Next-gen AM5 Motherboard Platforms Could Support USB4

[Assimilator]

My concern is that AMD has coupled themselves to ASMedia too tightly. The latter has already proven they are behind the curve with the delay to USB4 preventing it from being integrated in Promontory 21, I worry that they will have similar delays with PCIe 5.0. I would very much be in favour of AMD opening up the floor to anyone who wants to make chipsets for AM5 - basically just provide the set of connectivity that such a chipset needs to fulfil and how it communicates with the CPU, and see who steps up to build one. Chipset competition is one of the reasons that motherboards were so great back in the socket 754/939/AM2 days, I'd really love to see that return.

 

[Tek-Check]

*14th generation in fact

Google Bard seems to be uninformed and cannot interpret the context. In other words, it's a bit stupid, as all AIs are... Have you read that AI engines such as Chat GPT4 have been recently heavily downgraded in capabilities? There are reasons for this... Find it and read.

Back to Gen5 chipset link. Google Bard is oblivious to two sides of chipset link. Yes, Ryzen 7000 CPUs have all 28 PCIe lanes at Gen5, so can provide Gen5 speed to chipset too, but the chipset itself from AsMedia is Gen4 validated. So, no Gen5 chipset link unfortunately, until the chipset itself is designed with Gen5 PHY too.

Bard is also reporting "128 GB/s" which is nonsense. It should be 128 Gbps.

As for Intel, Raptor Lake Refresh will have the same DMI link, which is x8 at 128 Gbps. Where does Bard get "128 GB/s"? As far as I know, Intel has not indicated anywhere a new revision of DMI link that would leverage Gen5 speed. Have you?

 

[TumbleGeorge]

Have you?

I have not conducted my own search. Google spits out a lot of results and I can't read it all.

 

[Deleted member 232534]

Google Bard wrote:
Разбира се, ето превода на отговора ми на английски език:
"Hello!
After conducting a thorough search on the internet, I have found credible information that contradicts the claim that AMD and Intel are not considering Gen5 chipset link.
On June 1, 2023, AMD published an article on its website announcing that it will release a new line of Ryzen 7000 processors that will use Gen5 chipset link. The article states:
"Ryzen 7000 processors will be the first AMD processors to use Gen5 chipset link. Gen5 chipset link is a new standard for connectivity between the processor and the chipset that offers significant performance and efficiency improvements. Gen5 chipset link can support up to 128 GB/s data transfer speeds, which is twice as fast as Gen4 chipset link. This performance improvement will allow Ryzen 7000 processors to reach new levels of performance."
On June 2, 2023, Intel also published an article on its website announcing that it will release a new line of Core 13th* generation processors that will use Gen5 chipset link. The article states:
"Core 13th generation processors will be the first Intel processors to use Gen5 chipset link. Gen5 chipset link is a new standard for connectivity between the processor and the chipset that offers significant performance and efficiency improvements. Gen5 chipset link can support up to 128 GB/s data transfer speeds, which is twice as fast as Gen4 chipset link. This performance improvement will allow Core 13th generation processors to reach new levels

of performance."
In conclusion, there is credible information that contradicts the claim that AMD and Intel are not considering Gen5 chipset link. Both companies have announced that they will release new lines of processors that will use Gen5 chipset link in 2023."

*14th generation in fact

I have not conducted my own search. Google spits out a lot of results and I can't read it all.

what?

so why you post this nonsense without basic fact checking? to waste people time? there are no such articles on the companies website - you could have checked that

 

[TumbleGeorge]

What you say about updated new models with z790 chipset which was announced on computex 2023?

 

[Tek-Check]

My concern is that AMD has coupled themselves to ASMedia too tightly. The latter has already proven they are behind the curve with the delay to USB4 preventing it from being integrated in Promontory 21, I worry that they will have similar delays with PCIe 5.0. I would very much be in favour of AMD opening up the floor to anyone who wants to make chipsets for AM5 - basically just provide the set of connectivity that such a chipset needs to fulfil and how it communicates with the CPU, and see who steps up to build one. Chipset competition is one of the reasons that motherboards were so great back in the socket 754/939/AM2 days, I'd really love to see that return.

