I like it, Also as someone who has gone back to air cooling cpu's, CAMM2 opens up the possibility for new cpu air cooler designs that won't have to worry about blocking ram modules.
Yeah, it's freaking annoying having to find a cooler and RAM that work together well, and without making speed, latency, or other sacrifices. Now if they only spread it into two separate modules, and possibly 2 connectors on a MoBo, so you can start with like 32GB and get 64GB by adding a module, rather than replacing that which you payed good money for, and have no use for the old one, since with everything hardware speed related, the requirements and price go up faster than the efficiency goes down, if much at all. I remember running a 486DX80 (last predecessor of the K series), graphics card and all on a 250W power supply that was more than sufficient!
JEDEC is an industrial association consisting of over 300 member companies (not too different to the way the USB-IF works). A lot of the members do contribute resources (either funding or engineering work) to the association, and new standards are jointly developed and ratified with the help of member companies. Seeing as Dell Inc is one of the 300+ members of JEDEC, it's not unusual to see some of their development work ending up in final CAMM specs, and being credited for the work you do is presumably quite standard in this space.
Dell is part of but not with enough engineers in JEDEC to get a majority voting body of just Dell. So they had to convince others that CAMM2 was better than SO-DIMM. As SO-DIMM and DIMM are not that good CAMM was intended to fix it. JEDEC took it hopefully improved it and made it so anyone and everyone can use it. Having a 64GB module would be good and depending on cost i can see SIs trying to get consumers onto 32GB and 64GB instead of 4GB 8GB and 16GB. Of course in desktops not laptops as 64GB in a laptop is excess power usage for most that isn't needed.
@@DerekDavis213 I mean to consumers. it means you can finally get performance, energy efficiency and upgradability without sacrificing one. Lpddr5X is more energy efficient and way faster
Thank you for the update, I originally wrote off CAMM as a Dell trying to force people to buy more expensive RAM that is compatible with the motherboard.
Asus too, but at the time this was recorded, no others had revealed CAMM2 boards. I think I saw HUB say Asrock wasn't sure it would make it into production
OCuLink was never great, it‘s mechanically very fragile leading to signal integrity issues after a while and it only handles up to PCIe Gen4. “MCIO” is the connector type that got developed as a response to all the issues encountered and it should even be able to handle PCIe Gen6.
Since handhelds use a custom PCB anyways it just makes more sense to direct solder when the APU is direct soldered. Consumers are going to be far more concerned with increasing storage than increasing memory on a package they can't upgrade the APU. Unless you want a really thick handheld, of course. I've always felt I'd FAR rather return to inch thick laptops if I were able to upgrade the CPU and GPU, but of course laptop makers _don't_ want that cause you'd buy less laptops.
I'm really hoping CAMM2 takes off, We need it really bad for laptops but can see it working really well with desktop motherboards because it will make cooling memory so much easier
Yeah finding laptops that has upgradable RAM is becoming more difficult because every laptop manufacturer want the laptop to be as thin and light as possible. Even some of gaming laptops right now has entirely soldered RAM. I really hope CAMM really does change that
what? in reality it makes situation worse ram cooling wise. one side of this ram always be sided with mobo vs old one with air ventilation on both sides. Also socket wise, old 2 hooks no instruments ram out. now you need to unscrew it .... maybe there in future this camm2 can be useful, for now it have only huge disadvantages vs old one
Thank you Gordon and PC World for this. This deep dive was needed and will be valuable to refer to, going forward. CAMM2 is an a good and arguably, much needed innovation in that space.
I want camm2 on an itx motherboard. That is where I think this will be most useful outside of speed. I’m already things of ways to make my build better if I didn’t have to worry about dimms
In case you missed it, the module standard calls for quite a large board size to be supported, or rather, we don't know what the size is. As shown on the floor, CAM for ITX is too big, only the laptop version works. Or so it seems. We really don't know, since the "open standard" is closed. If you look carefully at what is shown on the floors, the MB is responsible for the desktop version cooling. It just appears to be too big. Also, back side connectors tend to run through the MB, so combining back side connectors with the desktop version of CAM, seem to be conflicting.
@@goblinphreak2132 You need to learn to read specs. Seriously learn to read them. That way, you might avoid looking as stup!d as this. If I asked you for the keep out sone on the back of an ATX board, would you even know where to look to be able to answer? Because that is a hint, of even if this folly was true, which it is absolutely not, it still would be a standard violation. Just one you do not even know exists.
@@Das_Tipples have you see how high the camm2 chips sit on the board. msi already showed it off. its basically the same height as m.2.... so yes, its already equal to m.2
Finally, something in the PC building space I can get into. I've not cared for fan stacking, side port PSUs nor hidden motherboard cables. Board makers have to be willing to maker low to mid-range boards with slots for these if they want it to be adopted. I'm not spending $250 for RAM then another $350 for a motherboard that works with it.
