CUDIMM Explained! Clock Driver RAM - DRAM IS CHANGING!

[@DevilbyMoonlight]

So basically the chip is like a 'repeater' in the old networking world where it corrects and retimes the signal at wire speed with 0 delay.. I hope it wont be a new point of failure.

[@ThatGuyPal88]

It definitely has “delay”, probably one of the reasons CAS latency is so much higher(other than the lower voltage)!

[@atomicskull6405]

It's basically a Schmitt Trigger, this is actually a common solution to degraded or noisy digital signals.

[@LeLe-pm2pr]

@@ThatGuyPal88 crucial has a reputation of making incredibly shitty ddr5 in terms of latency, so it doesn't seem that surprising that the latency numbers are that high, especially at 1.1v

[@Great_America]

No, there is measurable latencies. There is no such thing as zero latency in memory.

[@LeLe-pm2pr]

@@Great_America if it's done properly, you can absolutely redrive a clock signal with 0 latency

[@adriankoch964]

"Cu" means "asshole" in Portuguese, so CUDIMMs will bring a smile to ~300M people.

[@BBWahoo]

Brazillian Busssssy 🤤

[@specy_]

I can imagine how hilarious chemistry lessons are there

[@youkofoxy]

@@specy_ We do not have the Uranus joke. because is Urano here.

[@MeepMeep88]

"Cu" also means dck in Viet LOL..

[@DoTrO_NoL]

@@youkofoxy But we do have the Copper joke, which means CU.

[@AshtonCoolman]

CUDIMMs will be great for AM5 when support is added. I can't wait to see the results of 9000MT/s and faster at a 1:1 ratio.

[@danijelujcic8644]

I hope 8700G goes through benchmarks one more time.

[@WillFuI]

Do u mean 2:1? The memory controller is limiting 1:1 speeds. The apu’s are different so maybe them

[@profounddamas]

I'm afraid you will have to wait and wait and wait and....

[@ShaunaJagan]

Praying for Zen 6 have this by default

[@nerdynumen]

Your comment only applies to APUs, it doesn't matter if we connect the ram directly to a 7800X3D, or a 9950x, the IOD only going to 2200 is a massive bottleneck for the cores, and for the ram.

[@Catalinddm]

Cheers, Andy. Straight to the point. CUDIMM is quite a mouthfull...

[@stepbruv8780]

CUMM for shorter

[@Kumoiwa]

Clear and to the point, thank you!!!

[@PropaneWP]

That deserves a subscription

[@blackbirdpctech]

Great explanation, nice video … really like your channel

[@benjadot]

Well explained.

[@hasnihossainsami8375]

That's an interesting solution to the current DDR5 instability problems at high speeds. One of the purposes of CAMM2 design was to reduce the distance between CPU and memory, thereby alleviating the signal smoothening to some extent and allowing higher clocks. I guess we still need to see which solution is more effective, but if it turns out CUDIMMs are at least just as good as CAMM2 then I don't think the latter will make it into mainstream desktop any time soon. Mobile devices might still adopt CAMM2 more quickly because of the space-saving vs SODIMM.

[@georgejones5019]

We could see CAMM2 with a driver chip placed on it. The best of both worlds. Lower latency because the distance shorter, lower power, and better speeds.

[@notantogonist017]

CUDIMM + CAMM2 = CUCAMM2

[@atomicskull6405]

It's not original in any way it is *the* most common solution for noisy signals, just stick a Schmitt Trigger in there to clean it up. Yet they're acting like this is some innovation. It is not it's just what you do to clean up a weak or degraded digital signal.

[@rallyscoot]

CAMM is really painfull on upgradabillity on desktop systems.. Dimms are easilly replaced in smaller chunks. But for mobile and smaller devices where upgradabillity isnt prio camm should be better to use.

[@KevinAroni-uz5rl]

@notantogonist017 niente male come risposta, ma é CUM 2

[@atomicskull6405]

So they basically put a Schmitt Trigger on the memory module between the incoming clock signal and the memory controller, and they're going to price gouge for it. I can guarantee the price increase won't be proportional with the increase in cost of manufacture. This is very old technology that was invented in 1934 by Otto Schmitt (dude came up with it as a graduate student) that they have been using to clean up noisy or weak digital signals since forever.