I agree with this. However, it's not that simple. Devil is always in detail. And, you are introducing another, third USB4 product, beyond host and hub chips, which is the chipset itself. That's even more complex on the top of what they are trying to achieve now.

USB4 host and hub chips
RealTek has only recently announced, at Computex, that they are working on USB4 host and hub chips. Until then, AsMedia was the only vendor in the world developing the solution. ASM4242 host chip will be available around Xmas this year. RealTek possibly next year. It does look like that Intel, despite "donating" Thunderbolt spec to USB-IF, they did not tell them everything, hence delays with USB4. So, AMD didn't have anyone to open up to.

Promontory chipset with integrated USB4
This would be a completely new product that no one has heard about to date. This might arrive on AM6 platform and needs entire development, testing, validation and USB-IF certification cycle. This usually takes 2-3 years... Such chipset with integrated USB4 controller would be bigger and more expensive too, no doubt.

Also, I doubt they would ever place USB4 on the chipset itself as this complicates delivery of DisplayPort traces from CPU. Remember, USB4 unites DP, PCIe, USB and power delivery in one single port, so traces need to be as short as possible. Moving USB4 controller far away from CPU complicates matters, a lot. Discrete USB4 controller, like TB4 one, is usually placed as close to rear I/O on motherboard as possible. There are essential reasons for this.

My take on Lisa Su's recent visit to Taiwan and AsMedia was to decide whether Ryzen 8000 desktop CPUs would integrate USB4 controller, like mobile APUs do, or rely on discrete chip ASM4242 from AsMedia. They would have also decided whether to double the bandwidth of the chipset link. We will see in a few months once more leaks are out.

FYI, A Computer Guy, none of this is directed at you, just answering your point here, and giving a little education to the youngsters here, many of which might not know about the history of computers and just how much has changed in the last 3 decades.

Exactly. Most stuff have been integrated into CPU, USB and chipset, hence less need for more PCIe slots.

Again, a tiny minority who still need more PCIe slots have two options. One is to buy HEDT or WS system, and another one is to organise a crowd-funding community in collaboration with board vendors for custom semi-HEDT board. I am sure Supermicro or others would be on-board with such ideas. There just needs to be enough interest, I guess minimum 1000 units, to justify the effort.

What you say about updated new models with z790 chipset which was announced on computex 2023?

Is there anything to say about it?
We haven't heard that Intel is miraculously doubling the bandwidth of their DMI chipset link. Why would they do it on a good-bye platform?

 

[TumbleGeorge]

Is there anything to say about it?
We haven't heard that Intel is miraculously doubling the bandwidth of their DMI chipset link. Why would they do it on a good-bye platform?

I needed a comparison since the AM5* is not a goodbye platform and you're putting things off until the AM6?

*Because is on topic

 

[Tek-Check]

I needed a comparison since the AM5* is not a goodbye platform and you're putting things off until the AM6?

USB4 will come on AM5, but not as integrated solution in the chipset. USB4 will either be integrated into CPU, if there are two x4 PHY available in 1718 pin package, or as discrete ASM4242 chip from AsMedia. That much we know.

For doubling the chipset link from current 64 Gbps to 128 Gbps, no rumour has come out, as yet. It's up in the air. They either need to sacrifice one Gen4 M.2 drive on the chipset for x8 Gen4 chipset link, or get AsMedia to produce new Promontory version of chipset with another x4 Gen4 PHY. This is also what Lisa Su might have discussed in Taiwan.

 

[trsttte]

AMD is in more bizarre position because they first need to double the throughput. But, if they do this, they waste x8 Gen5 lanes on x8 Gen4 chipset link. So, next generation of Promo chipset for AMD platforms should have Gen5 connectivity, at least to CPU. Such chipset will work almost like PCIe switch. Gen3 connectivity on chipsets will disappear all together from both vendors.