Any new technology won't hit the mid range until it's the universal solution, let alone the low end. They need a carrot on a stick to upsell you with. We see it today with OLED. OLED is _not_ a new technology and it isn't more expensive to produce than LCD with an LED backlight. There's no reason for OLED to cost what it does other than they weren't successful in scamming people into thinking they needed 8k for their overpriced high end displays, so they're going to milk OLED as long as possible. They gatekeep high refresh the same way, putting 120hz panels in cheap TVs, but only let you use it with "motion smoothing" interpolation. They won't give you 120 _connections_ without paying for it, which is ridiculous because all those HDMI connectors cost MORE than royalty free DP. They're literally paying _more_ to gatekeep the feature to high margin displays. Obviously adding 6 more USB slots, another $10 dollars in VRMs, and $2 in RGB shouldn't magically make a $150 motherboard become $399, but they purposely remove $20 in _desired_ features to sway you away from the low margin boards. That way they can advertise "We have affordable boards" while leaving out "but we made sure you won't actually want it." CAMM2 tracing is _cheaper_ than 4xDDR5 slots, but trust me, they'll pay _more_ to make sure you don't have what's desirable without paying extra for it. Everyone thinks GPUs have ridiculous margins but they're actually one of the _lowest_ margin items in your system compared to the mobo, cooling, PSU, and peripherals. Several companies have openly stated that much. TLDR: don't hold your breath.
Interesting! I wonder if these CAMM2 modules could be stacked with a staggered configuration with a special staggered connector on the board with staggered mounting. I wonder how that would effect bandwidth too. I look forward to seeing performance benchmark comparisons on these versus the vertically mounted DRAM we use currently.
That's what I was thinking at first the way you often see it with sodimms, but with DDR5 _cooling_ is now a concern. For obvious reasons stacked+heat=bad day.
@@nadtz The latency is what I am more interested in. The ROG Ally runs something like 6400 CL19 stock due to having the RAM soldered right next to the CPU. CL19 on DDR5!!! I think CAMM will get us closer to that on desktop too. Overclockers already achieved over 10,000 MHz DDR5 with a Ryzen 5 8500G btw.
@@Violet-ui Overclocked RAM makes very little difference in real life performance. Remember, the CPU spends most of its time in ultra-fast L1 L2 and L3 caches. Main memory speed is secondary.
@@DerekDavis213 This is just wrong. The 5800X3D is about 20% faster in games than the 5800X because it has the cache to improve memory performance. That means that the 5800X without the extra cache is losing all that performance due to too slow memory. I personally have my memory OCed to where I get 10-15% more fps than XMP in CPU bound games. In some games you can get over 20% better performance with very fast RAM. This is why people still OC their memory. Memory OC does still significantly affect CPU performance.
When I saw CAMM2 during the last tech show, I knew this is groundbreaking moment we are witnessing It's been a minute since I've seen anything like that, last thing I can remember was SSDs
I just wonder if this is actually going to change anything in laptops and consoles since manufacturers will probably still just solder them to the motherboard rather than paying the extra cost for the modules and licensing
It's an interesting trade-off for small form factor, where it's great for creating room for CPU coolers, but terrible for taking up more precious board space. It'll be interesting to see if it ends up being adopted there, maybe just with limitations of the smaller modules.
It would be a cool engineering challenge to be shackled to the design restrictions of the iPad Pro M4 and see how far you can get making it upgradeable. De-serialisation would be a good start.
@@impuls60 AMD actually has pretty good IMC but their chiplet design is what's holding them back. Look at their monolithic APUs, they can go very high with memory speeds.
P.S. The variable mask around you and your physical background make it look like you have long, flowing locks blowing in the wind. You need to headbang with Steve of Gamers Nexus... for shits and giggles... windmill to the sound of the loudest consumer fans you've both ever heard.
I want this to be an option when I build my next rig... not only are there ALL the benefits you talked about here, but there is the whole "custom build" aspect of it too... Integrating the CPU cooler with the CAMM2 cooler opens up A WHOLE NEW DESIGN ASPECT to custom cooling/builds ... thin, long single plates going across the whole board sorta counters the thin, long line of the GPU and opens up a TON of new ideas for deign! But as I've said in many other discussion about CAMM2 the BIGGEST negative IMO is that you have to replace ALL your RAM if a single chip fails, whereas in DIMM systems you only have to replace 25% (or 50% depending on your setup) if a chip fails... and that's a BIG difference in pricing! PLUS you won't be able to run your machine on "just one stick" if you have a 2 stick setup and one of them dies... you're "not supposed to" run on just one stick (just like you're "not supposed to" used a mix of RAM speed OR manufacturers) but it WILL WORK... but you CANNOT do that with CAMM2.... if just one chip dies, you WILL NOT be able to use your computer AT ALL... so for the PCMR DIY crowd it would be smart to have a spare CAMM2 module for just this scenario..... BUT this might create more of a repair community that has the PROPER equipment to replace chips that use LGA connections :)
It true that at release we will only see non stackable variants. The standard support stackable variants but this will be further down the road for desktops at least
Open-source and yet you need to ask permissions to show these PDF schematics 🤧 THIS IS NOT how open-source works people of JEDEC 😤 Thanks as always for the awesome breakdown Gordon ❤ Have a great week ❤
So I think putting camm2 on their project zero MB would be a good choice because there are more expensive MB for aesthetics and I imagine that some people would like the cleaner look of this with a metal shield over it to make it look clean. You don't get BTF or Project Zero for performance or cost you get it for looks. I don't think it is about proof of concept. I think some exec was like that would look cool and then the engineers made it. That being said I really want CAMM on ITX boards since without the RAM height limiting coolers we might get better CPU coolers for SFF.
