Question Zen 6 Speculation Thread

Page 188 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Doug S

Diamond Member
Feb 8, 2020
3,296
5,728
136
I like how he states Zen 6 is just an evolution of Zen 5. No hype trains in sight.

Regarding the those memory controllers, It's known upcoming server CPUs support both DDR5 and MR-DIMM.

That wouldn't require two separate memory controllers. It wouldn't be that hard to make one that supports both since they are very similar - MRDIMM just requires a little "extra" to support the double pumped data transfers.

A totally separate controller would be for something completely different where you can't just add some stuff to your existing controller to support it - i.e. LPDDR6.
 
  • Like
Reactions: Vattila

Doug S

Diamond Member
Feb 8, 2020
3,296
5,728
136
If there are two memory controllers in the consumer Zen 6 IOD, then that's most likely for DDR5 and DDR6.

Could SERIOUSLY point to DDR6 launching for consumer in H2 2026!

Ridiculous.

The DDR6 spec isn't even announced yet, let alone officially released. Keep in mind that LPDDR6 was announced last year but wasn't officially released until yesterday. There will be zero consumer shipments of DDR6 in 2026, and other than maybe some early access / pre-standard/ testing type stuff under NDA to major customers like cloud providers, no enterprise shipments of DDR6 either.
 
Last edited:

Makaveli

Diamond Member
Feb 8, 2002
4,960
1,557
136
There is no way we are seeing DDR6 in 2026.

Most likely Late 2027 or Q1 2028 they may want to do that and go PCIe 6.0 at the same time for AM6.

Looks like all the "tech sites" are now running there own stories off what Detox posted above.

 

MS_AT

Senior member
Jul 15, 2024
733
1,481
96
BUT I could do an easy test on my Epyc with 256MB cache. You just need to tell me how to run that test step by step like a 5 year old
We would need to actually verify first what you claim to be able to devise correct measurement scenario;)

I think the basic claim was that 2nd x3D chiplet would help games?

Still, before you do measure something, it might be wise to look back what AMD did. For some reason, their people thought that limiting games to single high cache CCD via this funny Xbox driver was a better solution than letting games use both CCDs, even though in total 2 CCDs mean more cache. Then we have Zen2 -> Zen3 example, when 2x4cores CCXs got merged into one 1x8 core CCX even though the total L3 remained constant, this change was claimed to be one of the reasons for the big performance upgrade in games. Plus there are direct claims from AMD that having x3D on second CCD won't result in meaningful performance increase in games. I guess they are talking average, so there might be outliers, but still.

All in all, I think we have enough evidence from different sources to conclude that 9950x3DD won't be on average better gaming CPU for the suite of games available now. And I doubt software developers will care at large about thread placement unless 2 CCD SKUs will start enjoying much greater market share than they do now.
 
Jul 27, 2020
25,996
17,937
146
And I doubt software developers will care at large about thread placement unless 2 CCD SKUs will start enjoying much greater market share than they do now.
Good points overall.

However, most 9950X3D owners (majority of them) bought it because it's the most expensive CPU they could buy without breaking the bank and their back with a Threadripper or Xeon workstation CPU. If AMD is simply not giving us 9950X3DD because it doesn't offer much uplift on average, they are purposefully ignoring money on the table and they should answer to the investors if enough of them understand that this is what's happening. This gamer crowd will pay up to $1000 for a CPU. As many here have pointed out time and again, there are plenty of "normal" hobbies that cost a lot more than PCs. Heck, even maintaining a car can turn out to be more expensive over the life of the car than a frickin' gaming PC.

Anyway, I think AMD is soon going to experiment with multiple CCD V-cache parts in the Shimada Peak line-up, precisely to get data on how much having a lot of cache improves their workstation CPU sales.

There are at least 10 people on these forums who will buy the Threadripper 9000X3DD if it shows good performance in games. And those are only the ones I can think of. The ghost readers must be at least in the hundreds, if not thousands.
 

Kepler_L2

Senior member
Sep 6, 2020
877
3,590
136
Zen6 memory controller changes have nothing to do with DDR6/LPDDR6, they are changing the physical design/topology to allow for higher speeds, that's all (though I'm pretty sure you'll need a new motherboard to actually take advantage of those changes).
 

Joe NYC

Diamond Member
Jun 26, 2021
3,238
4,736
136
All in all, I think we have enough evidence from different sources to conclude that 9950x3DD won't be on average better gaming CPU for the suite of games available now. And I doubt software developers will care at large about thread placement unless 2 CCD SKUs will start enjoying much greater market share than they do now.

Which, they will not (enjoy greater market share), because the new single CCD Zen 6 with 12 cores, 24 threads will alone cover a lot of the ground, most of what 2 CCD Zen 5 does now...
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,644
2,462
136
They are currently using an outsourced memory controller IP where a single controller handles 128 bits, or 4 DDR5 channels (which, for historical reasons, are also called 2 channels).

Presumably, they are switching to a different memory controller IP, that handles 64 bits, or one dimm's worth per controller. And then they had to integrate two of them.

I don't think the switch from 1 to 2 controllers is significant other than that they switched the IP and it happened to be narrower. Presumably, the new controllers were chosen because they are faster.
 

MS_AT

Senior member
Jul 15, 2024
733
1,481
96
If AMD is simply not giving us 9950X3DD because it doesn't offer much uplift on average, they are purposefully ignoring money on the table and they should answer to the investors if enough of them understand that this is what's happening. This gamer crowd will pay up to $1000 for a CPU. As many here have pointed out time and again, there are plenty of "normal" hobbies that cost a lot more than PCs. Heck, even maintaining a car can turn out to be more expensive over the life of the car than a frickin' gaming PC.
Think of bad press Zen5 got on release. Now think of bad press they will get if they market 9950X3DD as gaming part, costing 300$ more for 0% average gaming uplift;) And then they will have trouble moving the inventory. Maybe as Halo part for some anniversary edition or whatever it would work, as these things don't have to make sense.

When it comes to TR or Epyc, I guess there was not enough demand for x3D parts this gen.
 
  • Like
Reactions: Tlh97 and inquiss