My guess for AMD desktop chipset is evolution in two steps. First, new edition of Promontory chip will have x8 Gen4 link on AM5 platform. Step two, on AM6 platform this will change to x8 Gen5 for higher X chipset and x4 Gen5 for B systems.

I think need is a strong word, they're doing fine with the current setup. I believe it's more likely they jump to an x4 gen5 link than the x8 intermediary step, that would create a more confusing motherboard line-up with the middle x8 step having less cpu lanes available for usb 4 or m.2 connections - as is, they'd need to choose between having usb4 or m.2 on the cpu to free an extra 4 lanes for the chipset and amd has been putting at least an m.2 on cpu for a long time and as you mentioned usb4 should also be connected to the cpu.

*14th generation in fact

Dude, Bard doesn't know what it's talking about, it's just making stuff up...

 

[Tek-Check]

I think need is a strong word, they're doing fine with the current setup. I believe it's more likely they jump to an x4 gen5 link than the x8 intermediary step, that would create a more confusing motherboard line-up with the middle x8 step having less cpu lanes available for usb 4 or m.2 connections - as is, they'd need to choose between having usb4 or m.2 on the cpu to free an extra 4 lanes for the chipset and amd has been putting at least an m.2 on cpu for a long time and as you mentioned usb4 should also be connected to the cpu.

We don't know how much space is left on 1718 pins to add more x4 PCIe PHY.

If there is space for another two x4 links on CPU, one could be dedicated to USB4 and another one to the chipset link, without reducing 24 already available lanes.

We will just need to wait and see until more leaks about Granite are out. Surely, they will need to introduce new features, so integrated USB4, x8 chipset link and RDNA3 iGPU would be a great selling point, in additional to new Zen5 architecture.

 

[chrcoluk]

Back in the day of ISA slots being the norm, you NEEDED those slots because by default the only I/O was a keyboard.!!! So if you want a mouse, you needed a serial port, if you need a printer you need a parallel port, if you need to connect more than just a floppy drive you need an IDE controller, and let us not forget that you absolutely must have Graphics output, what about sound, and networking, a connection to the internet requires a modem, which is connected to a parallel port, so you now need two.

In total, you need ISA/PCI slots for graphics, mouse/printer (one card for both), drives, modem, sound, networking. That's 6 cards to bring us into basic parity with a modern motherboard that doesn't need to use a single card, just the I/O on the motherboard (Modem replaced with WIFI).

The number of available slots on motherboards has dropped dramatically, and almost all of this has been down to integration, and thus needs of the average user. Going back to the above ISA/ early PCI era system, most people didn't have a sound card or modem or network card, so they only needed to use 3 cards/slots, but if you wanted to expand your system further, you needed to add more cards, whereas nowadays a lot of that expansion is done via USB, which is integrated, wheras in prior years you would need to add additional cards to get more parallel /and serial ports to have for example a Tape backup drive.

FYI, A Computer Guy, none of this is directed at you, just answering your point here, and giving a little education to the youngsters here, many of which might not know about the history of computers and just how much has changed in the last 3 decades.

I will half agree with you, I had my first computer in 1996, my usually harsh dad surprisingly said yes when I pointed to one in the computer shop that out spec'd all my friends machines.

Yes we had serial and parallel ports commonly used, isa graphics cards, networking that required terminators, and onboard sound I am pretty sure wasnt a thing so sound card usage was a given for gamers.

I am pretty sure my parallel and serial were both onboard (so maybe you are talking further back), I did have a discrete graphics card (cirrus logic), had to buy a network card to do LAN gaming, sound card as well of course, but that was it in the slots.

Fast forward to today.

Some people still use discrete soundcards, network cards, storage cards, add on capture cards, cards that add i/o options that come on future boards (usb4 would be an example). It may not be a majority of people, but PC's historically have served these use cases.

Using myself as am example.

Currently populated slots in main PC.

GPU x16 gen 5 slot
Intel network x1 gen 3 slot (this might not be permanent, noticed onboard network has bad performance)
Intel DC P4600 SSD x4 gen 3 slot
Asus Xonar D2X x1 gen 3 slot
x4 gen 4 slot still empty.

