Monday, November 22nd 2021

Samsung Talks DDR6-12800, GDDR7 Development, and HBM3 Volume Production

During Samsung's Tech Day 2021, the company presented some interesting insights about the future of system memory technologies and how it plans to execute its production. Starting with the latest DDR5 standard, the company intends to follow JEDEC documents and offer some overclocking modules that surpass the specification advised by JEDEC. While the DDR5 standard specifies memory modules with 6,400 MT/s, Samsung will develop modules capable of overclocking up to 8,400 MT/s. These are not yet confirmed as they are still in the development phase. However, we can expect to see them in the later life of DDR5 memory.

The company also talked about the DDR6 standard, which is supposedly twice as fast as DDR5. The new DDR6 standard is still in early development, and all we know so far is that the number of memory channels per module is seeing a twofold increase over DDR5 to four channels. The number of memory banks also increases to 64. In addition to DDR6 for desktop and server use cases, the company is also working on Low Power DDR6 (LPDDR6) for mobile applications. While the company's LPDDR5 memory goes into volume production using the 1a-nm process at the beginning of 2022, the LPDDR6 is still in early development. The base speed for DDR6 modules will allegedly arrive at 12,800 MT/s, while overclocking modules will join the party at up to 17,000 MT/s. Mobile-oriented LPDDR6 version is also supposed to come with up to 17,000 MT/s speeds.
Next up, Samsung talked about its memory offerings for graphics, where GDDR and HBM come into play. The new GDDR standard that is supposed to arrive is GDDR6+, which bumps the speed from 18,000 MT/s to 24,000 MT/s. Node of choice for GDDR6+ will be 1z nm, and the Korean giant wants to start manufacturing these modules this month. After this, the company's roadmaps show that the GDDR7 standard will replace GDDR6+ and offer rates of 32,000 MT/s. With GDDR7, there will also be a new feature present called "real-time error protection feature," which is still unknown. Presumably, it is some form of ECC for GDDR or something similar. We are yet to see.

In addition, the company also mentioned that its HBM3 memory will be ready for mass production in the second quarter of 2022, with speeds of 800 GB/s. This memory will target mainly AI applications, and Samsung is working with partners to equip new solutions with HBM3.
Source: ComputerBase.de
Add your own comment

56 Comments on Samsung Talks DDR6-12800, GDDR7 Development, and HBM3 Volume Production

#26
Punkenjoy
Don't worry, we will have few years of "Oh god this cas latency suck" and after that maybe a year of "it's not worth upgrading to DDR6 over DDR5" before it become mainstream.
Posted on Reply
#27
Wirko
PunkenjoyBack to the news I find it interesting that DDR6 will have 4 channel per dimm instead of 2. So a 2 Channel/DIMM system will have effectively 8 channel. If you pair that with a 16 core cpu, you end up with 2 core per memory channel versus 4 with DDR5 and 8 with DDR4. We will see how it go but that will bring back the memory channel in line versus what we had 2 decade ago where a single core CPU had 2 memory channel for itself.
Don't forget that these channels are half as wide, and those in DDR6 will be one fourth as wide. More channels will potentially bring a small advantage in random access, and that will be more likely in servers, where (and if) applications have long queues for memory access. More channels also need more complex IMCs (but at least the memory modules don't seem to be more complex because of that).
ARFYeah, 2017 for DDR5, and 2009 for DDR4?! Right?
Wikipedia even states (but I can't find other sources) that "JEDEC began working on a successor to DDR3 around 2005, about 2 years before the launch of DDR3 in 2007".
Posted on Reply
#28
bug
WirkoWikipedia even states (but I can't find other sources) that "JEDEC began working on a successor to DDR3 around 2005, about 2 years before the launch of DDR3 in 2007".
It would seem you're not wrong: tok.fandom.com/wiki/DDR4_SDRAM
Posted on Reply
#29
Punkenjoy
WirkoDon't forget that these channels are half as wide, and those in DDR6 will be one fourth as wide. More channels will potentially bring a small advantage in random access, and that will be more likely in servers, where (and if) applications have long queues for memory access. More channels also need more complex IMCs (but at least the memory modules don't seem to be more complex because of that).
Indeed
but if you compare a DDR4-3200 and a DDR5 6400, the same 64 bit word should arrive in the same timeframe on both memory. The DDR5 will send it 2x32 bit in the same timeframe the DDR4 will send the 1x64 bit. This will be the same for DDR6-12800 vs DDR5-6400.

