[SOLVED] What one motherboard feature is most important for you when upgrading?

[SHaines]

As builders we tend to weigh the value of having the latest technology without spending your life savings just to get a component that'll last you a while.

I tend to be a last generation type of person, so anything I buy is heavily tested and reviewed, plus it can be picked up for a somewhat reasonable price.

What do you think is the most important motherboard feature available today that will keep your system relevant for years to come (even if you need to spend a bit more to get it)?

We look forward to hearing from you!

 

[refillable]

Really? How many first-gen AM4 boards can actually support a Ryzen 3900X? From the little I've seen, motherboard vendors have tended only to provide BIOS updates supporting about 1 gen newer CPUs than the motherboard's chipset.

And then there's memory speeds to consider. Even if you could run a Ryzen 3900X on a 1st-gen AM4 board, would it hit memory bottlenecks with the fastest RAM speeds supported on that board?

Don't get me wrong - I like that AMD has kept AM4 around longer than Intel's customary 2-generation lifespan, but I'm left with the feeling of a promise not quite being kept.

* Note: I built a 7th PC with a motherboard that had a soldered CPU. So, I'm not counting that one.

You seem to be misinformed. Almost all X370 and B350 boards from the big four MB brands support Ryzen 3000. B450 and X470, even better - aside from minor hiccups here and there. Memory speeds are dependent on the CPU's memory controller, the chipsets don't even touch the CPU-Memory bus. Most promises are kept unless they decided to ditch the socket in this year's Vermeer launch.

 

[bit_user]

You seem to be misinformed.

That's entirely possible. Since I don't have a first-gen AM4 board, I didn't look into the matter very much. I think I just ran across a few mentions of boards not having such firmware updates available, but I'll readily defer to anyone who's an authority on the subject.

Memory speeds are dependent on the CPU's memory controller, the chipsets don't even touch the CPU-Memory bus.

Motherboards qualify memory for use with them, and they're designed & rated to work within a certain frequency range. Furthermore, it's the BIOS that configures the CPU's memory controller, so I have to question whether older boards will really even let you configure newer CPUs to use the fastest memory speeds supported by the CPU, except maybe as an unsupported "overclock" setting.

 

[drea.drechsler]

That's entirely possible. Since I don't have a first-gen AM4 board, I didn't look into the matter very much. I think I just ran across a few mentions of boards not having such firmware updates available, but I'll readily defer to anyone who's an authority on the subject.
....

You may be influenced by the furor surrounding the Ryzen 3000 release. The mobo vendors were hard pressed to update, test and release BIOS' with the AGESA updates as AMD fixed them across what amounted to a fairly large product portfolio of X570, X470, B450, B370 and B350 boards. It may seem logical, but many gave priority to their newest (X570) and then down the pecking order from there...so B350 owners did not feel the love. I can't say for certain, but I do believe A320 owners of many boards were indeed left abandoned for Ryzen 3000 support.

Being one, and a Zen2 early adopter to boot, I can say MSI owners felt it worse so it makes me a bit of an authority. I have a B350 MSI board, we finally got the latest AGESA for it but months after even B450 and X470 owners did (I also have a B450 MSI board) and it's still in BETA, never released.

We were left reading the posts from many a B450 or X470 board owner extolling the virtues of finally getting their Ryzen 3000 to boost as per spec and wondering when we get our turn. That, in turn, made for more than a few roasty burns on the net's directed at MSI engineering and marketing.

 

[refillable]

Motherboards qualify memory for use with them, and they're designed & rated to work within a certain frequency range. Furthermore, it's the BIOS that configures the CPU's memory controller, so I have to question whether older boards will really even let you configure newer CPUs to use the fastest memory speeds supported by the CPU, except maybe as an unsupported "overclock" setting.

Yeah sure, it's never going to be officially supported, but keep in mind that the chipset has nothing to do with the CPU-RAM connection. Potential major deterrents to faster memory speeds are whether the memory controller is able to handle the speed and the noises/interference on the lines of the bus. I've seen many people being able to bump up their RAM speeds just by upgrading the from a Ryzen 1000 to 2/3000 alone. BIOS is a fixable limitation that has to be ironed out by the AIBs. Lower RAM speeds are not an inherent chipset problem.