[@fyfaenihelvete]

I dont think price has been proportional with the cost of manufacture ever when it comes to components and tech in general.

[@amurtigress_mobile365]

Basically they reduce the number of digital inputs and thus loads from 8, 9, etc. to one, seen from the POV of the memory controller in the CPU. I wasn't expecting it to be that simple overall, that the clock signal alone is the problem. I was kinda expecting that additional signals need to be driven too.

[@rajeevmoghe5846]

a DIMM using a RCD is just a RDIMM

[@ading4510]

Thankyou for clear explanation, this is great, i always hope pc industry had pay more atention to RAM compatibility since problem with ram is always pain the butt to solve with those much variable

[@cj09beira]

great video, straight to the point, you got yourself a sub.

[@TechGaming-d4k]

Great explanation! Cheers.

[@AdmV0rl0n]

This was a good walk through. Thanks!

[@tymon5349]

good info but it would be alot more clearer if there was more on the screen supporting the story

[@ianjameslake]

3:24 Yeah, but they are CL52 (so very loose timings) at that 6400 speed, for that 1.1 voltage. Voltage is required for not only clock rate, but also tight timings. It's why you rarely see tight timing ram at low voltages.

[@vladislavkaras491]

Thanks for the short and informative video!

[@ScottGrammer]

Informative video with no BS. I like it!

[@jlrockafella]

I don't believe it will do much for gaming but might for other workloads

[@rajeevmoghe5846]

Yup the CUDIMM seems to be very similar to an RDIMM used for server workloads

[@SparklySpencer]

3:11 oh, okay, a signal amplifier... aka a repeater, fascinating

[@D3a7Hm1N3]

Do you need a new MB?

[@slaydog5102]

Maybe a bios update

[@darkman237]

So here's a series of questions: I'm planning a new build utilizing a Z890 MB and an Intel chip, but memory has changed. The board can take 9000MT DDR5, but I can't find any sticks for sale with that. I've found several 8000 and more 7200, but which is better? Does latency mean anything any more? These sticks seem to have high timing/latency values? Also the 8000 come in an unusual configuration 48GB in pairs rather than 16 or 32. Why is that? I hope you can answer my questions, so far I haven't received an answer from anyone else.

[@shazzz_land]

Cool idea, they implemented a signal repeater of sorts

[@Terry-cw9ty]

It's weird how we've only seen 1 or 2 camm2 boards so far, is there any idea when those will go retail?

[@xXx_Oshino_xXx]

I'm certain it's more common on laptops. Camm 2 has no real packaging advantages on desktop as desktops (even itx) dont benefit from flatter ram like a laptop does.

[@georgejones5019]

​@@xXx_Oshino_xXx An advantage would be CPU cooler compatability. CAMM2 also shortens the distance for the signal/data to travel. There are some advantages over traditional RAM sticks.

[@darkbit1001]

So, its like a buffer/amplifier, but not a buffer/amplifier?

[@angelg3986]

interesting for an additional $1 chip on the PCB, how much more money will they demand

[@zwipify]

Thanks for the informative video, much appreciated. I'll C U Jimmeh

[@jamespeterson7125]

You breifly mentioned that CAMM2 memory doesn't yet have clock chips yet, but I'd be more curious about how necessary they are. As far as as I can remember, better signaling was one thing they advertised with CAMM2.

[@RaminOhebshalom]

Thank you

[@rajeevmoghe5846]

The RCD(register clock driver) is made by Montage a chinese company. I did have a question how is a CUDIMM different than an RDIMM which were already being used for server applications

[@bumpahhi]

3:25 it seems like other CUDIMM memory has much more loosened timings when comparing these two

[@yamilmontes1365]

Micron makes horrible DDR5 chips, especially timings.....but hynix will be way better

[@johnwatson8004]

Very cool

[@pkeak1933]

Can ryzen 7950 run cudimm 🥺 ?

[@AlyxSharkBite-2000]

Do you think with CUDIMMs will let you run 4 DIMMs at the 6400MT/?

[@trukr817]

Will this work out better than the last time intel led with a new type of RAM, Rambus?

[@tanmayhembram9249]

Thanks.

[@Phil-D83]

Needs ecc as well

[@fedoresko]

Not clear what exactly causes compatibility issues. I see no clear explanation why AND should not be compatible, while Intel is. What particular CPU\chipset feature makes it work?

