V2Tommy
Member
Right? They're in the mirror practicing saying "I told ya so!" while doing nothing about it. Just like my poor friends still telling me I should've invested in Amazon 20 years ago.
Samsung raises DDR5 contract price by over 100% - "No stock"
- Thread starter adamsapple
- Start date
Right? They're in the mirror practicing saying "I told ya so!" while doing nothing about it. Just like my poor friends still telling me I should've invested in Amazon 20 years ago.
Disco Dave
Member
OmegaSupreme
advanced basic bitch
Did we ever get a fab facility in the states? I thought that was supposed to be a thing.
Memati187
Member
Why not get a third job and cancel sleep and freetime all together? Heck, Why don't cancel life while we are at it.
DaleinCalgary
Member
Yup Arizona, bad news is that it is Intel.
Lucie Goosey
Member
Amen. All the people whining and bitching like children who's parents just took their favorite toy away.
This was needed.
OLED_Gamer
Member
So glad I got my Pro and 9070xt day 1.
Can't wait to tell my grandkids about the times where waiting actually brought you better value.
Can't wait to tell my grandkids about the times where waiting actually brought you better value.
StereoVsn
Gold Member
Where do you think "Chinese RAM" is going ? Hint, China also needs data centers.
Last edited:
CentralScrutinizer
Member
We've hit a wall and the future is about efficiency and workarounds, not throwing more resources and power at it. It was gonna happen eventually but gaming is now getting forced into this situation.
MrAlwaysRight
Member
Fuck it bro's, I'm selling 16gb of my RAM, I won't need 32GB until about 2050 at this rate, so i can just run on 16gb, if i time it right, it will pay for my machine.
StereoVsn
Gold Member
CPU/GPU/etc - yes, TSMC. There is also Intel. Neither produce memory or nand.
Micron does but they just announced focus on enterprise customers (and really AI and hyperscalers).
Already happened. SSD (NAND) prices went up too.
DaleinCalgary
Member
Them "tools" make it inefficient and bloat how much ram we need. The lower level programming you do the less ram that is needed.
Ram usage is not getting smaller. Todays programmers don't understand effeciency
Last edited:
And we had people in this thread-
www.digitalfoundry.net
Telling us the companies would just eat the costs.
The AI Tech Crunch: Are We Looking At A "Dark Age" For Gaming Hardware?
The end of Crucial memory could be the start of years of pain.
www.digitalfoundry.net
Telling us the companies would just eat the costs.
Unknown Soldier
Member
Samsung refusing to sell memory to Samsung was peak hilarious side effect of all this
wccftech.com
Samsung Has Reportedly Rejected A DRAM Supply Request From Its Own Mobile Experience Division As It’s Focused On Maximizing Profitability
To maximize profitability, Samsung has had to make some hard choices, with one of them reportedly being rejecting a DRAM supply request from its own division
dgrdsv
Member
This isn't what they've announced. They've decided to bail from their own DIY business - shut down Crucial brand. That's all they've said.
I'm still quite certain that this is press induced panic more than anything substantial.
Those who like to talk about "AI bubble" should try looking at DRAM pricing changes and apply the same logic there.
Radical_3d
Member
Represent.
Represent(ative) of bad opinions
So is now still the best time to buy a pc?
Hohenheim
Member
48fb is fine for a long time. I Have 64 GB in both my builds, and have another 64gb in spare, in case a module stops working for some reason.
Should be more than enough until I upgrade to DDR6 in 5 years time.
Last edited:
Hudo
Member
Yes. And so the Chinese won't buy from Samsung et al.
DenchDeckard
Moderated wildly
It was Black Friday like I said. Game over now.
If you can still grab a last minute pre build deal id say do it.
Last edited:
Cyberpunkd
Member
It will burst because behind it there is no viable business model. It's the same as dot com boom where you just mentioned a website and gotten money. Everyone is putting money in AI because nobody wants to go against the common sentiment - also most of the people in charge are filthy rich and have pay packages structured ina way they won't lose that much if it doesn't pan out, and can potentially gain a lot. So most of the time top management has very little reason not to do it:
1 - you win? You get filthy rich
2 - you lose? Not your fault, everybody lost as well
People have difficulty understanding that in business you don't need to go for perfect, just the upside has to be larger than the downside.
Where are the truly transformative AI features ? I work with startups and see hundreds of SaaS platforms on a weekly basis - all AI in them is fancy LLM with a lot of if/then trigger conditions. Almost none of this is adding any value and creates a ton of information overload (you have no idea how productivity in tech was devastated by adoption of Slack).
Last edited:
Felessan
Member
LLM will never answer this unless explicitly told (found answer somewhere) as they don't physically read. For simplicity and speed of LLM this step is omitted and they go straight to meanings and garlic as a meaning have no R.
Honest answer would be "I don't understand/know" but LLM tailored to give answer even if it's a shitty one.
They think it's temporary uptick in demand and a part of normal super-cycle. Demand goes like sine function with upper than average and lower than average take turns and it's unwise to overreact to extremes.
For example Hynix only plan to put extra efforts if this part of super-cycle will last longer than expected and it'll be probably no earlier than 2027. Before that they are sticking with gradual expansion according to long term trends of the market.
Hudo
Member
I know. It's still funny, tho.
GHG
Member
Meanwhile you have some people saying LLM's have become the "Internet" for them and have replaced search.
We're cooked.
Mokus
Member
Lol. For that question ChatGPT needs more RAM.
Last edited:
apple_of_enlightenment
Member
It's up to Unity&Unreal, the ball is in their court.
Hudo
Member
Okay, I laughed. Well done, Sir.
P:axMagellanic
Member
China isn't going full retard with circular finance and pure desperation like the American firms are. Also if anyone knows how to flood a market it's them.
Radical_3d
Member
And buy a PS5pro and a house.I'm fine with 32gb as of now and made the wife buy a new minipc with 32gb for the workplace, sincerely thinking of it's worth to upgrade to a PS5pro now and resell the gaming pc next year
Bojji
Member
GPU - yes.
Felessan
Member
Don't give them an idea
It's actually is - you need 2stage LLM, one for recognizing letters as objects with meaning and another to convert letters into objects they represent, making whole LLM bigger and more hungry for memory
Toots
Member
Physiognomonics
Member
They can do whatever they want as long as they all do the same price increase. There is no stock because they don't want to increase production.
This is why we really need China (or others) to start producing more DDR5 to really compete against SK Hynix, Micron and Samsung.
This is why we really need China (or others) to start producing more DDR5 to really compete against SK Hynix, Micron and Samsung.
Last edited:
The thing about this test and test like those is that it uses loophole in how LLM's work. LLM's work on basis of tokens aka bits of words not single letters. So stuff like Strawberry might be divided into Stra - wbe - rry. Moreover LLM doesn't see tokens as words like "Stra" but a number so it would be something like 24522 - 232111 - 43332. The amount of tokens is preset and it is called vocabulary.
The point is that it's an issue with underlying architecture not with LLM intelligence itself.
Models with per letter tokens should solve that without an issue or just higher intelligence LLMs that became aware of their own underlying architecture.
The reason why we use tokens instead of letters is because tokens are much cheaper to run performance wise. For the same reason why for example japanese kanji characters can represent whole concepts in one sign rather than writing multiple words or sentences to describe it.
For sure we will have per letter LLM's. Doing so this now would trade too much performance and size.