Unless you're talking about older sockets that still has a northbridge, though.

 

[USAFRet]

"What one motherboard feature is most important for you when upgrading?"

Lack of RGB.
The less the better.

 

[bit_user]

Lower RAM speeds are not an inherent chipset problem.

Right, but I was thinking more along the lines of board layout, design, fabrication, and testing.

Anyway, I concede the point, my primary concern having been addressed by your earlier post and @drea.drechsler.

 

[drea.drechsler]

You can still have Molex and SATA connectors on the PSU.

I see it as a directive for the motherboards more than anything. Only thing that really needs to change on the PSU is the main connector and the standby power voltage.

Wouldn't you need buck convertors on the motherboard to develop 3v and 5v for USB and SATA? Developing the required voltage for the known and predictable load of on-board consumers is a far cry from doing the same for highly variable and unpredictable off-board consumers.

Not to mention the precious real-estate on the motherboard already under stress from the huge number of traces used for PCIe lanes and memory lanes (4 channels!).

I have visions of upgrading motherboards to add a new HDD even when there are ample SATA data ports available. Either that or a major amount of cost and complexity thrown onto the motherboard, so it can handle the huge mix of power loads that could be thrown at it, while taking a small cost out of the PSU.

I think I'd rather keep it in the PSU, the ideal package for the robust components needed to do this sort of thing with ample margins and where it belongs.

 

[lespj]

nvme been around for a while now, how about banishing slow sata and increasing nvme connectivity, less cables less to cause connectivity issues,

 

[USAFRet]

nvme been around for a while now, how about banishing slow sata and increasing nvme connectivity, less cables less to cause connectivity issues,

$$$.
Just like SATA SSD, they will get cheaper. Just not yet.

 

[refillable]

$$$.
Just like SATA SSD, they will get cheaper. Just not yet.

I think they have already supplanted faster SATA SSDs. NVMes such as SN550 and 665p are competitive $/GB-wise to the Samsung 860 EVO, while having at least twice the data rate. I agree with lespj, I think next gen boards should put more emphasis on NVMe and less on SATA. I don't see any reason why there should be any more than 4 SATA connectors.

 

[Peligro Raton]

There is no such thing as a future proof motherboard. 27 years of PC building has taught me that.

 

[msroadkill612]

Z370 is a solid option IMO, even though Intel doesn't support chips over several generations or obviously you could go the AM4 route.

Depending on what he's using his system for, I would start mid-range with something like an i5-8400 and a z370 board or a Ryzen 5 and b350. Both of these set-ups would give you some upgrade options to more cores for the future (Intel has 8c/16t coming to Z370 boards, Ryzen 7 already has 8c/16t on AM4). But honestly any 4-core or more CPU with relatively recent architecture will be up to the task for 95% of use cases.

If it's a gaming system, upgrading the graphics card is the most important thing. You can run older hardware on the newest games if your card is up to par and that would be the only upgrade he would need for years.

it is hard to imagine worse advice than a b350 for future proof AMD.

 

[msroadkill612]

I had an epiphanie recently.

I discovered cheap (~$6 on Ali?) pcie x1 nvme adapter (bridge?) cards.

A recurring point here is anachronistic SATA... it does seem a dogs breackfast for SSDs.

IE. only get sata ssd if u lack an nvme port, & if u invest in ~inferior/obsolete sata, u have ~wasted money.

Most am4 mobos have very limited nvme ports, but seem to have largely unused pcie x1 slots, each yielding 500MB/s on pcie 2 400 series mobos, & either 1000MB/s pcie 3, or 2000MB/s pcie 4 with x570 mobos on a regular 4 lane nvme device.

 

[bit_user]

There is no such thing as a future proof motherboard. 27 years of PC building has taught me that.

It's a relative thing, though.

When I built my Sandybridge-E workstation, NVMe didn't exist. Thankfully, with its abundance of PCIe 3.0 connectivity, I was able to just drop a NVMe 1.0 drive in one of my CPU-connected x4 slots.