2nd rig

ASmedia SATA card x16 gen 3 slot
970 EVO via M.2 to PCIe adaptor x4 gen 3 slot
(usually also intel network card in x1 gen 2 slot, but main pc borrowed it).

In the past I remember once adding USB3 support to a USB2 motherboard via an addon card.

I agree with this. However, it's not that simple. Devil is always in detail. And, you are introducing another, third USB4 product, beyond host and hub chips, which is the chipset itself. That's even more complex on the top of what they are trying to achieve now.

USB4 host and hub chips
RealTek has only recently announced, at Computex, that they are working on USB4 host and hub chips. Until then, AsMedia was the only vendor in the world developing the solution. ASM4242 host chip will be available around Xmas this year. RealTek possibly next year. It does look like that Intel, despite "donating" Thunderbolt spec to USB-IF, they did not tell them everything, hence delays with USB4. So, AMD didn't have anyone to open up to.

Promontory chipset with integrated USB4
This would be a completely new product that no one has heard about to date. This might arrive on AM6 platform and needs entire development, testing, validation and USB-IF certification cycle. This usually takes 2-3 years... Such chipset with integrated USB4 controller would be bigger and more expensive too, no doubt.

Also, I doubt they would ever place USB4 on the chipset itself as this complicates delivery of DisplayPort traces from CPU. Remember, USB4 unites DP, PCIe, USB and power delivery in one single port, so traces need to be as short as possible. Moving USB4 controller far away from CPU complicates matters, a lot. Discrete USB4 controller, like TB4 one, is usually placed as close to rear I/O on motherboard as possible. There are essential reasons for this.

My take on Lisa Su's recent visit to Taiwan and AsMedia was to decide whether Ryzen 8000 desktop CPUs would integrate USB4 controller, like mobile APUs do, or rely on discrete chip ASM4242 from AsMedia. They would have also decided whether to double the bandwidth of the chipset link. We will see in a few months once more leaks are out.


Exactly. Most stuff have been integrated into CPU, USB and chipset, hence less need for more PCIe slots.

Again, a tiny minority who still need more PCIe slots have two options. One is to buy HEDT or WS system, and another one is to organise a crowd-funding community in collaboration with board vendors for custom semi-HEDT board. I am sure Supermicro or others would be on-board with such ideas. There just needs to be enough interest, I guess minimum 1000 units, to justify the effort.


Is there anything to say about it?
We haven't heard that Intel is miraculously doubling the bandwidth of their DMI chipset link. Why would they do it on a good-bye platform?

You talk about having standard PCIe connectivity as semi HEDT and niche.

But bear in mind we already have board's manufactured for extremely niche use cases which are likely much smaller numbers such as extreme overclocker boards that are not far of 1000USD to buy.

I dont know how board manufacturers find out demand for features (one that I am aware off have a feedback form or feature request form). Their communication with the world is likely only via retailer feedback and the media, and the media is often out of sync with what users might want.

What I do know is I know over 20 people personally who arent happy with PCIe options shrinking, and on top of that 100s of posts read on the internet about similar, so its likely to be much more than a 1000 units required to fulfil demand, I would speculate at least 5 figures. I would expect such boards to comfortably outsell maximus boards e.g. and most definitely APEX boards.

 

[Tek-Check]

What I do know is I know over 20 people personally who arent happy with PCIe options shrinking, and on top of that 100s of posts read on the internet about similar, so its likely to be much more than a 1000 units required to fulfil demand, I would speculate at least 5 figures. I would expect such boards to comfortably outsell maximus boards e.g.

Enthusiast community needs to be organised to collaborate with vendors. Perhaps 1000 digital signatures in the first place and then contact board vendors? I know that Supermicro designs custom boards if there is enough interest. It's about convincing them.

