Not in client.My speculation would be complete removal of L3 SRAM from main CCD die.
Venice-D is indeed N2, you can guess the rest.And moving the CCD of Zen 6 to the most advanced note (N3P or N2) to harness all of the density improvements.
mainboard maker — — — — — — — — — > distributor — — — — — — — — — > retailAnd do you have any idea why its so hard to find a Turin motherboard ?
What is RDL?I think hybrid bonding will only be used for X3D and MI400, so cache wil stay on the ccd.
The second half is RDL, probably.
And do you have any idea why its so hard to find a Turin motherboard ?
Whoa, that would be a bloody massacre. Dense cores with the L3 of the P cores while not wasting expensive bleeding edge wafer space on non-area efficient cache. What's that, N2 stacked on N4C cache dies?Do i understand correctly that L3 will be removed from 32 core CCD only?
absolutely massive backorders by huge cloud providers
same for other new #1 performing hardware getting released i.e. nvidia Blackwell, huge B2B supply before releasing client/retail
this is the reality today
What's wrong with Nvidia's ARM cores? Why they still needing Intel CPU's for their AI hardware?What is the new Blackwell platform? The old Hopper platform with Sapphire rapids is probably a little too old.
What's wrong with Nvidia's ARM cores? Why they still needing Intel CPU's for their AI hardware?
I'm in this situation!Still, when you're with AM4, new mobo+new ram +new CPU is a bit hard to justify is many cases IMO. It would be nice but....
I watched the Chips and Cheese video where they interview the AMD chief architect. I find it really interesting and smart that Zen 5 is basically a wider "reset" of Zen. Now they will being to integrate some of the smarter features of Zen 2-4 into Zen 6 and beyond. They really seem to have their house in order.
There are obviously advantages to having one "core" architecture. Less, more focused resources and a unified direction and goal for the team.I think so. So does the Intel E-Core team.
But Intel P-Core team seems to be lost. Probably overwhelmed by its complexity, unable to complete a similar re-vamp.
I have personally seen this to be true over my decades of engineering management.There are obviously advantages to having one "core" architecture. Less, more focused resources and a unified direction and goal for the team.
I was under the impression that Intel is large enough to develop two architectures simultaneously, and they are, but they don't seem to be able to coordinate the teams.
I wonder how many people at the end of the day are actually the ones making the big decisions regarding this CPU's? I'm sure you've heard of the Pareto distribution, which essentially claims that the square root of the number of workers do half the work. So if you have 100 engineers on a project, 10 are doing 50% of the work.
Yes, as much as we want more and better engineers at these companies, good management is very important as well.I have personally seen this to be true over my decades of engineering management.
Additionally, more is not always better. It slows things down when you have to continuously keep 100 engineers on track.
As for multiple core designs, I believe that it is getting increasingly difficult to validate a design on an advanced node. Efficiency is important in design teams as well as the design itself.
Everybody gets excited... it's actually released... twelve CCDs... one of them has V-cache.Threadripper might be getting 3D V-Cache. Interesting.
What you do if you have the resources:Yes, as much as we want more and better engineers at these companies, good management is very important as well.
Do a little scouting, find the best path and send the whole team down it. AMD.
Do a little scouting, can't find the best path but instead 2 or 3 that look promising, split up the teams and send them. Intel.
But how many scouts do you send, how far do they go before they turn around, and how much of the team should follow? Those are important management decisions.
AMD seems to have found a good path and went all in on it. Intel seems to be a little lost currently.
Everyone (including me) gets excited because we want a proper HEDT part. Threadripper addresses a small market because AMD treats it like an enterprise or workstation part. If they released a proper 24+ core part that hit 5.7+ ghz (+X3D) at a somewhat reasonable price without exotic $1,000+ motherboard prices, and if it were unlocked, they’d sell the parts. I’d definitely pay $1,200 for a 32 core 5.7ghz (peak) chip with quad channel memory and say 32 lanes of PCIE 5.Everybody gets excited... it's actually released... twelve CCDs... one of them has V-cache.
It makes sense. Genoa-X is a thing, and so was Milan-X. No reason why they can't sell premium Threadripper parts based on those products.Threadripper might be getting 3D V-Cache. Interesting.
i think its more viable now with the improvements to clock rate with zen5s new cache die arrangement. TR always had very boost clocks compared to full on server parts.It makes sense. Genoa-X is a thing, and so was Milan-X. No reason why they can't sell premium Threadripper parts based on those products.