AMD's Future Chips & SoC's: News, Info & Rumours.

Page 85 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

InvalidError

Titan
Moderator
Board manufacturers can't exactly guarantee that their motherboards will be compatible with future chips based on an architecture and process that didn't exist yet at the time the boards got designed. At best, they can over-engineer their boards to some extent and hope future chips won't stray too far from the first-gen specs. Unfortunately, it doesn't make much sense for AMD to compromise its future chip designs only to maintain compatibility with more marginal boards, so I am fully expecting most manufacturers to have a fair number of boards with limited or no Zen 2 support.
 
Board manufacturers can't exactly guarantee that their motherboards will be compatible with future chips based on an architecture and process that didn't exist yet at the time the boards got designed. At best, they can over-engineer their boards to some extent and hope future chips won't stray too far from the first-gen specs. Unfortunately, it doesn't make much sense for AMD to compromise its future chip designs only to maintain compatibility with more marginal boards, so I am fully expecting most manufacturers to have a fair number of boards with limited or no Zen 2 support.
They totally can as they receive samples and spec sheets WAY in advance. It's all a matter of how they qualify for certification and who owns the responsibility if it goes south. Remember they are making the new boards as well, so they know perfectly well how the previous chipsets are like compared to the new ones. I'm sure most of the wiring is the same and some electrical signaling changed, which can be either re-routed or just done with previous versions of the boards.

Also, do you really expect the MoBo layout for the new stuff to be that different while using the same socket? I'm sure it's not that much different, if at all.

Cheers!
 

InvalidError

Titan
Moderator
They totally can as they receive samples and spec sheets WAY in advance.
It is impossible for board manufacturers to have any data on Zen 2 well over two years ago since AMD had no data on TSMC's 7nm process to base any such data on.

The motherboard layout may not change but VRM specifications on the other hand can change quite a bit due to process changes, architecture changes, multi-chip design, changes to how the chips change power states, etc. Changes in VRM performance requirements appear to be the main reason why at least some board manufacturers are being cautious about promising any sort of forward compatibility until older board models have been fully qualified for the new CPUs.

This is similar to how Haswell and newer Intel CPUs require higher performance PSUs to cope with faster transients from cores going from sleep to full-clock and back an order of magnitude faster than before. If Zen 2 has more aggressive power management across its multi-chip design, it will require a VRM able to cope with likely faster and larger transients than what AMD put in the Zen/Zen+/AM4 specs and what board manufacturers may have engineered their boards for, especially at the lower-end where the over-engineering effort may be slim to non-existent.
 
Who mentioned anything about time? And why do they need to know about Zen2 when all they care about to start modifications are spec sheets?

But to humor you about times: when they move to Assembly and QA they need CPUs to test with. I would assume, from design, verification and qualification for build is a good 6 months or more, with some prototyping in-between. The timelines do add nicely for them to make the decision BEFORE LAUNCH to know whether or not what they had out in the wild will be good for the new CPUs.

Why are you being so contrary to the positive idea of backwards compatibility? Still gutted over the Z170 fiasco/shenanigans? It is no secret Intel blocked that, but here AMD seems to be OK with them doing it :p

Cheers!
 

InvalidError

Titan
Moderator
But to humor you about times: when they move to Assembly and QA they need CPUs to test with. I would assume, from design, verification and qualification for build is a good 6 months or more, with some prototyping in-between. The timelines do add nicely for them to make the decision BEFORE LAUNCH to know whether or not what they had out in the wild will be good for the new CPUs.
There was no Zen 2 CPU for motherboard to do any testing with when 300-series motherboard were designed three years ago, TSMC didn't even have its 7nm up and running for qualification to give AMD performance and power characterization data for AMD to finalize Zen 2's specifications through simulations. In other words, there was no way for motherboard manufacturers to know back then what their motherboards would need to guarantee compatibility with Zen 2 and beyond.

Yes, motherboard manufacturers may be able to update their support list now that engineering samples are available, but that won't help people who already own first-gen motherboards that won't make the cut.

I couldn't care less about Intel's 100-series ' fiasco'. I'm still using an i5-3470 for my main PC and whatever I may upgrade to in the next year or two I likely won't be upgrading again within the following five years. I'd be perfectly fine with single-generation chipsets and motherboards or even soldered CPUs if it improves reliability, performance, stability, efficiency, etc. while reducing total cost.
 
I'm not even sure what you're arguing about anymore... It's kind of obvious a Company can't manufacture something and be 100% sure it will work (be compatible) for products years down the line. That's not the point here.

AMD made a simple assessment: "we'll keep using AM4". Fair enough. MoBos for Ryzen "1K gen" were made aplenty. Power characteristics and pin alignments of Zen2 were given to MoBo manufacturers and realized "hey, our old MoBos can actually handle this with a BIOS update!". It's not more complicated than that (well, it is, but...).

