DDR4 and despite buying the exact same modules from the same manufacturer I ended up with a different revision that had different memory chips and I still had to manually adjust timings.

 

[Gezzer]

Neato...

The big question isn't if or how much of a better standard it is, but how easy it'll be to shift the market to the new standard. Remember BTX, the form factor to improve on ATX? Better lay out, simpler and more efficient at keeping everything cool. And wasn't Dell a big backer of the standard even releasing a few BTX models?
Just saying...

 

[ZedRM]

Most people dont upgrade their RAM, ever.

You have no idea what you're talking about. Are you a retailer? Do you regularly service and upgrade Workstations/PCs/Laptops? I do. RAM upgrades are a very regular thing, ESPECIALLY for OEM boxes that only have 4GB/8GB/12GB/16GB installed in them. It happens a lot. A LOT!

 

[ZedRM]

Neato...

The big question isn't if or how much of a better standard it is, but how easy it'll be to shift the market to the new standard. Remember BTX, the form factor to improve on ATX? Better lay out, simpler and more efficient at keeping everything cool. And wasn't Dell a big backer of the standard even releasing a few BTX models?
Just saying...

Excellent point.

 

[ZedRM]

And clearly the "current dynamic" has reached its limits.

No, it has not. CAMM is not a proper solution to the problem of timing latency. A proper solution maintains all current functionality while allowing for improvements. CAMM does not do that. Dell/JEDEC needs to go back to the drawing board and do a better job of engineering a solution instead of coming up with something that is so blatantly limiting and wasteful.

 

[Endymio]

You have no idea what you're talking about. Are you a retailer? Do you regular service and upgrade Workstations/PCs/Laptops? I do. RAM upgrades are a very regular thing... It happens a lot. A LOT!

A pest control specialist may remove a lot of dead rats, but it doesn't mean everyone has them.

Once when a colleague stated he was having performance problems on a low-memory PC, I suggested he open the machine and install some additional memory. He -- and the rest of the room -- looked at me as I were a three-headed alien.

"No," he replied, in the tones of one explaining why ordering takeout food is better than the cheaper option of picking up a roadkill corpse on the way home, "I'll simply purchase a faster one."