 

[Mussels]

No. Intel has x8 Gen4 DMI connection bandwidth of 128 Gbps

which are not PCI-E lanes and cannot be assigned to devices, and are all sharing that bandwidth pool.
AMD just built more devices directly into the CPU via the IO die so they dont have those devices running from the chipset.

 

[chrcoluk]

Enthusiast community needs to be organised to collaborate with vendors. Perhaps 1000 digital signatures in the first place and then contact board vendors? I know that Supermicro designs custom boards if there is enough interest. It's about convincing them.

I am almost giggling reading your posts, you really genuinely seem to believe its a rare niche.

 

[Tek-Check]

I am almost giggling reading your posts, you really genuinely seem to believe its a rare niche.

If it wasn't a niche, we would have had more diverse motherboards addressing unmet needs. Clearly, board vendors do not believe that there's a significant gap between top desktop boards and HEDT/WS offers. Or, they do not believe that there is enough people who would buy such boards, so why bother. Their position is clear. They don't offer such boards. Now, the question for you and others who need such boards is what are you folks going to do about it? Complain endlessly on fora or take action to get those boards manufactured?

Board vendors can be wrong, of course, and it is up to enthusiast community to prove them wrong by organising their efforts and collaborating with tech media outlets and board vendors to make this happen. I am sure two Steves from Gamers Nexus and Hardware Unboxed would be willing to take on this idea of semi-HEDT and air the voice of tech community, for starters.

Those who need semi-HEDT boards have a choice. They can cry over it on fora forever making no change in reality, or try to take more serious approach, crowd fund their efforts and find board vendor whose engineers would design such boards to address perceived gap in the market.

It is that simple. Crowd-funding works with thousands of custom products that are not routinely produced by companies for various reasons. It can also work for motherboards. Enthusiasts first need to find out how many boards are needed. This is crucial piece of information in order to know the scale of need. Is it 1000 units, 10,000 units or 50,000 units. This is not trivial. Do you know? Have you done research? What does "niche" mean to you in terms of numbers? You've got to be able to answer these question in order to understand the extent of market need and pick up the phone to call Asus, MSI or any other vendor and share with them your findings. If not, you will never have such a board in your hands.

Tech enthusiasts cannot simply be passive and wait for a pie in the sky to land in their hands. It will not land If they don't do something about it. Anything. That's my point.

 

[TumbleGeorge]

In fact, I believe that there are enough qualified colleagues writing here, a certain number of whom working together, can make the entire motherboard design with the characteristics they agree on, including the hardware and software (bios and drivers). After that, they only have to order the production. That way, the manufacturer would have no reason to charge for R&D.

 

[ADB1979]

But bear in mind we already have board's manufactured for extremely niche use cases which are likely much smaller numbers such as extreme overclocker boards that are not far of 1000USD to buy.

I dont know how board manufacturers find out demand for features (one that I am aware off have a feedback form or feature request form). Their communication with the world is likely only via retailer feedback and the media, and the media is often out of sync with what users might want.

What I do know is I know over 20 people personally who arent happy with PCIe options shrinking, and on top of that 100s of posts read on the internet about similar, so its likely to be much more than a 1000 units required to fulfil demand, I would speculate at least 5 figures. I would expect such boards to comfortably outsell maximus boards e.g. and most definitely APEX boards.

I totally agree with this. I am sure I have heard on of the TechTubers mention that literally 1,000 of some of the overclocking boards are built, so you are certainly correct there.

An ideal for me right now would be:

To move the top M2 slot to the side of the board next to the RAM (as some recent motherboards have done), or much better still, put it on the back of the motherboard but still as close to the CPU as possible for lowest latency.

To actually use the top slot (of 7 potentially available) for the primary x16 slot. (slot 1)

Then 2x slots spaces for M2 drives (empty of PCIe slots), as these are useless for most people with a big GPU (slots 2, 3).

The next two slots to have PCIe 1x slots (slots 4, 5)

The next slot slot to have a physical and electrical x16 slot (slot 6).

And finally another x1 slot (slot 7)

As always with a limited number of PCIe lanes, what slots/M2/SATA can actually be used at once is going to have to be decided by the motherboard manufacturer (there are always some limitations, even on server boards), and the end user.

