Question Zen 6 Speculation Thread

[gdansk]

I forgot that AMD removed SATA controller from the Ryzen SoCs, it slipped my mind.

It seems there will be AM5 Zen 6 which suggests you could still use your X670 board (probably) with its Prom21 chipset. But for new boards, would it need a chipset if they add more PCIe lanes to the SoC as requested? That is my question...

 

[DavidC1]

Stripping few legacy features from motherboards aren't gonna bring the cost of $300 boards to $200, probably not even $250.

Motherboard margin on $100 boards are really low. They know this, there are seriously lot of stuff there. So they try to make money to differentiating on features and the profit is on $200+ boards. Which enough people seem to buy for manufacturers to keep making them. The raw cost difference between a $100 board and even a $300 board is probably not much.

You expected prices to go down after we got just one main chip on the board right? Well, nothing like that happened. AMD/Intel probably knows this too, that if they completely eliminate the chipset it's just going to kill their partners. The thing is they probably can eliminate the PCH, with the main chip becoming an SoC.

Think of it in terms of a mobo manufacturer. Why would you push for lower selling cost boards when it's just a death march to bankruptcy?

 

[gdansk]

Think of it in terms of a mobo manufacturer. Why would you push for lower selling cost boards when it's just a death march to bankruptcy?

No chipset = lower BoM and consumers will still see total features they care about increasing so you don't have to lower your ASP. But the consumers are really paying for the new features because it's on the SoC now! It's all win for them. And even for consumers it seems like if you want AMD to care about its power draw, if you want it cooled properly, and if you want it on a newer process you want it on the SoC.

 

[burninatortech4]

Exactly!

Chipset-less design.

There is a configuration of I/O that would be satisfactory to 95% of users, and AMD could just put it in the CPU.
- I don't need 28 USB headers that x670E chipset offers. I don't even need half of that.
- SATA will be useless by Zen 6
- 1 network port covers 95% of users. Make it 10 GB/s
- WIFI, Audio a little more controversial, could still be separate chips with PCIe lane coming in.
- more PCIe directly from CPU

This could cut the prices of motherboards from $200, $300 (and up). Reduce them to $100 for basic mobo. Many of which could be mATX

Funny thing is that the original Zeppelin die had 10Gb ethernet MAC's. I don't know of many designs that used them though. Definitely hoping to see it become exposed more often outside of a strictly homelab/hobbyist context.

 

[Seba]

SATA ports (usualy 4) are present even on $50/50EUR motherboards (new ones), so it can't be that expensive to implement.

 

[gdansk]

SATA ports (usualy 4) are present even on $50/50EUR motherboards (new ones), so it can't be that expensive to implement.

SATA ports (usually 2) are present also on small, cheap boards without Prom21. Board makers can optionally add a (cheaper) SATA controller instead.

 

[Seba]

One SATA port takes about the same space on the motherboard as an A-RGB header. And it is much more useful.

So those arguments about cost and motherboard space are not valid.

Yeah, I know that shiny lights sell PCs now. And reviewers love them. Just saying what I think.

 

[Thibsie]

Killing SATA when you still need stuff like LPC ? Nah...
Last desktop "corporate"-type PCs we got a school still have PS/2 ports. lol

 

[LightningZ71]

A couple things:
1) They make M.2 slot add-in PCIe SATA controllers that can drive upwards of 4-5 SATA drives. They are DIRT cheap to buy. For those SOCs that provide 16+4+4(or even + another 4) PCIe lanes, I'd MUCH rather the board just feature 2 or even three M.2 slots and let me get one of those M.2 Sata adapters if I decide that I need it.
2) The 10GbpsE controllers on the Zepplin die were only ever exposed on a handful of embedded products. They were one of those "you only pay for the IP license if you expose the functionality" add-ons that just made sense to add to a jack of all trades processor.
3) There's enough I/O out of these SOCs to build an mATX board without an I/O chipset, expose all the ports on the SOC for the I/O headers, and drive an x16 slot and 2 or 3 x4 slots. Those x4 slots can hold any combination of multi-I/O card you could ever need, ethernet cards and even multi SSD cards.
4) Promontory is nothing more than just another PCIe host. It certainly does a lot, but, for most of it's functions, it doesn't need to be embedded on the motherboard. It is possible to build it on a PCIe add-in card and get 90% of it's functionality. This isn't some pie in the sky idea as it's ALREADY BEEN DONE... https://www.tomshardware.com/news/asus-rog-x670e-i-has-a-unique-design

There could be a lot of neat boards out there that are cheap to build, cheap to sell, but offer the flexibility for end users to customize how they need.