[@DragoMorke]

I was wondering if the current intel 200s was interesting to me but maybe I wait another generation (alder lake now) until lpcamm2 modules and boards are out with this tek. The lower power needs are interesting but this reminds me when ddr5 was just out and expensive and sold out.

[@D.u.d.e.r]

Cool DIMM enhancement, however with it I would expect current CPUs to support top DDR5 speeds of 8800MT and not just 6400. It’s ridiculous that LPDDR5X can run on higher MT speed than desktop DDR5 (CU)DIMM modules.

[@Buddy-D]

Noob here, any chance to see a production of CUDIMMs for DDR4?

[@aboomination897]

no

[@KevinAroni-uz5rl]

@aboominación897 eres bueno

[@micahthewise1]

Here from a ZachTechTurf short to learn about CUDIMM!

[@alienb1212]

Yay it's Rambus II.

[@influentialgurning]

Really soon. At least there was no gotten or going ahead.

[@profounddamas]

How many thousands of euros will cost 8 GB? Anyway I'm not convinced with those 52 timings. If it's so great why such horrible timings? Why is it when new memory arrives it always sucks?

[@kelownatechkid]

First frame of this video has incredible Giorgio A. Tsoukalos energy lmao. Thank you for making this overview, this is an important development

[@vincentsmith9180]

Doesn't that increase latency? I mean, the Clock Redriver Chip having to process the signal from the Memory Controller on the CPU?

[@1990JRW]

Right, it isn't exactly "free". Looking at other redrivers (USB, PCIe), the latency should be less than 1ns. So marginal effect, given the clock rate.

[@Demonoid1990]

I'm getting the gist that it acts more as an amplifier or like a UPC. Just something simple to boost and clean the signal. Some might use a UPC if they live in an area with shoddy electrical. It cleans up the signal from the wall, and gives you a nice smooth sine wave. I'm no engineer but I get the general idea of it's purpose. My dad would certainly know more than me being an electrician hehe. I'm just a trucker and a tech enthusiast/hobbyist.

[@handlemonium]

Restoring optimal frequency signal is probably worth it if the increased latency is only several nanoseconds.

[@fnorgen]

If it's just reproducing the clock signal then at worst the phase will be a bit off, which will have a negligible impact at these frequencies. The clock signal is really boring. It doesn't transmit any information. It's just a clock to set the pace and keep everything synchronized. From what I understand the control and data signals still take the direct path. It's just that feeding the memory chips a cleaner clock signal makes all the other signal processing way more consistent.

[@computersales]

Cudimm is exactly what I say when a stick of memory fails memtest.

[@cadcock15]

Cant wait until we have ryzen pushing well past 10000 whether its with cudimm or camm2. Memory performance is really holding back their architecture.

[@xlr555usa]

You will be waiting a long time

[@dan8t669]

Highly doubt this will result in high frequency + low lantency + low voltage DIMMs. Can only have 2 out of 3.

[@thomaslindell5448]

Wouldn’t this make more sense on the motherboard

[@danixsc]

Huh so that's why Kingston ram modules cost twice the G.Skill ones. I don't see that much advantage if you have to pay double the price.

[@TonyMasters-u2w]

We DON'T need obsolete DDR not scalable architecture we need HBM 🤗

[@dnakatomiuk]

Yet cost at the start will be huge at first and they will be slow

[@nick066hu]

And the real 'ingenuity' is not even being mentioned in any of the reviews of the CUDIMM. 'ingenuity' no.1: That client clock driver chip costs (for example, Part Number RG5C172B0C0GBX) $0.267 a piece (it is the actual price, I looked it up) and this additional cost of $0.267 / memory stick makes you the whole RAM package $200 more expensive. Brilliant, isn't it ? 'ingenuity' no.2: to exclude competition so that no manufacturer sells this 'improvement' at less than 500 times the costs, it is patented, so everybody must pay a hefty royalty to the 'inventor' of this scam. This trivial thing is patented despite the fact that it has been used for decades to reproduce the clock signal in a component.

[@piratesephiroth]

Don't be silly, these are early models made in limited quantities and the price includes the cost of research and development as well. It's gonna drop to normal RAM prices once it goes mainstream.