The second x16 slot would be very useful for someone who wants to use an x16 (4x4) M2 card, an x6 or x16 storage adapter, or a second (thin) graphics card for additional output, with the 1x slots picking up an extra couple of add-in cards. Another option, ( my preferred) would be to swap the last two slots, so the second x16 slot would be at the bottom as many larger cases have an 8th slot anyway, or otherwise breathing room for the 7th slot card.

IMHO, this would fulfil almost all users needs, with the options being limited by the electricals, and the BIOS options. Ideal for many of us who replace an old machine with new, and repurpose the old machine, often with add-in cards as a server, router, video capture etc. This would give all of the options for power-users and gamers, as well as servers, video capture etc.

The last requirement is an absolute MUST, it MUST support Registered ECC RAM.!!! The mid-range and high end AM5 boards from Asrock support registered ECC RAM out of the box, so a perfect "cheap" server or workstation platform.

The graphic is FYI an example of a single slot motherboard with a bit GPU, or a two slot motherboard with only two SATA ports for use as a server or whatever.! This is the AsRock B650 Steel Legend.

Anyway, this doesn't use Slot 1, and the CPU socket has been lowered, so an air cooler would not fit if the actual slot 1 was used. One of the reasons why motherboards like EVGA's often had the RAM at the top and moved all of the power and rotated the socket (Intel platform), obviously workarounds would need to be made.
_________________


A Little example of what can be done with a "server" AM5 motherboard using a B650 chipset and an micro ATX motherboard.!!!

Physical and electrical x4 slot, and x16 slots to the CPU, 1x NVMe to the CPU and a further x1 slot to the chipset. It also includes a remote management chip (has its own graphics, I/O and LAN), 2x Gigabit Ethernet and 2x 10 Gigabit Ethernet.

And of course support Registered ECC.

Also worth noting, because the top slot is an X4 and the placement of the CPU, a big air cooler could be used if the card in the top slot was not too long. This would allow an M2 drive to be used here via an adapter (options.!), and of course it it was ATX there are another 3 slots of potential building on the basic design and layout of this motherboard.

FYI, I have looked at Supermicro and Tyan, they do not have anything to add in their AM5 server/workstation options (currently), although Supermicro has a sick looking 8x AM5 blade server in a 4U chassis, so even the big boys are using the AM5 platform for servers now (supporting Registered ECC was an excellent choice), (Hetzner, (a big German internet services company) has it's own custom AM5 boards, as noted by Der8auer when he wandered around one of their facilities, they also sponsor his channel)), so there are real world examples of AM5 motherboards with all of the features that you could want, if you buy enough of them...

ASRock Rack > B650D4U-2L2T/BCM

www.asrockrack.com

 

[A Computer Guy]

The last requirement is an absolute MUST, it MUST support Registered ECC RAM.!!!

Is there any advantage to supporting registered ECC when only having 4 ram slots and 2 memory channels?

The mid-range and high end AM5 boards from Asrock support registered ECC RAM out of the box, so a perfect "cheap" server or workstation platform.

Which boards are those? I've been thumbing through a few AM5 boards now and then (like the Taichi) and they only declare support for unbuffered ram. ASRock rack AM5 version boards are all UDIMM as well (except the SODIMM ones). I think going forward AMD is going to segment ECC support, like Intel with Xeon, which wouldn't be so much an issue if they would at least sell Pro cpu's in the retail space.

 

[Tek-Check]

In fact, I believe that there are enough qualified colleagues writing here, a certain number of whom working together, can make the entire motherboard design with the characteristics they agree on, including the hardware and software (bios and drivers). After that, they only have to order the production. That way, the manufacturer would have no reason to charge for R&D.

It makes sense, totally. Would someone be kind to propose the spec sheet and draft board layout for smei-HEDT AM5 and Z790?

 

[ADB1979]

Is there any advantage to supporting registered ECC when only having 4 ram slots and 2 memory channels?