Likewise, PCIe 3.0 has been nice for GPU upgrades in that machine, as well.

Finally, due to the quad-channel memory configurataion, it so far stands as the only machine in which I've never upgraded the RAM. I went with 16 GB, from the get go, and that's still working just fine for my purposes.

 

[bit_user]

I don't see any reason why there should be any more than 4 SATA connectors.

I'm not saying every board should have more than 4, but I have a machine with 4x HDDs in a RAID + BD-R for making archival-grade backups of the highest-value stuff. That's 5. The boot drive also happens to be a SATA SSD (because I had a spare one available), so there's 6.

Now, you could argue that I'd do fine with an add-in card, and I don't have a strong argument against it (other than a couple bad experiences with them, in the past). But, if I can just buy a motherboard with 6 SATA ports, then that's what I'll do. In fact, I did have exactly that option and I went for the 6-port board!

 

[bit_user]

A recurring point here is anachronistic SATA... it does seem a dogs breackfast for SSDs.

IE. only get sata ssd if u lack an nvme port, & if u invest in ~inferior/obsolete sata, u have ~wasted money.

NVMe doesn't automatically mean awesome performance. I'd take a Samsung 860 Pro (SATA) over an Intel 600p (NVMe), just about any day. Yeah, the Intel drive has faster peak sequential and IOPS, but is slower in some real-world scenarios. And that's not even bottom-of-the-barrel. There are even worse NVMe drives...

People have this association with NVMe and speed, but to a large extent because the most of the first-gen NVMe drives were also high-end models that sold at a significant price premium.

And you're ignoring the scenario I just mentioned in the post above, where I had an old SATA drive just lying around from a machine that I had already upgraded. So, the SATA drive cost me $0, whereas NVMe would've cost me the drive + at least $6.

Most am4 mobos have very limited nvme ports, but seem to have largely unused pcie x1 slots, each yielding 500MB/s on pcie 2 400 series mobos, & either 1000MB/s pcie 3, or 2000MB/s pcie 4 with x570 mobos on a regular 4 lane nvme device.

Yeah, it makes sense if you have at least one spare slot that's not blocked by a GPU and don't want to use it for anything else (like a multi-gigabit NIC).

It's certainly a worthy option, for some. However, it doesn't suddenly make SATA unnecessary for everyone (not that you said it did).

 

[USAFRet]

NVMe doesn't automatically mean awesome performance. I'd take a Samsung 860 Pro (SATA) over an Intel 600p (NVMe), just about any day. Yeah, the Intel drive has faster peak sequential and IOPS, but is slower in some real-world scenarios. And that's not even bottom-of-the-barrel. There are even worse NVMe drives...

Indeed.
I've said this before in here...
In my actual use case (Adobe Lightroom), my Intel 660p is no faster or slower than the various Samsung 840/850/860 in this system. Close enough to the exact same result.
There are many other things going on rather than just raw drive speed.

The main reason that 660p is in this system is because it was 1TB for $88...

 

[bit_user]

Indeed.
I've said this before in here...
In my actual use case (Adobe Lightroom), my Intel 660p is no faster or slower than the various Samsung 840/850/860 in this system. Close enough to the exact same result.

The 660p is not too bad. Much better than the 600p. If you'd gotten the latter, or some of the more recent low-end drives from other brands, I don't expect you'd be so indifferent.

In fact, according to the benchmarks from the 660p review, even the Crucial MX500 (SATA) convincingly smoked the 600p, in a number of cases.

https://www.tomshardware.com/reviews/intel-ssd-660p-qlc-nvme,5719-2.html

The main reason that 660p is in this system is because it was 1TB for $88...

Intel SSDs also have a pretty good reputation for reliability, and traditionally command a price premium beyond what their performance warrants. So, that was definitely a good deal. I was eyeing the 760p, probably around that time, but decided to hold off.

 

[USAFRet]

Oh, don't get me wrong.
This 660p is great.

Its just that people get all starry eyed over the 3x or 5x Sequential speed over a SATA III SSD, expecting that to be a universal 5x speed improvement.
1.0 sec to open something (SATA III) does not always translate into 0.2 sec (NVMe).

