INTEL WONNED
>>109638431May I see it benchmarked?
>>109638431Really? You mean I can buy it right now for under $500 ready to go ?
>>109638431>chipletsIntel eating AMDs homework.
>>109638431>256 threads >shipping 2027H2uhhh AMD said they're shipping 512 thread CPUs next quarter?
>>109638431>designed for AIDid every tech company forget that computers are also used for things other than AI or something?
>>109638819You have to see it from their perspective. Their vision is an utopia where you buy "intelligence" from them. They won't, and they can't, let this opportunity slip.
>>109638583AMD is still 256 cores.
>>109638858Yes and how many servers using 256 core CPUs aren't throughput focused
>>109638431>intel>256 coresYou'd probably need a small nuclear reactor to power up this thing
>>109638870SMT only comes into play for when whatever you are doing doesn't involve lots of math.HPC and AI are both memory and to some extent fabric bound. This is likely what Intel is targeting as it's kinda what saved them after companies realized that the Infinity fabric is shit.
>>109638931no, silly there's a reason Vera has SMT
>>109638431>AVX10let me guess, AVX20 next?
>>109638552It's particularly funny when their whole marketing strategy a while back was shitting on chiplet designs.
>>109638431>agentic AIthis is so retarded lmao>>109638856>They won't, and they can't, let this opportunity slip.especially since amd still has this loser mentality where the copy whatever intel and nvidia are doing for a bit less dollar, bunch of fuking retards, amd with competent leadership would be a 2 trillions dollar company.
>>109638971It's weird to see someone see Intel directly copying EPYC and say AMD has the loser mentality. When is Nvidia's unified memory laptop shipping btw?
>>109638971>>agentic AI>this is so retarded lmaoImho it would be cool if it actually worked but the problem with it is it degrades after some time ("meltdown") which defeats its entire purpose.
>>109638945and Intel's next server coreARM will be the only odd man out in the server world without SMT
>>109638431>That'll be $16,500 please and thank you
>>109638945Vera has a different kind of SMTIt can essentially split a core into 2 isolated paths versus literal time sharing of execution units. This basically helps guarantee a thread can execute its work immediately instead of possibly being halted waiting for its time.It can also adjust bandwidth allocation at will to give threads fully unfettered access to each core.
>>109639056And Zen has statically partition parts too (like the entire front end)DMR not having SMT is not a good thing it put Intel at a competitive disadvantage for no good reason. they thought it would let them fit more cores in a socket but they end up with the same core count as AMD
>>109638552>>109638968Bahahah.
Glad my AMD 9000 cpu is free of the ai nonsense, probably the last gen to be free of it
>>109638819We are not the target demographics anymore. They want to sell to AI companies because they believe personal computing is dead.
>Next Gen Agentic AIGotta grift those investors with meaningless buzzwords while you can eh.
>>109638819You could run 6 gorillion Docker containers. lmao
>>109638819>>109639254>>109639266besides the marketing bullshit, what part do you think is actually ai designed?
>>109639287I guess all the dedicated 3D matrix math processors are what they call "AI cores" for "AI", but if SHTF they could reuse them to do regular calculations instead so that the hardware doesn't get devalued over another tech revolution downsizing after the hype bubble pops.
>>109639275It's not meaningless"next gen" = not here yet"agentic" = it can act on its own"AI" = All In = I'm feeling lucky
>>109638431I don't get it, my computer doesn't use the cpu at all when using ai, so what is the usecase?
>>109639547your bot is going to make 20 billion git commits per second by 2030 and you'll need more CPU for that
>>109638431doesn't matter, i am not buying a chip from a company owned by the US government
>>109638476Don't be ridiculous.
>>109638431>1.6TB/sSo is this better than current GPUs for AI?
>>109638431why 256 cores?I don't really see how this is useful for AI workloads, local or datacenter.
>>109640320dont tell me about my software needs
>>109638819Not for long.
>>109638968lmao no way is that real
>>109638583256 x ~3.3GHz be ticking near 845GHz how far is tserenkow glow?
>>109640728>we now have newfags so new that they don't remember the early Ryzen dayshttps://www.techpowerup.com/235092/intel-says-amd-epyc-processors-glued-together-in-official-slide-deck
>>109638431diddit dooded?
>>109639571>Intel still sucks oxidative assOk
>>109640817Sad, isn't it>>109640728You have to be 18 to post here