[@nick066hu]

​@@piratesephiroth I remember using 'clock conditioner' ICs since 2006 in the circuits and PCBs I have personally designed. (Texas Instruments Part No. LMK03000 Family) Just because they gave the solution a new, fancy name and deceived the patent office (as if it were a newly invented thing, it was lucky they didn't apply for the Nobel Prize) it didn't require much development and research.

[@piratesephiroth]

@@nick066hu Still the research has been done and even though this solution is simple and cheap, that price can include the cost other failed projects that lead to it.

[@nick066hu]

@@piratesephiroth They are perfectly entitled to use any excuse they like as a price driver, it is none of our business. I usually add a hefty contribution to my next BMW when I make invoices. What upset me is that, as I have witnessed hundreds of times in my professional life, when a company uses the Patent Office only to restrict competition and there is no invention at all.

[@WillFuI]

Comparing the Corsair kit to the crucial kit is a bit like comparing apples to oranges. The timings on the Corsair kit were a lot lower while the crucial kit they timings are super loose. I but if u loosened out the timings on the Corsair kit u could lower the voltage down to 1.1v or near

[@droberts7725]

Why can't someone make an 8086 processor with HBM? I wonder why Intel didn't add HBM to try to save their dying product line.

[@paulboyce8537]

CUDIMM up to 10000 and more near future. PCIe 5.0 with direct lane to CPU "REBAR on streoids". Selected motherboards with additional 8 pin power to the PCIe 5.0. Elephant in the room what is missing? Hint it is better than 5080.

[@MrAtomUniverse]

if it's not a SERVER technology, i dont think's gonna be groundbreaking

[@melheno]

In AM5 even if you use 9000mts the memory bandwidth is not improving so this CUDIMMs can only improve intel cpu.

[@edgari0kas]

1000x 500 where

[@starofeden9277]

the new intel motherboard allready build to take cudimms check Assrock

[@aliyuabba4575]

I thought camm2 is the future

[@jcgamer892]

Oh great, a video about condoms 😅

[@marioinacio9274]

x870 e variants will support cudimms msi asrock asus will suport them there is news discusing this

[@vulcan4d]

Sure I'll buy it if AMD will run at the higher speeds but for now nahhhhh

[@AlbiDartanan]

3:03 reminded me on : Ludicrous speed?! Sir, we've never gone that fast before! I don't know if the ship can take it! Dark Helmet: What's the matter, Colonel Sandurz? Chicken? Colonel Sandurz: (cracking) Prepare the ship... Prepare ship for ludicrous speed! Fasten all seat belts, seal all entrances and exits, close all shops in the mall! Cancel the three-ring circus! Secure all animals in the zoo--

[@MrChomiq]

To people at Crucial - please bring back Ballistix kits

[@Shuttterbugg]

Dude CL52...IDC WHAT THEY DO THATS INSANE

[@Kneedragon1962]

A tech nerd who speaks English, likes open Sauce, and explains things in a fairly logical way. Bonus! Subscription added. Hello from Brisbane Australia. One question ~ Does this change the 'sweet spot' speed for AM5? If you have the clock on the memory DIMM, can you turn up the Infinity Fabric speed by 50% ? So we can run 9,ooo Mega Transfers / sec (instead of 6ooo) and still have a 1:1 ratio? My understanding, was the slow part of the Infinity Fabric, was the connection out to system RAM. The stuff on the CPU die ~ that could be turned up by a factor of 8 or 16 or something.

[@jacquesmertens3369]

What a cheap trick to set a new "compatibility" standard and make us buy new RAM. I'm perfectly happy with my 6000MT DDR5. Can't see how a "middle man" chip would make things faster or more stable. I guess it will do neither of that. It just adds an extra step.

[@zlibz4582]

auto AI motherboard OC for cpu + RAM is desperately needed by the customers all around the world for a stable , quick , fast , reliable ,safe OC so that customers dont have to spend a week or two researching about OC info and trying out different bios settings and waste their time , hope motherboard manufacturers provide a solid AI OC ability in built the mid to ihgh range mobos also why the f are they not providing a ram cooling fan and a bracket beside the ram to cool these high end OC ram ?? dont they know that ram also needs active cooling when its OCd ??