In the end, on a single channel, the bandwidth doesn't change much, but you double the number of channel each generation.
Posted on Reply
#30
chrcoluk
FreedomEclipseDDR5 isnt even super mainstream yet and they are already talking about DDR6... Sheesh. Might aswell keep saving my dollarydoos and wait for the DDR6 platform since DDR5 seems dead in the water so soon.
Its the world of tech, hopefully it never gets as bad as smartphones where the latest model is only the latest for barely a few months now.

Sadly I do think we going to see accelerated transitions between DDR generations. Like how we do now on CPU generations, the companies have realised it makes more money.
Posted on Reply
#31
ARF
chrcolukIts the world of tech, hopefully it never gets as bad as smartphones where the latest model is only the latest for barely a few months now.

Sadly I do think we going to see accelerated transitions between DDR generations. Like how we do now on CPU generations, the companies have realised it makes more money.
Companies don't know that the customers can avoid the new "shiny" products because their old are already good enough.

Just stay with Windows 7 or Windows 10 with the hardware that you currently have, and forget about the upgrades.

The smartphones sell faster because many people break them quite often and easily.
Posted on Reply
#32
Richards
Gddr6+ for rdna 4 plus infinity cache will bring orders of magnitude of bandwidth
Posted on Reply
#33
ARF
RichardsGddr6+ for rdna 4 plus infinity cache will bring orders of magnitude of bandwidth
RDNA 4?! There isn't even RDNA 3 anytime soon, which means RDNA 4 is a thing from the very very distant future. Maybe its development or brainstorming of ideas has just started, 4-5 years before its future launch in 2025-2026...
Posted on Reply
#34
noel_fs
why are they talking ddr6 when ddr5 motherboards just came out
Posted on Reply
#35
R0H1T
RichardsGddr6+ for rdna 4 plus infinity cache will bring orders of magnitude of bandwidth
You obviously forgot HBM3 :rolleyes:

I have internal sources confirming they'll release something that'll shock Intel/Nvidia/Apple heck their own shareholders, you just wait!
ARFThere isn't even RDNA 3 anytime soon, which means RDNA 4 is a thing from the very very distant future.
Not true, they now have enough money to develop multiple GPU designs simultaneously like they did for Zen.
Posted on Reply
#36
ARF
noel_fswhy are they talking ddr6 when ddr5 motherboards just came out
It really doesn't make any sense, TBH.
R0H1TNot true, they now have enough money to develop multiple GPU designs simultaneously like they did for Zen.
They can have the money, but the progress doesn't or hasn't worked like that. They plan and follow product release cycles. They, for sure, won't launch RDNA 3 and RDNA 4 alongside each other.
More likely 2.5-3 years apart them.
Posted on Reply
#37
R0H1T
Of course they didn't release zen3 & zen4 side by side did they, even though work on them progressed simultaneously? Obviously zen3 had more work done on it by the time zen4 first underwent development, similarly zen5 wrt zen4 developing alongside it.
Posted on Reply
#38
Fouquin
FreedomEclipseDDR5 isnt even super mainstream yet and they are already talking about DDR6... Sheesh. Might aswell keep saving my dollarydoos and wait for the DDR6 platform since DDR5 seems dead in the water so soon.
As has always been the case. Whatever is out now, expect next gen has been in the works already for a year or more. That's how we keep up on improvements, not by taking a vacation when the new standard is published but by working on the next standard right away.
Posted on Reply
#39
Richards
R0H1TYou obviously forgot HBM3 :rolleyes:

I have internal sources confirming they'll release something that'll shock Intel/Nvidia/Apple heck their own shareholders, you just wait!