This is brilliant but the question is cost of the CAAM2 modules because I doubt it will be cheap for say 16Gb I hope it is. But off topic I wish board manufacturers for the 24 pin connector etc would turn them in their sides to make the cable less sticky out. I know someone did it but it should be standard now
Nice stuff, I hope everything can eventually go to one RAM standard with CAMM2, & we can do away with soldered RAM in laptops, & maybe even Android tablets, but having said that, I'm still good with running desktops, & laptops with DDR4 RAM for my needs, & can't see myself upgrading for at least a few years now having 64GB in my 2 desktops, & 24GB in my laptops for less intensive taks.
The issue the performance of CAMM2 allows for much higher speeds. So desktops will want it, especially given that CAMMs can benefit from bigger and more robust cooling options. We'll see what the future holds.
The 2DPC problem with low memory capacity of SODIMM has been one of the most frustrating points to consider when building a new DDR5 computer. Looking forward for when CAMM2 fully replaces SODIMM in the DIY space.
I mean, if you REALLY that concern with OCing your RAM, you don't want to stick any other than same exact 2 stick of DIMM anyway. So just combined it to single CAMM2 and everything is just as the same. Beyond that, they still sells DIMM motherboard if you don't care about that shit
Like I said in another channel with sticks you can buy 2 sticks and then add 2 more later but with this thing you have to do the full commitment on getting the exact capacity of memory for the rest of the life of your PC. If you get one of these modules that is 16GB for example, you need to buy a whole thing with 32GB then try to sell the old one. On laptops this makes sense but on desktop...
Alleged upcoming new gen of high performance APUs are gonna benefit greatly by this new standard setting in now. can't wait to have a the ultimate sff build in the future
Except with APUs having it on package like Lunar Lake is doing will be far more beneficial. This is my concern with Strix Halo. 40CUs will be choked hard in random games if they have to rely on system memory, regardless if it's DDR5 or CAMM2DDR5. No matter how close it's physically impossible to be closer than surrounding the die as GPUs currently do it.
@@zodwraith5745 Current camm2 capability is around 8500mt/s. micron on next year already planning on releasing 9600 mt/s camm2. even more with years to come. so unless integrated ram is blowing through 10k mt/s, i don't see why not. there's certainly demand for it
The 4 studs are probably made for a small fan or heatsink. Eventually they will have their own little cooler/heatsink combo right next to the cpu cooler lol. Probably will look pretty cool i bet.
I don't see how this will work well in servers, with 8/10/12 memory channels per cpu. Even with "quad" modules, that's still 2-4 large flat modules that need to be squeezed into a very limited area.
With the modules we've seen, yeah, doesn't seem space-efficient. But maybe the plan is to use the shorter card with its better connections and put something more dense than regular DDR5 packages on it.
"Battery life is not as important on gaming laptops" you take that back! Lots of people use gaming laptops on battery. And since in them the components are generally more power-hungry than in ultrabooks, it's even more important to squeeze every bit of efficiency as possible from all components, including RAM. However, RAM power usage is severely overestimated. Each So-dimm module in my gaming laptop consumes just 0.13 watts at idle and during basic tasks like browsing or watching movies, so in total 1\4 a watt combined. Compare that to the CPU which consumes from 2.5 to 5 watts in the same tasks or display which consumes from 2 to 10 watts depending on brightness setting. There are also ports, wifi module and so on, so in total, the percentage of power consumed by RAM is really small. Maybe 5%, and that's in gaming laptops, probably even less for LPDDR5.
That's me. JEDEC requires the copyright statement but I thought I'd spice it up with some comic sans. You can also find papyrus too. ua-cam.com/video/jVhlJNJopOQ/v-deo.html
I have 64GB RAM in my system with two empty slots for 2 more 32GB modules for a total of 128GB, so way more than I have ever used. This CAMM2 sounds good, but they may want to put two sockets on MoBo's, to make upgrading easier, so you don't have to replace all of it at once, and a a high cost.
Thank you for the insight. info on this new ramware helping when deciding to buy more faster/expensive dimm ddr5 now and use the sticks in futurer z890, or wait for this camm tech to mature, or just upgrade to dimm ddr6 in future
In these photos the CAMM2 seems to take a lot of space in terms of width compared to a tall narrow DIMM. My concern is whether this will fit well on a Mini ITX board. Which don't have much width between CPU socket and edge of the board.
I'm certainly hoping CAMM2 will be available on some MATX or ITX boards off the bat. Though MSi's MPower motherboard is interesting too. Hoping that may get carried over to Z890. Kinda hard to imagine why they'd debut it for Z790 when Intel 15th is around the corner. 🤔 I'm just getting anxoius to do a new build. The 9700k and 2080 in my Alienware Area 51m laptop has carried me for this long, but I'm ready for an upgrade, and I'm going back to a desktop system, albeit a MATX system.
So I assume by this that memory makers assume they will have no problem increasing speeds but have a real problem increasing the number of channels for bandwidth?
