What an incredible technical achievement!
One good thing about the AI boom is that it’s flooding capital into ram development and ram production. Once the bubble pops all these ram companies will have a glut of cheap high performance memory to sell to anyone who can buy it in cash. The question is if the bubble popping triggers a full blown recession will anyone have any cash to buy.
RAM demand ain’t going away anytime soon, unless major computer architecture changes happen.
There is always people with money to buy
“…and you definitely won’t be able to afford it”
I was watching a video on this and they mentioned the possibility of 128 and 256 gb consumers grade sticks and i almost laughed out loud at the absurdity of being able to afford them.
I can’t wait to buy my grandkid a 256GB stick for his 18th birthday, at which point he will thank me and his dad will have to sit him down and explain that i am old and don’t know that’s not even enough to play the upcoming GTA VII, reminding him i probably had to cash in a large chunk of my retirement fund just for that.
But GTA is no longer interesting or fun, not even the great peanut can make that game enjoyable for more two shakes of a BBQ lamb stick. And if we can’t fix government there won’t be a PC, all cloud and subscription.
Only $1,000,000 per gigabyte, think of the performance!
This makes it official, I’m skipping DDR5. I have ryzen 5700 and 32 GiB of DDR4, it’ll hold me over for a few more years. When I finally upgrade, it’s going to be at least 128 GiB.
also just go ahead and switch over to Linux , Steam has Linux figured out and at this point MS is state sponsored spy ware.
Same here; I’m running a Ryzen 3700X, with 64GB of nice DDR4. By the time I need to upgrade DDR6 should be well an truly established…maybe even at reasonable prices.
Right there with you but just rocking the Ryzen 3600X, hasn’t limited me in anything I care to do since 2019.
I’m probably skipping DDR5 as well at this rate. My Ryzen 2700 could use an upgrade, but I think I can survive with it until memory prices regain a little sanity. And I expect DDR6 will hit before that happens.
when will they go back to optimizing software to run in the available ram, rather than continuing to increase ram size?
When the devs run out of available RAM.
Or if you’re 100 Rabbits and you live on a boat powered by solar so you’re actively limited by available electricity so you write tools based on your home-made tiny VM.
Who are “they”?
The Stonecutters, obviously. https://youtu.be/fI4bRqjobbw
Who rigs every Oscar night?
who builds software?
Governments, corporations, military, startups, communities, hobbyists…
If you’re able to optimize AI tools to run larger models in less RAM, you’ll have a lucrative future ahead of you.
i would not use ai for anything. too many errors and falsehoods come from them. and theyre now out there hacking other systems because nobody reins them in
It’s KINDA being done in the ways we run MoE models. Strata makes more efficient use of limited VRAM if you have a decent amount of RAM, but you don’t even need to keep the entire model in RAM with it, it can keep the least frequently used experts on disk instead (said disk being an NVMe SSD hopefully, otherwise if you hit those experts, good luck)
I hear superyachts are pretty nice too.
I hope that one day I can afford a ticket to an exhibition where we all gather around a security-glass display case with a stick of one of those in it and bask in its presence.
DDR6 has no official launch date. Many believe 2029 is a realistic target for commercialization. However, in the current state of things, it would be a miracle if DDR6 had no setbacks.
One can only hope the bubble pops before 2029.
It will, it’s coming next summer/fall unless it comes earlier. If the AI industry continues on in its current state until 2029 without popping, then it won’t pop. It means they stabilized their debt and financing.
The bubble will pop, unless it doesn’t. Noted.
I firmly believe it will based on the financials. I am just saying if it’s 2029 and it “hasn’t happened yet” then it won’t happen. It’s coming within the next year, I would bet money on it. Not my life savings, but a chunk. In fact I have bet some money on it.
One can only hope the bubble pops before 2029.
and yet, you can only sell so many kidneys
Slight correction, you can only sell so many of your own kidneys. But there’s no limit otherwise!
oh, I wasn’t talking about mine !
In this economy ?
its the refrigerant cost that really get you.
Global warming and ai are both fucking honest to goodness ol’ kidney harvester’s on that one.
Some 16 billions would be the hard limit, this year.
Don’t forget about pig kidneys. Innovation!
Xenotransplantation for the win! So many kidneys 😁!
Maybe the AI bubble will burst before they can buy up the next decade’s worth of production capacity of DDR6 to manufacture scarcity and it will end up being cheaper than DDR5?
all hope isn’t lost
🙏🙏🙏🙏🙏😭😭😭😭😭😭
probably too late lol
8GB is only 2,762.89 before taxes. Act now, limited quantities.
I’ll buy 1TB, then have AI build a house for me to live in. It’s fool …proof.
Great how real RAM development starts the moment nobody but a handful of billionaires can afford it.
Ddr6 is nothing new tho. Look at your gpu . It had it for years at this point.
GDDR versioning doesn’t necessarily line up like that. For instance, GDDR7 is technologically akin to LPDDR5X, just with a different set of configuration tradeoffs.
At this point it doesn’t matter anymore if it were DDR99 with 578644.7 GT/s modules as we, consumers, won’t see them again in real life
Electron apps will still be slow.
Since limited waver space (because HBM takes so much of it at the same capacity) is one of the main driving factors for the Ram-apocalyse, this might actually help. And the performance looks actually great (the comparison to the DDR4 i am running shows some parameters being 10x better, this is actually a meaningful upgrade), meaning that some applications that currently use HBM might be able to just use DDR6 instead, causing a greater allocation of wavers towards that.
Wafer
Thanks, will change it in the future! (not gonna ninja edit that lol)
I generally edit my comments like this:
Since limited
waverwafer spaceSo that is fixed for the next reader but the comment chain still makes sense.
Generative ML inference isn’t so dramatically bandwidth limited anymore, either, so the “demand” for HBM isn’t as clear.
Eh the really big models want a lot of bandwidth still, otherwise you just don’t make use of all the compute power you have.