Not true, they now have enough money to develop multiple GPU designs simultaneously like they did for Zen.
Forreall ?.. but i do believe amd will take the performance crown from nvidia.. nvidia can't beat a mcm gpu with a monolith one
Posted on Reply
#40
Vayra86
TheoneandonlyMrKWhat, no one told me should I stop building rigs for people?!.
PC DIY has died about as often as PC Gaming. And every time it comes out as a more profitable business. Strange how that works.

We're persistent m*fkers.
Posted on Reply
#41
bug
Vayra86PC DIY has died about as often as PC Gaming. And every time it comes out as a more profitable business. Strange how that works.

We're persistent m*fkers.
Still, with the money I could have built a decent rig 5-7 years ago, today I can buy a decent video card :(
Posted on Reply
#42
Vayra86
bugStill, with the money I could have built a decent rig 5-7 years ago, today I can buy a decent video card :(
Yep... just holding on to what I have right now :P
Sign of the times, let's face it. Everything's under pressure. I don't think this is a PC thing.
Posted on Reply
#43
bug
Vayra86Yep... just holding on to what I have right now :p
Sign of the times, let's face it. Everything's under pressure. I don't think this is a PC thing.
Somehow it feels like PC has been at the forefront. GPUs started to become untouchable since Turing, over 3 years ago.
Posted on Reply
#44
Bomby569
This crazy race for new standards makes no sense when the market can't supply any of them. And it will only drive prices even higher, something i suppose it's what they all want, cash in just like they are doing with gpu's.
Posted on Reply
#45
bug
Bomby569This crazy race for new standards makes no sense when the market can't supply any of them. And it will only drive prices even higher, something i suppose it's what they all want, cash in just like they are doing with gpu's.
There's no crazy race, this is how the industry works. When you buy Zen3, someone is finishing up Zen4 and somebody else is already looking at Zen5.
In case of standards, they take years to get right and qualify. Not to mention you need to correlate them with what you can actually build.

The only crazy thing here is people thinking only what they see on the store shelves matters.
Posted on Reply
#46
Bomby569
bugThere's no crazy race, this is how the industry works. When you buy Zen3, someone is finishing up Zen4 and somebody else is already looking at Zen5.
In case of standards, they take years to get right and qualify. Not to mention you need to correlate them with what you can actually build.

The only crazy thing here is people thinking only what they see on the store shelves matters.
all you have to do is look up how long the PCIe or DDR generations lasted to see times are shortening fast
Posted on Reply
#47
bug
Bomby569all you have to do is look up how long the PCIe or DDR generations lasted to see times are shortening fast
And that is a problem how?
Posted on Reply
#48
windwhirl
bugWhen you buy Zen3, someone is finishing up Zen4 and somebody else is already looking at Zen5.
... They're already working on Zen 8, btw
Posted on Reply
#49
Vayra86
Bomby569all you have to do is look up how long the PCIe or DDR generations lasted to see times are shortening fast
This is FOMO talking, rather than common sense, I think.

If you have PCIe 2.0 you can still make do on an x16 lane with current day GPUs, go figure. There is no need to yell 'first' every time something gets released, but somehow, in our infinite consumerism, people have gotten that idea.

Let the race be the race, and upgrade when there is an actual NEED instead of just an abstract WANT. Cheaper, more sensible, and you won't feel like you're not keeping up. There's nothing to keep up with, here, except your own use case.

For example I see people upgrade CPUs every other gen... for gaming... and they gain what. 5%? Best case? Similarly, upgrading GPU every gen to gain 20-30%... totally nuts. To then follow up with 'its going too fast' is pretty twisted.
Posted on Reply
#50
TranceHead
windwhirlDoes it matter? Memory has gotten faster and faster, never slower.
DDR4-4200 is faster than DDR5-4200.
Don't be too hasty saying 'never'. In TPU's own tests, DDR5 kit tested was only 3.2% faster than DDR4-3200.
Early adopters get screwed in price and speed.
Posted on Reply
Add your own comment