I'd just like to interject for a moment. What you're referring to as CAMM2, is in fact, DDR5/CAMM2, or as I've recently taken to calling it, DDR5 plus CAMM2
Camm makes much more sense for signal path routing, but I see many many optimisations to come, firstly it’s all up against one side where by all rights the pins could be right under each chips legs and distributed evenly. Secondly it could be on the back side of the motherboard and almost mapped directly into the cpu pin for pin with mm’s of trace
Actually te studs around the module is needed to click in cooling on top of the module. without the cooling, they will burn up. Sure, this cooling probably will got some rgb goodies on it. But main reason is cooling mounting
I am kind of surprised that they don't connect this to the back of the motherboard directly opposite the CPU. True cases would need to be re-designed for access but would make the module much closer to the CPU
@@LongliusYep, & I'm not gonna crap on this one, even if it's gonna be costly at first, as any standard is better than soldered RAM that so many systems are using now, or only having 1 slot for a RAM upgrade with a low amount like 8 GB soldered on the board(thanks Lenovo 🤦) meaning you end up with weird configs like 24 GB, or 40 GB when the board, and processor in theory could take 64 GB, or even 128 GB.
CAMM 1 was proprietary for DELL. It had a lot of problems. Lenovo, Micron and other OEMs took over and overhaul it into CAMM 2 which is open. Now the entire CAMM connector is replaceable. It uses the same bendable pins like the CPU sockets but if you bend it -> just replace it. This has to be introduced for CPU sockets as well.
@@mowtow90 I hadn't even considered modular CPU sockets. I spent hours bending back LGA 1700 pins on my current mobo, because I couldn't prove that it wasn't me that bent them. Good ole' Newegg return policy.
Do we know how much restate CAMM2 is going to need in non-Z dimensions? I can see that the single MSI motherboard example doesn't seem to have the big CAMM2 installed or denoted as to which size CAMM2 was installed in the picture.
I don't understand why some people are excited about this. It doesn't seem to offer anything better than what we already have outside of being low profile while being substantially more expensive. Seems more like it's focused on higher profit margins for manufacturers.
Gordo! Hope you're kicking it dude. I mean those slacker kids' butts, LoL. Come on we, old guys need to use our walkers together! Hey mine's got wheels!
I like it, Also as someone who has gone back to air cooling cpu's, CAMM2 opens up the possibility for new cpu air cooler designs that won't have to worry about blocking ram modules.
Yeah, it's freaking annoying having to find a cooler and RAM that work together well, and without making speed, latency, or other sacrifices. Now if they only spread it into two separate modules, and possibly 2 connectors on a MoBo, so you can start with like 32GB and get 64GB by adding a module, rather than replacing that which you payed good money for, and have no use for the old one, since with everything hardware speed related, the requirements and price go up faster than the efficiency goes down, if much at all. I remember running a 486DX80 (last predecessor of the K series), graphics card and all on a 250W power supply that was more than sufficient!
Agreed, cooler manufacturers will be very happy about this.
noctua baby
Gordon is such a pro! This man can break down complex specifications and explain it to a layman audience. Very enjoyable and informative presentation.
The Dell engineers are credited in the JEDEC documents? That is amazing.
JEDEC is an industrial association consisting of over 300 member companies (not too different to the way the USB-IF works). A lot of the members do contribute resources (either funding or engineering work) to the association, and new standards are jointly developed and ratified with the help of member companies.
Seeing as Dell Inc is one of the 300+ members of JEDEC, it's not unusual to see some of their development work ending up in final CAMM specs, and being credited for the work you do is presumably quite standard in this space.
Dell is part of but not with enough engineers in JEDEC to get a majority voting body of just Dell. So they had to convince others that CAMM2 was better than SO-DIMM. As SO-DIMM and DIMM are not that good CAMM was intended to fix it. JEDEC took it hopefully improved it and made it so anyone and everyone can use it. Having a 64GB module would be good and depending on cost i can see SIs trying to get consumers onto 32GB and 64GB instead of 4GB 8GB and 16GB. Of course in desktops not laptops as 64GB in a laptop is excess power usage for most that isn't needed.
I'm so glad computing is getting interesting again.
How is this memory module any more interesting than SO-DIMMs that we've been using for years?
@DerekDavis213 supports LPDDR5X so laptop memory manufacturers have no excuse to Sauter ram anymore
@@Westlee_ There is a big excuse to solder ram: charge very high prices for ram upgrades, because the customer cannot upgrade memory themselves.
@@DerekDavis213 I mean to consumers. it means you can finally get performance, energy efficiency and upgradability without sacrificing one. Lpddr5X is more energy efficient and way faster
Shorter Traces, less latency, better signal integrity; easier to hit higher frequency/tighter timings then sodimm, or regular dimms. @DerekDavis213
Thank you for the update, I originally wrote off CAMM as a Dell trying to force people to buy more expensive RAM that is compatible with the motherboard.
Gotta give kudos to Dell for not keeping this proprietary, having laptops with upgradeable low power memory makes me happy.
MSI isn't the only one with a CAMM2 board, AsRock showed one off at Computex as well, I believe the AsRock was "Z890".
Asus too, but at the time this was recorded, no others had revealed CAMM2 boards. I think I saw HUB say Asrock wasn't sure it would make it into production
So the camm2 boards are releasing soon, pricing?
@@xlr555usa Probably around the time DDR6 starts coming out. I'm sure some will do it in DDR5 as well.
Excited about the new standards, CAMM2, Oculink, EDSFF. all neat.
OCuLink was never great, it‘s mechanically very fragile leading to signal integrity issues after a while and it only handles up to PCIe Gen4. “MCIO” is the connector type that got developed as a response to all the issues encountered and it should even be able to handle PCIe Gen6.
Love the Comic Sans + Papyrus slide.
