Question Zen 6 Speculation Thread

Page 334 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DrMrLordX

Lifer
Apr 27, 2000
23,159
13,254
136
really AMD sees such profit opportunity in having 4p+8c rammed literally everywhere in 5 product families instead of classic 12 full cores ccd?

N2 wafers are stupid expensive. Cost/transistor isn't going down anymore. And supposedly, at least in mobile power envelopes, the dense cores can maintain high enough clockspeed in 9t+ scenarios that there's little-to-no advantage for 12p standard core CCDs. Though I'd like to see some comparisons in mt clockspeeds within variable power envelopes between the two different CCD types. Maybe someone will get us that data, eventually.
 

Thunder 57

Diamond Member
Aug 19, 2007
4,244
7,038
136
AMD users laughed at Intel ADL's big.LITTLE, yet AMD was just a gen or two late with big.LITTLE...

What AMD really needs to step up its game in mobile is to provide a top-noch SW/HW support and viable (complete) platforms to OEMs. OEMs juggling with 7 different SKU lines doesn't seem like it.

That's horse shit and you know it. I don't know anyone who laughed at big.LITTLE. Troll outside of an AMD thread.
 

itsmydamnation

Diamond Member
Feb 6, 2011
3,113
3,960
136
phat.DENSE
DR DENSITY here....... time to horsecock some loads

Look Thick as Frick
 

fastandfurious6

Senior member
Jun 1, 2024
867
1,020
96
yeah I don't get it

server zen6 EPYC cores are both dense and N2

client zen6 has N3P instead???

as with zen5, the flagship-on-mobile-package will be both best deal and best performer

24 full phat N2 cores + 2LP + 3D (X2)
 

fastandfurious6

Senior member
Jun 1, 2024
867
1,020
96
10965HX3D (24 full cores) + dGPU > Medusa Halo

if there's 5 medusa CPU families before Halo, then Halo will be priced 3000+++ and won't be worth vs full 10965HX3D on mobile

only for LLMs basically
 
  • Like
Reactions: ToTTenTranz

ToTTenTranz

Senior member
Feb 4, 2021
872
1,466
136
1 IOD, 1 CCD, 3 SoCs, 2 GMDs (which are also used for dGPUs)

This is less modular than one would think. If the memory controllers are in the GMDs, why exactly do the MDS Premium and Halo use different SoCs? Higher MT CPU performance on either could be had just by adding a CCD.

I get that MDS Point needs a different SoC because it's a cost-focused monolithic oriented design (even if the 8xCU RDNDA3.5 iGPU seems like a very dumb idea). But if the other two MDS aren't using the same IOD, then the general approach isn't making a lot of sense to me.


Olympic Ridge: IOD + CCD
Medusa Point: SoC + (optional) CCD
Medusa Premium: SoC + AT4 GMD
Medusa Halo: SoC + AT3 GMD

Isn't the Magnus SoC + AT2 config missing here, as it follows the same SoC + ATx scheme?





10965HX3D (24 full cores) + dGPU > Medusa Halo

if there's 5 medusa CPU families before Halo, then Halo will be priced 3000+++ and won't be worth vs full 10965HX3D on mobile

only for LLMs basically

That was when large amounts of DRAM were reallistically available to consumers and prosumers, which they aren't at the moment and could still not be a year from now.
Which leaves MDS Halo in a very awkward position.
 
  • Like
Reactions: marees

ToTTenTranz

Senior member
Feb 4, 2021
872
1,466
136
That's not part of client Zen6
The only difference between Magnus and the rest is it'll initially be sold to a single and exclusive client, Microsoft. It it's using AT2 which is also going into dGPUs, then it's safe to assume there are no custom blocks in the silicon (not that there were any relevant ones in Scarlett and Lockhart anyway).


Even the client exclusivity may end, as AMD eventually sold Series X chips as 4800S and even XBOne chips as A9-9820. The 4800S is basically the Series X with the iGPU disabled, so the Magnus equivalent to that would be just the SoC without the AT2 attached.




That seems unfortunate that they don't have an N2P version of the SoC for Halo/Halo mini. It'll only be a problem for Halo mini I guess.
I'd say it's more to gain in having the GMDs made on N2P than the SoC.
 
  • Like
Reactions: marees