 

[chaz_music]

Good points, though that was a sad thought about M.2 reliability issues.


Exactly. If the OS could query the PSU's capacity and current load, then graphics drivers could limit the GPU to avoid overtaxing the PSU and triggering a random reboot. Instead, they could pop up a notice, telling the user that the GPU performance is being limited by their PSU and suggesting an upgrade.


Back in 2006, you also could build an Intel system with any CPU, and have ECC. At that time, the only requirement was to use a suitable motherboard. This was true until they moved the memory controller into the CPU, in the Nehalem generation.


These are already possible, and even exist, to some extent. You're really just talking about software features & functionality.

And with regard to thermal capacity of CPU cores, this is already a done deal. In AMD CPUs, it's being monitored and the OS is steering work on the basis of which cores have the most thermal headroom. I know some Intel CPUs have certain cores designated for lightly-threaded workloads, but I'm not sure if they do dynamic monitoring and reassessment (though I'm sure they will, in the near future).


I think IOMMUs fixed this. I believe it's no longer the case that any device can simply write to any physical address, though I could be wrong about that.


I've certainly read about how malicious USB devices can fry a host, but you don't get -200 VDC without a big capacitor and a DC-to-DC converter, right? You're not simply talking about a defective USB stick or one that's physically damaged?


Do you think the non-gaming DIY PC community is really that big, or even growing much? I wish I shared your optimism, but all of the long-term PC market trends are headed in the wrong direction, even if the last few years have been decent.

Believe it or not, the malicious USB hack has been around for awhile. There are many SOBs out there:
https://elie.net/blog/security/what-are-malicious-usb-keys-and-how-to-create-a-realistic-one/

People will do the darned things, and not just with PCs. There was a story many years ago about some guys making a trebuchet in their backyard. They heard a noise in the middle of the night - they left it armed, and the trigger failed. Chopped the house in two, as the story goes.

On the DIY front, that is how I got into engineering. I built an IBM XT compatible. Used an NEC V20.

Yes, DIY is still a market but it is not going to be anywhere as profitable as the gaming and overclocking is now. But I would pay good money for a solid home built system built the way I want it. I did that with my NAS. Used a SuperMicro server board and FreeNAS with ZFS. The system idles at 48W with 7x8TB HDDs in RaidZ3 with a spare. Internal copies to an SSD was over 800MB/sec. Too much time on my hands.

 

[chaz_music]

https://en.wikipedia.org/wiki/ATX#ATX12VO

​

I'm not a fan of this, as it makes even more of the system's potential expandability dependent on the motherboard (and further crowds precious board space).

I do like there being a standard for OEM PSUs, however. It will be nice to have easy upgrade steps & mulitple options, when faced with a Dell/HP/Lenovo. But, I really hope the DIY market stays away from this. I prefer to have modularity in the PSU, itself.

The ATX12VO is a great idea from a system stability standpoint. The motherboard manuafcturer is on the hook to make the local lower voltage supplies to fit whatever the CPU needs. This helps future proof the PSU connector style, as any fast di/dt requirements can evolve as the CPUs need it. It also make a 12V UPS solution much easier.

I agree with you that having all of the system power go through the MB is not a good plan. For high power loads, they could just move to a 48VDC solution anyway to cut down on the wire gauge and to allow the main PSU is get more efficient. Maybe even change the whole system to 48V. But then you get into trouble with safety standards like UL and CE ...

 

[chaz_music]

There are many ways for SATA to disappear ... you just need to be innovative .. for example , make HDD all ALL NVME interface by adding onboard 32GB NVME Cache SSD on all xTB harddisks. no one said Mechanical harddisk should disappear , I said SATA !!!

no need to lecture me about "just because you this and that" and stop assuming people dont know about NAS and what ever.

And you could just populate the MB with one of the 4 way NAS type SATA breakout headers and make the header hot in the bios. That way you don't eat up PCB real estate with SATA headers.

 

[drea.drechsler]

The ATX12VO is a great idea from a system stability standpoint. The motherboard manuafcturer is on the hook to make the local lower voltage supplies to fit whatever the CPU needs. This helps future proof the PSU connector style, as any fast di/dt requirements can evolve as the CPUs need it. It also make a 12V UPS solution much easier.
...