[@Hullbreachdetected]

Yes, reliable AI based optimisation and OC is the future of reliable PC sales. The generic customer struggles to optimise a PC let alone OC. Hardware will be always an issue, but at least the software side of it should be a few click basic thing.

[@SaccoBelmonte]

RAM Speed is so overrated. Just a few single digit percent difference between 5400Mt/s and 8000Mt/s. Only after they bump the speeds/bandwidth considerably then it will make more sense.

[@bloeckmoep]

Honestly, nothing of that was a problem prior to socket 1366 or am2/2+/3. Why you ask? Simple answer, Cpu communicated with a dedicated RAM controller chip which was not in the Cpu thus was closer to the RAM itself and all of those controllers had the benefit to REDRIVE the signals and also opamp them when needed. Another benefit of external RAM controller was independent cooling. So now we have dimms with an opamp onboard, to amplify an already s** **y signal... sounds about right as a solution regarding the bad standards of the entire HF industry. Instead of again making independent on board RAM controllers with redriver and opamp features in the middle, nope do an opamp at the far receiving end. 😂😂😂

[@artosbear]

I want hardware that remains relevant for 30+ years, with software that runs On said 30-year-old hardware. I don't need faster processors. I don't need better graphics in games. I want shorter games with worse graphics made by people paid more to work less. I don't want AI. I don't want things made to enable AI. I want things to be more power efficient without worrying about being faster than ever. We don't need to make new hardware constantly. I want hardware made from hardware that was thrown away in the vast endless plains of electronics landfills and that the whole world should go and retrieve and clean up and restore those places. It's time for our nerdery to Stop being in service to giant corporations that only want our money, our energy, and our lives. They don't care about the future and it's time for us to actually start cultivating the future.

[@Jjanssoni]

6400 cl52 at 1.1v thats not great. I bet any m die or a die can do the same with 1.1v.

[@michaelmcconnell7302]

meh... im more than happy with my ddr4 4000

[@Great_America]

I won’t be buying CUDIMM - at least not at those prices. CUDIMM is more than four times the price of conventional memory. No real performance difference and just lower voltages. Pass.

[@masterinico]

From what I’ve read so far AMD 9000 series CPUs support it and MSI’s X870E boards will support it. Still digging though

[@adriankoch964]

CUDIMM and CAMM2 are probably the last spasms of user replaceable memory on the consumer side before memory becomes widely part of the SoC. Apple has proven with their M series chips that the bandwidth increase can yield some great performance and power efficiency gains and it's down to Intel and AMD to deal that blow to the rest of the industry. OEMs will be all for it, since this can allow them to force consumers in more reliable product lifecycles or charge a big premium for models with a long-lasting amount of memory.

[@rkan2]

AMD already doing it with HBM I guess

[@kevinerbs2778]

ze-ed is not a word also this RAM tech isn't new this about 30 to 40 years old technology

[@terpfen]

Downvoting because you padded what should have been a 2 minute explanation into a nearly 7 minute video by intentionally speaking at a snail's pace.

[@cliffordcooley1273]

Get a life! He was not speaking slowly. He was speaking clearly.

[@Sheerwinter]

Ddr4 is cheap now. Go get them budget gamers this video is not for u. 2x32gb is enough for u right? Quantity over speed is important. Even if it's 2x the performance of ddr4. That which is not the case here i guess. Even if it's 2x... It ll not be worthit. I mean just imagine having 16 gb ddr5 or 32 gb ddr4. Or 32gb ddr5 vs 64gb ddr4. What will u get? Ddr4 speed is enough for basic computing. And it's just enough for 99.99% am i wrong? And add ur extra dollars to ur graphic card to get more fps. Or invest in cpu for more processing speed. I'm super disappointed. 3060 12gb to 4060 8gb (wtf less vram?) and how much performance uplift u are getting? 10% to 15% at maxed. My 3060 i gave it to my relatives because the 4060 was on the market. I bought that sh*t. And I'm regretting it. I asked my relative to give me my 3060 and get the trash instead. But no... He said the 12 gb vram was important for him. 💀 F#@* u nvidia. (They are literally giving us sh*t in the name of innovation)

[@ekayn1606]

I can't wait for the next big jump in speeds! I want 10,000Mhz so I can put a Vegeta figure crushing his scouter in my PC🤓🤌

[@prashanthb6521]

Well explained.