MSI said they will provide support for some boards and AMD hasn't said anything back (AFAIK) about it. Consumers win. The amount of actual MoBos that do end up with "official" support is just a side-thought that's not really important, I'd say.

On the other hand, what would AMD be really missing by not changing the socket like Intel has done almost every gen since Sandy Bridge? What is there, technologically speaking, that would make a good Consumer-side argument to force them to change almost the full platform for their new CPU? DDR5? PCIe4? More PCIe lanes? Tri/Quad Channel memory?

Cheers!
 

InvalidError

Titan
Moderator
On the other hand, what would AMD be really missing by not changing the socket like Intel has done almost every gen since Sandy Bridge?
Keeping the same socket is pointless if changes in electrical specifications break backward compatibility with a possibly significant portion of existing boards that were designed prior to new chips' specifications becoming available. This is made much worse by the fact that AMD has no plan to launch A5xx/B5xx-series chipsets to go along with Ryzen 3xxx, which means anyone who does not want to pay the X-series premium will have to use 300/400-series boards, which means people shopping for boards will have to check for compatibility on a board-by-board basis.

Depending on how widespread compatibility issues between older boards and Zen 2 are, this could easily turn into a PR nightmare worse than only committing to two-years sockets.
 
Keeping the same socket is pointless if changes in electrical specifications break backward compatibility with a possibly significant portion of existing boards that were designed prior to new chips' specifications becoming available.
But all sockets have un-used pins for that specific scenario. If the power requirements change, they can swap power pins around or play a bit with the capabilities inside the CPU to regulate power (Intel did this with Haswell, IIRC; not sure AMD does it though).

My questions (the ones you ignored) still stand though.

This is made much worse by the fact that AMD has no plan to launch A5xx/B5xx-series chipsets to go along with Ryzen 3xxx, which means anyone who does not want to pay the X-series premium will have to use 300/400-series boards, which means people shopping for boards will have to check for compatibility on a board-by-board basis.

Depending on how widespread compatibility issues between older boards and Zen 2 are, this could easily turn into a PR nightmare worse than only committing to two-years sockets.
Well, in all fairness, if the chipset doesn't need to change, why force a new version of them? You call it a PR nightmare, I call it smart use of resources. There's nothing technologically awesome they need to include in the mainstream chipset to necessitate a new one, is there?

On the other hand, the power requirements for the Ryzen "2k gen" are already high and voltage wise, I'm pretty sure power regulators in current Motherboards (even from 2012) can cope with such a change without needing to use new regulators; specially higher end models with VRM using 10+ phases. And from the looks of it, the power characteristics of the Ryzen "3k gen" will be similar-ish to the "2K gen", as per MSI's announcement (otherwise, they wouldn't do it?).

And they're not going for a 5xx chipset series? That's actually cool. I can't remember what where the changes from 3xx to 4xx anyway, so I'm sure they just don't need anything better than the 4xx chipset series for Ryzen "3K gen"?

Well, when they give us more information about the CPU itself, we'll be able to figure out more, I guess.

Cheers!
 

InvalidError

Titan
Moderator
On the other hand, the power requirements for the Ryzen "2k gen" are already high and voltage wise, I'm pretty sure power regulators in current Motherboards (even from 2012) can cope with such a change without needing to use new regulators; specially higher end models with VRM using 10+ phases.

And they're not going for a 5xx chipset series?
A 100 phases 2000W VRM won't help you if it was designed for 1A/us transient response speed and the new chips end up requiring faster transient response to keep up with more aggressive power management across an increased number of cores.

AMD will be launching the 570X chipset around the same time as the 3000-series but no B5xx/A5xx until later. This is bound to be problematic as it means people who don't want to pay the X-series tax will have to buy 300/400-series boards which won't have out-of-the-box support for 3rd-gen and need to borrow 1st/2nd-gen CPUs to update the BIOS on eligible boards first. That's a deal-breaking hassle in my book even with AMD's loaner CPU program, which is itself a rather expensive fix to that particular PR nightmare.
 
A 100 phases 2000W VRM won't help you if it was designed for 1A/us transient response speed and the new chips end up requiring faster transient response to keep up with more aggressive power management across an increased number of cores.