 

[Joe NYC]

A couple things:
1) They make M.2 slot add-in PCIe SATA controllers that can drive upwards of 4-5 SATA drives. They are DIRT cheap to buy. For those SOCs that provide 16+4+4(or even + another 4) PCIe lanes, I'd MUCH rather the board just feature 2 or even three M.2 slots and let me get one of those M.2 Sata adapters if I decide that I need it.
2) The 10GbpsE controllers on the Zepplin die were only ever exposed on a handful of embedded products. They were one of those "you only pay for the IP license if you expose the functionality" add-ons that just made sense to add to a jack of all trades processor.
3) There's enough I/O out of these SOCs to build an mATX board without an I/O chipset, expose all the ports on the SOC for the I/O headers, and drive an x16 slot and 2 or 3 x4 slots. Those x4 slots can hold any combination of multi-I/O card you could ever need, ethernet cards and even multi SSD cards.
4) Promontory is nothing more than just another PCIe host. It certainly does a lot, but, for most of it's functions, it doesn't need to be embedded on the motherboard. It is possible to build it on a PCIe add-in card and get 90% of it's functionality. This isn't some pie in the sky idea as it's ALREADY BEEN DONE... https://www.tomshardware.com/news/asus-rog-x670e-i-has-a-unique-design

There could be a lot of neat boards out there that are cheap to build, cheap to sell, but offer the flexibility for end users to customize how they need.

That's an interesting product, interesting idea.

Getting to much cheaper motherboards (with chipset-less designs, enabled by adding some additional functionality to Zen 6 could re-invigorate the DIY market. I think the ~ $300 prices of mobos are suffocating this market. Also, one of the reason why socket AM4 refuses to die. There is a good selection of cheap mobos.

BTW, this seems to be repeating on the Intel side. I see people ordering $499 mobos...

 

[inquiss]

How many people use more storage than what their PC comes with? 5%? What's wrong with forcing them to get the more expensive board/case, if they want that HDD internal, or a $10 SATA to USB converter if they're OK with having it external? Why do we need to design something into every PC that an ever decreasing percentage of people will have any need for?

Legacy data SSDs. Like, you're in a tech forum interested about tech. Your use cases are likely gonna be niche. SATA isn't going away soon

 

[igor_kavinski]

SATA isn't going away soon

Yeah. Gamers and even non-gamers need to hoard "stuff" on HDD JBODs or RAID arrays...

 

[MadRat]

You can easily just $1000 motherboards if you have a killer app driving sales. Let us face it, you get so much more these days than we did in 1995. We used to spend the equivalent of $500 on motherboards in those days, throwing serious money at L3 cache, dual sockets, new peripheral integrations, etc. The problem today isn't the value, its the lack of excitement around new parts. Crap from 5 years ago runs everything good enough.

 

[marees]

https://www-chiphell-com.translate.goog/thread-2648487-1-1.html?_x_tr_sl=auto&_x_tr_tl=en&_x_tr_hl=en-GB

 

[misuspita]

Soooo... Halo will have stacked V Cache?

 

[marees]

Soooo... Halo will have stacked V Cache?

I am thinking halo would be a nuc / desktop model

Probably a strix point successor could have 3d v-cache ?

 

[Joe NYC]

I am thinking halo would be a nuc / desktop model

Probably a strix point successor could have 3d v-cache ?

MALL cache could be stacked.

AMD apparently had to remove MALL cache from Strix Point in order to accommodate the NPU.

 

[Thunder 57]

MALL cache could be stacked.

AMD apparently had to remove MALL cache from Strix Point in order to accommodate the NPU.

Boneheaded decision. Unfortunatly they pretty much had to because AI!

 

[MadRat]

Shouldn't AI be a dedicated chiplet?

 

[FlameTail]

Kepler says Zen6 is to be expected in late 2026/early 2027.

https://twitter.com/x/status/1854196214390030539

 

[Gideon]

Kepler says Zen6 is to be expected in late 2026/early 2027.

https://twitter.com/x/status/1854196214390030539

So they still fail to deliver faster than 22 months? This is more like 26 months even!

If true, that's terrible for two reasons:

• Lisa Su promised multiple times to adhere to a 12-15 month schedule. How can you promise that and then miss it 3 times in a row? By 7+ months? (Yeah yeah covid, etc, but it didn't bother others)
• The competition (at least the ARM competition) will release twice in that timeframe. If it's even half of the M2 -> M3 -> M4 uplift, AMD will fall even further behind. Unless the rumored 10% IPC uplift isn't true

 

[Meteor Late]

So they still fail to deliver faster than 22 months? This is more like 26 months even!

If true, that's terrible for two reasons:

• Lisa Su promised multiple times to adhere to a 12-15 month schedule. How can you promise that and then miss it 3 times in a row? By 7+ months? (Yeah yeah covid, etc, but it didn't bother others)
• The competition (at least the ARM competition) will release twice in that timeframe. If it's even half of the M2 -> M3 -> M4 uplift, AMD will fall even further behind. Unless the rumored 10% IPC uplift isn't true

I expect Zen 6 to get a very healthy frequency bump, though, so 10% IPC uplift ain't bad at all if it clocks above 6GHz, and I expect it to clock that high, it's a straight jump into at least N3P.

 

[igor_kavinski]

Why do people think AMD won't release a Zen 5 refresh in the intervening period?

Ya know, the XT series aka 9950XT etc.

 

[igor_kavinski]

So they still fail to deliver faster than 22 months? This is more like 26 months even!

I think they are choosing to go slowly and carefully rather than rush things through and failing to catch some devastating bug (like the Arrow Lake regression in games).

 

[MS_AT]

I think they are choosing to go slowly and carefully rather than rush things through and failing to catch some devastating bug (like the Arrow Lake regression in games).

Even if so, it still means they are falling behind ARM competition that is able to iterate faster without obvious issues.