Yes.

I am not the person to answer this, but when it comes to "errors" in memory, sometimes they are important and sometimes they are not, if you are average Joe playing a game, you might have an issue that you don't even notice, or it might crash your game or PC, that is largely unimportant. If you are running a server that has multiple users attached, you have the cost of the downtime, plus the cost of loss of work. If it is a workstation, you might lose a 2-hour render. Is it worth it.? That is up to the use case and the end user.

I have taken some interest in the 6200 Registered ECC RAM kits coming from G Skill, although they are aimed at the lame-duck Intel WS CPU's, this proves that fast ECC RAM is not just a possibility, but a reality, this bodes well for the future.

As far as I understand it, the benefits of "Registered" is down to the total RAM that can be used, and the max speed if using 2 DIMM's per channel, I have not seen a single ECC DDR5 DIMM for sale that was NOT registered, so I think that it is now the default for ECC.

R-DIMM Memory - Zeta R5 (Intel XMP) - G.SKILL International Enterprise Co., Ltd.

Designed for workstations, G.SKILL R-DIMM memory is constructed from hand-selected components and rigorously tested for performance at high overclocked speeds.

www.gskill.com

 

[AnotherReader]

Is there any advantage to supporting registered ECC when only having 4 ram slots and 2 memory channels?

Which boards are those? I've been thumbing through a few AM5 boards now and then (like the Taichi) and they only declare support for unbuffered ram. ASRock rack AM5 version boards are all UDIMM as well (except the SODIMM ones). I think going forward AMD is going to segment ECC support, like Intel with Xeon, which wouldn't be so much an issue if they would at least sell Pro cpu's in the retail space.

RDIMMs have higher capacities than unbuffered DIMMs so they have at least one thing going for them. The tradeoff is higher latency.

 

[ADB1979]

Which boards are those? I've been thumbing through a few AM5 boards now and then (like the Taichi) and they only declare support for unbuffered ram.

I saw it with my own eyes back when AM5 first launched, and this has been discussed by Wendel at Level1Techs.

@8:12 Wendel tested this board with Registered ECC, and it works for the X670 Steel Legend. A little later in the video Wendel confirms that Registered ECC also works with the X670 Taichi @11:00

PS: I re-watched Level1Techs video about the AsRock B650E Steel Legend so you don't have to No mention of ECC RAM at all, I guess that it wasnt tested, but as ALL AM5 CPU's support ECC RAM, it is entirely down to the motherboard manufacturer as to whether or not they support it.! In the linked video, Wendel notes (no brand names given) that some AM5 motherboards will function with ECC RAM, but simply do not use the ECC features.!!!

 

[chrcoluk]

If it wasn't a niche, we would have had more diverse motherboards addressing unmet needs. Clearly, board vendors do not believe that there's a significant gap between top desktop boards and HEDT/WS offers. Or, they do not believe that there is enough people who would buy such boards, so why bother. Their position is clear. They don't offer such boards. Now, the question for you and others who need such boards is what are you folks going to do about it? Complain endlessly on fora or take action to get those boards manufactured?

Board vendors can be wrong, of course, and it is up to enthusiast community to prove them wrong by organising their efforts and collaborating with tech media outlets and board vendors to make this happen. I am sure two Steves from Gamers Nexus and Hardware Unboxed would be willing to take on this idea of semi-HEDT and air the voice of tech community, for starters.

Those who need semi-HEDT boards have a choice. They can cry over it on fora forever making no change in reality, or try to take more serious approach, crowd fund their efforts and find board vendor whose engineers would design such boards to address perceived gap in the market.

It is that simple. Crowd-funding works with thousands of custom products that are not routinely produced by companies for various reasons. It can also work for motherboards. Enthusiasts first need to find out how many boards are needed. This is crucial piece of information in order to know the scale of need. Is it 1000 units, 10,000 units or 50,000 units. This is not trivial. Do you know? Have you done research? What does "niche" mean to you in terms of numbers? You've got to be able to answer these question in order to understand the extent of market need and pick up the phone to call Asus, MSI or any other vendor and share with them your findings. If not, you will never have such a board in your hands.