AMD will be launching the 570X chipset around the same time as the 3000-series but no B5xx/A5xx until later. This is bound to be problematic as it means people who don't want to pay the X-series tax will have to buy 300/400-series boards which won't have out-of-the-box support for 3rd-gen and need to borrow 1st/2nd-gen CPUs to update the BIOS on eligible boards first. That's a deal-breaking hassle in my book even with AMD's loaner CPU program, which is itself a rather expensive fix to that particular PR nightmare.
But what's important in the power delivery of the chipsets is the lower voltage over the grunt amperes it can deliver, isn't it? That's why I mentioned the phases instead, as that's what helps with voltage drops and lower voltages, right? My memory about power delivery is a bit out of date, so I could be wrong, but that's what I remember :p

In any case, I'm still confident in my comment over power delivery for current chipsets. I'd even say current 3xx and 4xx lower end boards could get away with the requirements for the equivalent mid-low range of Ryzen "3k gen". Well, time will tell, but like I said, I'm confident :p

And thanks for that clarification. I thought you meant "no 5xx chipset at all". If they intend to support PCIe4 immediately, I would imagine they'd need a new PCH so they don't fall back to PCIe3, I guess?

I have to say I'm not a fan of forcing upgrades with no good reason. I'm still very salty about Llano (FM1 to FM2 sockets), the 890FX to 990FX (chipset, not socket) forced upgrade for Piledriver over Derpdozer and, more recently, Intel shenanigans since 2012 with Sandy/Ivy. I could go even further back and say you had the choice of using s370 CPUs in slot A! I mean, if you can do that in the 2000s, why can't we now? /rant

Cheers!
 

InvalidError

Titan
Moderator
But what's important in the power delivery of the chipsets is the lower voltage over the grunt amperes it can deliver, isn't it? That's why I mentioned the phases instead, as that's what helps with voltage drops and lower voltages, right?
More phases only spread out VRM's heat dissipation across more devices and a larger area to facilitate cooling, it does very little for transient response which has more to do with bandwidth in the VRM's feedback and control loop. If the VRM is too slow, core voltage will overshoot when cores slow down or go to sleep and undershoot when clocks go up or cores wake up, which increases the likelihood of malfunctions and crashes. All the power in the world is meaningless without sufficiently fast transient response to keep Vcore stable through the worst possible CPU power state transitions. As you pointed out yourself albeit for different reasons, this likely means many boards will get demoted (if not disqualified) in terms of VRM greatness for the 3000s.

As for PCIe4, 300/400-series board may get it for the first x16 slot as long as it does not have signal switches for x16/x8x8 or any other active electronics between the CPU and the slot and the board's signal integrity is good enough to allow it.

Putting new chips in old boards designed before the process used to make the new chips existed comes with too many caveats, compromises and unknowns for my liking.
 

jdwii

Splendid
Just figured i'd say that MSI did update their statement and said that the 300 series will work with Zen 2.

PR and random people at customer support don't always know everything that goes on in the background. Motherboard manufactures have already started rolling out updates to support the newer CPU's on older boards.

Obviously 16 core CPU's with unlocked multiplayer's won't work out well on 4 phase motherboards lol hell most of the 4 phase b350 boards are hitting 100+C with Ryzen 1700 owners who OC.
 
From the article: "At this point, we are still performing extensive testing on our existing lineup of 300- and 400-series AM4 motherboards to verify potential compatibility for the next-gen AMD Ryzen CPUs". Which implies: "we have the new CPU in our labs and we're making sure it works in previous gen stuff".

Not too bad. Looks like the new Ryzens are cooking well?

Cheers!
 

jdwii

Splendid
From the article: "At this point, we are still performing extensive testing on our existing lineup of 300- and 400-series AM4 motherboards to verify potential compatibility for the next-gen AMD Ryzen CPUs". Which implies: "we have the new CPU in our labs and we're making sure it works in previous gen stuff".

Not too bad. Looks like the new Ryzens are cooking well?

Cheers!
Agreed when Amd says 2020 they mean 2020 nay sayers have been trying to ruin every launch by saying otherwise yet here i am i bet i will be able to pop a 3000 series zen 2 chip in my Asrock X370 taichi
 
The only thing I'm not liking so far, is that it appears AMD has decided to mimic the horrible and stupid Intel "numbering" for their tiers... Ryzen 3,5,7 and 9. Because sticking to thousands numbering is too complicated for the average user!

Cheers!
 

InvalidError

Titan
Moderator
The only thing I'm not liking so far, is that it appears AMD has decided to mimic the horrible and stupid Intel "numbering" for their tiers... Ryzen 3,5,7 and 9.
I'm not a fan of redundant information in model numbers either. There is no point in R3/5/7-xxxx when R3/5/7 are synonymous with x0xx-x3xx/x4xx-x6xx/x7xx-x8xx respectively. Things are more confusing on Intel's side with overlapping model number ranges and product names between sockets.
 
Yeah... I don't even know why they think it's going to help. It just makes it even MORE confusing for regular people with all the numbers and letters to remember... They already have their own internal codes for products, so they don't really need a friggin' long naming nomenclature for their stuff.

Ugh!

/rant

Cheers!
 

ASK THE COMMUNITY

TRENDING THREADS