Hope they use this on future handheld pc
Since handhelds use a custom PCB anyways it just makes more sense to direct solder when the APU is direct soldered. Consumers are going to be far more concerned with increasing storage than increasing memory on a package they can't upgrade the APU.
Unless you want a really thick handheld, of course. I've always felt I'd FAR rather return to inch thick laptops if I were able to upgrade the CPU and GPU, but of course laptop makers _don't_ want that cause you'd buy less laptops.
Excited to see how CAMM2 does on desktop.
I'm really hoping CAMM2 takes off, We need it really bad for laptops but can see it working really well with desktop motherboards because it will make cooling memory so much easier
Yeah finding laptops that has upgradable RAM is becoming more difficult because every laptop manufacturer want the laptop to be as thin and light as possible. Even some of gaming laptops right now has entirely soldered RAM. I really hope CAMM really does change that
what? in reality it makes situation worse ram cooling wise. one side of this ram always be sided with mobo vs old one with air ventilation on both sides.
Also socket wise, old 2 hooks no instruments ram out.
now you need to unscrew it ....
maybe there in future this camm2 can be useful, for now it have only huge disadvantages vs old one
@@vasilypupkin5533 Heatsinks are coming soon. It's inevitable.
@@vasilypupkin5533how often do you change ram sticks?? It would take like 2 minutes to swap.
also less interfering with CPU coolers
Thank you Gordon and PC World for this. This deep dive was needed and will be valuable to refer to, going forward. CAMM2 is an a good and arguably, much needed innovation in that space.
Thanks Gordon, great video with tons of info.
Very very interesting, let's see where it goes!
And of course thanks Gordon for the analysis.
Gordon knows and explains like nobody else. We who are about to nerd salute you!
I want camm2 on an itx motherboard. That is where I think this will be most useful outside of speed. I’m already things of ways to make my build better if I didn’t have to worry about dimms
In case you missed it, the module standard calls for quite a large board size to be supported, or rather, we don't know what the size is. As shown on the floor, CAM for ITX is too big, only the laptop version works. Or so it seems. We really don't know, since the "open standard" is closed. If you look carefully at what is shown on the floors, the MB is responsible for the desktop version cooling. It just appears to be too big. Also, back side connectors tend to run through the MB, so combining back side connectors with the desktop version of CAM, seem to be conflicting.
@@FrodeBergetonNilsen no the board size diesnt matter. You can fit the new camm2 on the back of an itx where nothing exists. Computing is evolving
@@goblinphreak2132 You need to learn to read specs. Seriously learn to read them. That way, you might avoid looking as stup!d as this. If I asked you for the keep out sone on the back of an ATX board, would you even know where to look to be able to answer? Because that is a hint, of even if this folly was true, which it is absolutely not, it still would be a standard violation. Just one you do not even know exists.
@@goblinphreak2132 Clearance would have to be in equal spec of a rear M.2
@@Das_Tipples have you see how high the camm2 chips sit on the board. msi already showed it off. its basically the same height as m.2.... so yes, its already equal to m.2
I'm looking forward to getting CAMM2 on my next machine. Hopefully there are CAMM2 boards at the Arrow Lake launch later this year.
When Arrow Lake launches, next gen Ryzen will be out there, and superior too.
Arrowlake seems like a move in the right direction. But I'll wait a gen or two for Intel to prove they can make more efficient and stable CPUs.
MSI had one at their computex booth this year. Looks like a nice feature.
Finally, something in the PC building space I can get into. I've not cared for fan stacking, side port PSUs nor hidden motherboard cables. Board makers have to be willing to maker low to mid-range boards with slots for these if they want it to be adopted. I'm not spending $250 for RAM then another $350 for a motherboard that works with it.
Any new technology won't hit the mid range until it's the universal solution, let alone the low end. They need a carrot on a stick to upsell you with. We see it today with OLED. OLED is _not_ a new technology and it isn't more expensive to produce than LCD with an LED backlight. There's no reason for OLED to cost what it does other than they weren't successful in scamming people into thinking they needed 8k for their overpriced high end displays, so they're going to milk OLED as long as possible. They gatekeep high refresh the same way, putting 120hz panels in cheap TVs, but only let you use it with "motion smoothing" interpolation. They won't give you 120 _connections_ without paying for it, which is ridiculous because all those HDMI connectors cost MORE than royalty free DP. They're literally paying _more_ to gatekeep the feature to high margin displays.
Obviously adding 6 more USB slots, another $10 dollars in VRMs, and $2 in RGB shouldn't magically make a $150 motherboard become $399, but they purposely remove $20 in _desired_ features to sway you away from the low margin boards. That way they can advertise "We have affordable boards" while leaving out "but we made sure you won't actually want it." CAMM2 tracing is _cheaper_ than 4xDDR5 slots, but trust me, they'll pay _more_ to make sure you don't have what's desirable without paying extra for it.
Everyone thinks GPUs have ridiculous margins but they're actually one of the _lowest_ margin items in your system compared to the mobo, cooling, PSU, and peripherals. Several companies have openly stated that much.
TLDR: don't hold your breath.
Love the font choices : )
Interesting! I wonder if these CAMM2 modules could be stacked with a staggered configuration with a special staggered connector on the board with staggered mounting. I wonder how that would effect bandwidth too. I look forward to seeing performance benchmark comparisons on these versus the vertically mounted DRAM we use currently.