That sounds like ATX12V to me. There are 12V buck convertors scattered all over my motherboard supplying necessary power for CPU, memory, chipset, audio CODEC, NIC, and inside the GPU.

All I know +5 and +3.3 are used for is SATA, out the USB ports and legacy COM and LPT ports. Off-board users, one and all. So dropping or changing those voltages really entails changing those consumers. COM and LPT...ok, drop those from modern boards, except in special cases. But is it practical to locate a 12V buck convertor inside a 2.5inch SATA package to develop necessary voltage?

And what about USB devices?

Then there's the SB voltage...to keep memory hot? Not sure. The mnotherboard could do that since the consumer is on-board, but that means a +12V source that's hot all the time. Safety hazard? or another low-current +12VSB rail just for the purpose. What is this accomplishing again?

I just see a big cascade effect trying to go +12VO in desktop systems. If going forward you need to provide these voltages, from a systems engineering perspective it makes far more sense for them to be generated inside the PSU as they are now.

I have no idea what modern systems use -12V for, btw. I'm also curious how it makes 12V UPS easier.

 

[chaz_music]

That sounds like ATX12V to me. There are 12V buck convertors scattered all over my motherboard supplying necessary power for CPU, memory, chipset, audio CODEC, NIC, and inside the GPU.

All I know +5 and +3.3 are used for is SATA, out the USB ports and legacy COM and LPT ports. Off-board users, one and all. So dropping or changing those voltages really entails changing those consumers. COM and LPT...ok, drop those from modern boards, except in special cases. But is it practical to locate a 12V buck convertor inside a 2.5inch SATA package to develop necessary voltage?

And what about USB devices?

Then there's the SB voltage...to keep memory hot? Not sure. The mnotherboard could do that since the consumer is on-board, but that means a +12V source that's hot all the time. Safety hazard? or another low-current +12VSB rail just for the purpose. What is this accomplishing again?

I just see a big cascade effect trying to go +12VO in desktop systems. If going forward you need to provide these voltages, from a systems engineering perspective it makes far more sense for them to be generated inside the PSU as they are now.

I have no idea what modern systems use -12V for, btw. I'm also curious how it makes 12V UPS easier.

Yes! Get rid of the -12V. I agree wholeheartedly. Left over from many years ago. I believe the -12V was for the RS232 port. and some old PCI audio cards, I think.

The cascade effect is correct, but a completely redesigned structure would lower cost and raise reliability. Like when VESA video went to AGP, then PCIe. But yes, there would be a break from legacy power rails for SATA, USB, etc. This is what happened when we went from the 4 pin Molex power connector to the present SATA. Having the 5V and 3.3V made right on the motherboard lowers the chance of poor regulation and voltage drops, especially for long wires and high di/dt loads (inductance).

12V UPS: If 12V is the only power rail that is fed into the motherboard, using an "ideal diode" IC branched from a simple 12V battery makes the UPS function super easy. You would also need a charger. There are several vendors that makes these diode OR ICs now (TI, Analog, Maxim). Using MOSFETs, they decrease the diode forward voltage very low (typically 30mV) instead of the the 0.6-1.0V drop that a high current diode would cause (Schottky and standard rectifier types). There used to be patents against this kind 12V UPS architecture, but I believe they have all expired. Many companies patented these ideas, and then made very few products.

You pointed out the need for sleep power - good point. But you could use the same 12V feed if you wanted. Just make sure there is either di/dt slope limit at power up or have plenty of reserve for a full power load step ($$$).

There is another potential improvement: If the power supply is made smart, it could control the fans direct and receive commands over I2C or any chosen protocol. That keeps the fan power and connectors off of the motherboard and opens up more MB real estate.

 

[drea.drechsler]

...
The cascade effect is correct, but a completely redesigned structure would lower cost and raise reliability.
...

WOW...that's going just a tad further than 'what ONE motherboard FEATURE is most important'. Sounds like all the way to 'what would you do if you could bottom-up redesign the entire personal computer system'.