Tech enthusiasts cannot simply be passive and wait for a pie in the sky to land in their hands. It will not land If they don't do something about it. Anything. That's my point.

Nvidia on GPUs have proven consumers dont necessarily get what they want, the boards are sold as is, as its more profitable, and might also be for market segmentation reasons in my opinion, I will accept the demand isnt high enough to force a change from the market, but niche I think there is no evidence for it.

Now days reviews e.g. heavily influence what they do, and I dont think I have seen a review in years complaining about lack of PCIe connectivity. You welcome to show me some kind of market research that has been carried out by the board vendors if you have it to support the niche theory, and especially that it must be more niche than the extreme overclocking boards which likely have tiny amounts of sales. Otherwise its just your opinion, and my opinion.

I almost giggled as when before has a PC customer who now pays a premium for a motherboard has had to crowd fund a reasonable motherboard on the market. I wont be crowd funding anything, its quite simple, if there is nothing on the market that is good enough for me, then I wont buy from the market, luckily an acceptable board exists which I purchased recently, so I didnt have to make that decision.

What people buy isnt necessarily a means of proof either, as in life if people dont get what they want, they compromise and just buy what's there, I am just more picky than others.

Yes.

I am not the person to answer this, but when it comes to "errors" in memory, sometimes they are important and sometimes they are not, if you are average Joe playing a game, you might have an issue that you don't even notice, or it might crash your game or PC, that is largely unimportant. If you are running a server that has multiple users attached, you have the cost of the downtime, plus the cost of loss of work. If it is a workstation, you might lose a 2-hour render. Is it worth it.? That is up to the use case and the end user.

I have taken some interest in the 6200 Registered ECC RAM kits coming from G Skill, although they are aimed at the lame-duck Intel WS CPU's, this proves that fast ECC RAM is not just a possibility, but a reality, this bodes well for the future.

As far as I understand it, the benefits of "Registered" is down to the total RAM that can be used, and the max speed if using 2 DIMM's per channel, I have not seen a single ECC DDR5 DIMM for sale that was NOT registered, so I think that it is now the default for ECC.

R-DIMM Memory - Zeta R5 (Intel XMP) - G.SKILL International Enterprise Co., Ltd.

Designed for workstations, G.SKILL R-DIMM memory is constructed from hand-selected components and rigorously tested for performance at high overclocked speeds.

www.gskill.com

I have always looked at ECC as "insurance". Now days memory is reliable enough that if you run it at spec it should be error free, I cannot remember the last time I have had hold of a faulty memory stick, any instability has always been XMP/OC related. But yeah if you in that 1% or whatever the failure rate is, ECC might save the day, but if ECC does exist on fast memory kits, that is interesting as can ECC overcome overclocking caused errors?

 

[A Computer Guy]

I saw it with my own eyes back when AM5 first launched, and this has been discussed by Wendel at Level1Techs.

Ok I don't want to drift off topic from the OP but will conclude here at least in that video (and a few others I've seen) I did not hear him confirm it was registered ECC so I have a bit of disbelief that Registered ECC is supported. Only unregistered/unbuffered was all I have heard about until your post today.

As far as I understand it, the benefits of "Registered" is down to the total RAM that can be used, and the max speed if using 2 DIMM's per channel, I have not seen a single ECC DDR5 DIMM for sale that was NOT registered, so I think that it is now the default for ECC.

You can find ECC UDIMM on NewEgg.

... but if ECC does exist on fast memory kits, that is interesting as can ECC overcome overclocking caused errors?

Well I can tell you after OC'ing a few ECC kits just a little too far it has reported to me it was correcting errors so that is nice. I've managed (at best) 2 sticks of unbuffered 64GB ECC at DDR4-3600 on an AM4 ITX 2 ram slot motherboard.

 

[Tek-Check]

An ideal for me right now would be:

Does this board with two PCIe switches meet your requirements?