That's what I was thinking at first the way you often see it with sodimms, but with DDR5 _cooling_ is now a concern. For obvious reasons stacked+heat=bad day.
OC cooling may require Vapor-Chamber heat spreaders on the CAMM2 bottom side.
@@zodwraith5745
CAMM2 on desktop. Probably going to hit 8000MHz CL20 or something crazy like that
I heard 8400 mentioned in another video
@@nadtz The latency is what I am more interested in. The ROG Ally runs something like 6400 CL19 stock due to having the RAM soldered right next to the CPU. CL19 on DDR5!!!
I think CAMM will get us closer to that on desktop too.
Overclockers already achieved over 10,000 MHz DDR5 with a Ryzen 5 8500G btw.
@@Violet-ui It will be interesting to see. Wonder how long before the stacked designs come out and how much of a difference that will make.
@@Violet-ui Overclocked RAM makes very little difference in real life performance. Remember, the CPU spends most of its time in ultra-fast L1 L2 and L3 caches. Main memory speed is secondary.
@@DerekDavis213 This is just wrong.
The 5800X3D is about 20% faster in games than the 5800X because it has the cache to improve memory performance. That means that the 5800X without the extra cache is losing all that performance due to too slow memory.
I personally have my memory OCed to where I get 10-15% more fps than XMP in CPU bound games. In some games you can get over 20% better performance with very fast RAM.
This is why people still OC their memory. Memory OC does still significantly affect CPU performance.
Fantastic overview.
Gordon crushing it during computex. He's doing everything for PCWorld coverage besides smelling the handhelds.
When I saw CAMM2 during the last tech show, I knew this is groundbreaking moment we are witnessing
It's been a minute since I've seen anything like that, last thing I can remember was SSDs
Thanks Gordon!
I just wonder if this is actually going to change anything in laptops and consoles since manufacturers will probably still just solder them to the motherboard rather than paying the extra cost for the modules and licensing
It's an interesting trade-off for small form factor, where it's great for creating room for CPU coolers, but terrible for taking up more precious board space. It'll be interesting to see if it ends up being adopted there, maybe just with limitations of the smaller modules.
thank you for this deep dive
This was great, thankyou
I"m keen to see this in NUC style devices for the home lab,
It would be a cool engineering challenge to be shackled to the design restrictions of the iPad Pro M4 and see how far you can get making it upgradeable. De-serialisation would be a good start.
Price will basically determine how quickly it is adopted.
It'll be expensive until economies of scale kick in
That and if Intel/Amd fixes their memory controller.
@@impuls60 AMD actually has pretty good IMC but their chiplet design is what's holding them back. Look at their monolithic APUs, they can go very high with memory speeds.
The other is if you can get stable higher speed ram also. Performance will drive demand driving down costs.
@@Forlong21 Yap.. I just hope to see the cas latency finally go back down. I want the same latency as ddr3 or ddr4 ram with ddr5 bandwidth.
I hope you're doing as well as you can, Gordon. ❤
P.S. The variable mask around you and your physical background make it look like you have long, flowing locks blowing in the wind. You need to headbang with Steve of Gamers Nexus... for shits and giggles... windmill to the sound of the loudest consumer fans you've both ever heard.
I want this to be an option when I build my next rig... not only are there ALL the benefits you talked about here, but there is the whole "custom build" aspect of it too... Integrating the CPU cooler with the CAMM2 cooler opens up A WHOLE NEW DESIGN ASPECT to custom cooling/builds ... thin, long single plates going across the whole board sorta counters the thin, long line of the GPU and opens up a TON of new ideas for deign!
But as I've said in many other discussion about CAMM2 the BIGGEST negative IMO is that you have to replace ALL your RAM if a single chip fails, whereas in DIMM systems you only have to replace 25% (or 50% depending on your setup) if a chip fails... and that's a BIG difference in pricing! PLUS you won't be able to run your machine on "just one stick" if you have a 2 stick setup and one of them dies... you're "not supposed to" run on just one stick (just like you're "not supposed to" used a mix of RAM speed OR manufacturers) but it WILL WORK... but you CANNOT do that with CAMM2.... if just one chip dies, you WILL NOT be able to use your computer AT ALL... so for the PCMR DIY crowd it would be smart to have a spare CAMM2 module for just this scenario..... BUT this might create more of a repair community that has the PROPER equipment to replace chips that use LGA connections :)
I'm certainly looking at camm2 and wondering if this will be the new standard when ddr6 comes out
Good work Gordon 🤩
It true that at release we will only see non stackable variants. The standard support stackable variants but this will be further down the road for desktops at least
So we all agree it's now called camm2
Yo nice lecture man !!
i hope to see this becoming a new standard. I wanna see this being used in ITX systems too.
Would love to see it for Zen 5 :)
Open-source and yet you need to ask permissions to show these PDF schematics 🤧
THIS IS NOT how open-source works people of JEDEC 😤
Thanks as always for the awesome breakdown Gordon ❤
Have a great week ❤
CAMM seems like a perfect complement to full cover water blocks.
So I think putting camm2 on their project zero MB would be a good choice because there are more expensive MB for aesthetics and I imagine that some people would like the cleaner look of this with a metal shield over it to make it look clean. You don't get BTF or Project Zero for performance or cost you get it for looks. I don't think it is about proof of concept. I think some exec was like that would look cool and then the engineers made it. That being said I really want CAMM on ITX boards since without the RAM height limiting coolers we might get better CPU coolers for SFF.
This is brilliant but the question is cost of the CAAM2 modules because I doubt it will be cheap for say 16Gb I hope it is. But off topic I wish board manufacturers for the 24 pin connector etc would turn them in their sides to make the cable less sticky out. I know someone did it but it should be standard now
It would be cool if CAMM2 type D has the ports on the front and the back side of the motherboard instead of stacking.
Nice stuff, I hope everything can eventually go to one RAM standard with CAMM2, & we can do away with soldered RAM in laptops, & maybe even Android tablets, but having said that, I'm still good with running desktops, & laptops with DDR4 RAM for my needs, & can't see myself upgrading for at least a few years now having 64GB in my 2 desktops, & 24GB in my laptops for less intensive taks.
The issue the performance of CAMM2 allows for much higher speeds. So desktops will want it, especially given that CAMMs can benefit from bigger and more robust cooling options.
We'll see what the future holds.
Those studs likely for some type of memory cooling fan which will absolutely be needed with this configuration.
Can you imagine to have the same RAM for desktop and laptop? Mmmm not sure if that is remotely possible, but would be great.
The 2DPC problem with low memory capacity of SODIMM has been one of the most frustrating points to consider when building a new DDR5 computer.
Looking forward for when CAMM2 fully replaces SODIMM in the DIY space.
about time we update the physical ram interface, it's been the same for like 40 years ? To have it flat and secured with screws is so much better
camm2 makes sense to me, so I like it
Looks to me that if a memory increase is wanted, you have to discard the old one and insert a new larger capacity one.
Someone on ebay will want your 32GB module when you upgrade it to something else.
I mean, if you REALLY that concern with OCing your RAM, you don't want to stick any other than same exact 2 stick of DIMM anyway. So just combined it to single CAMM2 and everything is just as the same.
Beyond that, they still sells DIMM motherboard if you don't care about that shit
Like I said in another channel with sticks you can buy 2 sticks and then add 2 more later but with this thing you have to do the full commitment on getting the exact capacity of memory for the rest of the life of your PC. If you get one of these modules that is 16GB for example, you need to buy a whole thing with 32GB then try to sell the old one. On laptops this makes sense but on desktop...
SI will want this because reduce mechanical issues with memory dislodging and generating support costs.
Appreciate the explanation
Alleged upcoming new gen of high performance APUs are gonna benefit greatly by this new standard setting in now. can't wait to have a the ultimate sff build in the future
clus will too msi claimed about 30% lower latency which is huge
Except with APUs having it on package like Lunar Lake is doing will be far more beneficial. This is my concern with Strix Halo. 40CUs will be choked hard in random games if they have to rely on system memory, regardless if it's DDR5 or CAMM2DDR5. No matter how close it's physically impossible to be closer than surrounding the die as GPUs currently do it.
@@zodwraith5745 Gpus dont use on package memory unless its hbm genius
@@zodwraith5745 Current camm2 capability is around 8500mt/s. micron on next year already planning on releasing 9600 mt/s camm2. even more with years to come. so unless integrated ram is blowing through 10k mt/s, i don't see why not. there's certainly demand for it
@@zodwraith5745 oh and it also has 256bit bus, so no worry of choking there
I really want to buy a AMD AM5 Motherboard with CAMM2. I'm pretty excited for it! Hopefully prices will be good when it comes out.
I wonder if they would try to enforce camm2 format for global release of ddr6, new platform, new CPU, might as well try to make the jump then ?
If it can give me 8000+ mt/s with decently tight timings and actually be stable thanks to the shorter traces then I’m all about it
The 4 studs are probably made for a small fan or heatsink. Eventually they will have their own little cooler/heatsink combo right next to the cpu cooler lol. Probably will look pretty cool i bet.
I don't see how this will work well in servers, with 8/10/12 memory channels per cpu. Even with "quad" modules, that's still 2-4 large flat modules that need to be squeezed into a very limited area.
With the modules we've seen, yeah, doesn't seem space-efficient. But maybe the plan is to use the shorter card with its better connections and put something more dense than regular DDR5 packages on it.
"Battery life is not as important on gaming laptops"
you take that back! Lots of people use gaming laptops on battery. And since in them the components are generally more power-hungry than in ultrabooks, it's even more important to squeeze every bit of efficiency as possible from all components, including RAM.
However, RAM power usage is severely overestimated. Each So-dimm module in my gaming laptop consumes just 0.13 watts at idle and during basic tasks like browsing or watching movies, so in total 1\4 a watt combined. Compare that to the CPU which consumes from 2.5 to 5 watts in the same tasks or display which consumes from 2 to 10 watts depending on brightness setting. There are also ports, wifi module and so on, so in total, the percentage of power consumed by RAM is really small. Maybe 5%, and that's in gaming laptops, probably even less for LPDDR5.
What's up with the Comic Sans in the JEDEC doc?
That's me. JEDEC requires the copyright statement but I thought I'd spice it up with some comic sans. You can also find papyrus too. ua-cam.com/video/jVhlJNJopOQ/v-deo.html
I hope we get DDR5/6 CAMM2 modules for the DIY space sooner than later.
This would be my first Msi board to get ! With the new x870
DDR5 gets a bit warm. What do the thermals look like? Are we going to see CAMM2 tower coolers?
I have 64GB RAM in my system with two empty slots for 2 more 32GB modules for a total of 128GB, so way more than I have ever used. This CAMM2 sounds good, but they may want to put two sockets on MoBo's, to make upgrading easier, so you don't have to replace all of it at once, and a a high cost.
Thank you for the insight. info on this new ramware helping when deciding to buy more faster/expensive dimm ddr5 now and use the sticks in futurer z890, or wait for this camm tech to mature, or just upgrade to dimm ddr6 in future
I would imagine if it kicks off the prices would naturally come dowm to be more competitive
In these photos the CAMM2 seems to take a lot of space in terms of width compared to a tall narrow DIMM. My concern is whether this will fit well on a Mini ITX board. Which don't have much width between CPU socket and edge of the board.
I can't believe dell would allow others to use their patent
Looking forward for 128GB CAMM2 modules.
most exciting thing yet from computex.
We will see CAMM sockets directly on CPUs once glass packages become the standard.
Two of those might fit in an OAM(OCP accelerator module) under the heat spreader between the screws, along with the CPU's VRM parts...
it helps so much with trace layout and tuning timing / power ,,, dell did good.
I'm certainly hoping CAMM2 will be available on some MATX or ITX boards off the bat. Though MSi's MPower motherboard is interesting too. Hoping that may get carried over to Z890. Kinda hard to imagine why they'd debut it for Z790 when Intel 15th is around the corner. 🤔
I'm just getting anxoius to do a new build. The 9700k and 2080 in my Alienware Area 51m laptop has carried me for this long, but I'm ready for an upgrade, and I'm going back to a desktop system, albeit a MATX system.
I’m all for this if they can up the memory bandwidth. We desperately need this for ai.
HI GORDON!!!
ddr5 is difficult to seat already so yea probably time to switch.
So I assume by this that memory makers assume they will have no problem increasing speeds but have a real problem increasing the number of channels for bandwidth?
I'd just like to interject for a moment. What you're referring to as CAMM2, is in fact, DDR5/CAMM2, or as I've recently taken to calling it, DDR5 plus CAMM2
how will manufacturers put heatsink on the underside?
Camm makes much more sense for signal path routing, but I see many many optimisations to come, firstly it’s all up against one side where by all rights the pins could be right under each chips legs and distributed evenly.
Secondly it could be on the back side of the motherboard and almost mapped directly into the cpu pin for pin with mm’s of trace
I guess I'm safe to build a workstation with normal RAM modules 2025
Actually te studs around the module is needed to click in cooling on top of the module. without the cooling, they will burn up. Sure, this cooling probably will got some rgb goodies on it. But main reason is cooling mounting
I am kind of surprised that they don't connect this to the back of the motherboard directly opposite the CPU. True cases would need to be re-designed for access but would make the module much closer to the CPU
Lets get BuildZoid a CAMM2 setup ASAP 😁
Gordon in 2022: "This new memory standard is controversial for being proprietary actually"
CAMM: (literally the coolest thing ever)
Well it's not proprietary anymore now that it's a JEDEC standard.
@@LongliusYep, & I'm not gonna crap on this one, even if it's gonna be costly at first, as any standard is better than soldered RAM that so many systems are using now, or only having 1 slot for a RAM upgrade with a low amount like 8 GB soldered on the board(thanks Lenovo 🤦) meaning you end up with weird configs like 24 GB, or 40 GB when the board, and processor in theory could take 64 GB, or even 128 GB.
CAMM 1 was proprietary for DELL. It had a lot of problems. Lenovo, Micron and other OEMs took over and overhaul it into CAMM 2 which is open. Now the entire CAMM connector is replaceable. It uses the same bendable pins like the CPU sockets but if you bend it -> just replace it.
This has to be introduced for CPU sockets as well.
@@mowtow90 I hadn't even considered modular CPU sockets. I spent hours bending back LGA 1700 pins on my current mobo, because I couldn't prove that it wasn't me that bent them. Good ole' Newegg return policy.
@@Longliusthe "Literally the Coolest Thing Ever" meme didn't exist back when CAMM was standardised. Is the Internet incapable of humour or something?
what is SI channel?
Signal Integrity.
You stab the memory modules in? How do you do that without breaking the mobo?
What
CAMM2 DDR5 will probably do not be to popular DDR6 version I can be interested .
I thought MSI Z890 that was presented this week, will be on the market soon.
the density of de each ram chip, will define more cappacity in the future.
Do we know how much restate CAMM2 is going to need in non-Z dimensions? I can see that the single MSI motherboard example doesn't seem to have the big CAMM2 installed or denoted as to which size CAMM2 was installed in the picture.
It looks like most of the boards support the larger CXXX dual channel modules from what I have seek from afar.
I don't understand why some people are excited about this. It doesn't seem to offer anything better than what we already have outside of being low profile while being substantially more expensive. Seems more like it's focused on higher profit margins for manufacturers.
Do you know the trace width on the board? I used to make boards.
Gordo! Hope you're kicking it dude. I mean those slacker kids' butts, LoL. Come on we, old guys need to use our walkers together! Hey mine's got wheels!
Good call for Dell, they can live off this patent for the next decade, and fund more innovation, fingers crossed for 32x PCIE.
they can do trace layout to ram chip itself until its perfect
Thanks.