Motherboard reviews: What matters the most to you?

[Mussels]

It's already leaked out in other parts of the forum, I'm TPU's latest motherboard reviewer.
Part of the changes being brought in is to speed up the reviews - finding ways to streamline the process so we get reviews out faster without compromising the content.

Personally, I think there's a lot of duplicate testing going on with CPU and GPU testing and want to trim that down, but it's a fine line of where to stop.
We already have detailed 14900K and RTX 4090 reviews, how much do we need re-tested every motherboard review?
Don't we need to just test enough to make sure they're not being throttled or held back in some way, instead of re-running every single benchmark and test again?

I'm after feedback and thoughts on this, so that my time is spent where it matters to the readers.


A lot of time goes into the photos and getting the article itself looking good.
Things take a lot longer than you'd assume there - those graphs and charts take manual time and effort although I am working with W1zz on automating more of that.
He's a machine, but I personally don't speak HTML as a second language - just assume that however long you think these things take, at least triple it.


CPU testing and redundancy:
Stock BIOS settings vs power limits changed/removed is obviously important, for both AMD and Intel platforms these days.
That said, we already have the performance values for this from the CPU reviews themselves.

Unless its heavy multi core workloads it doesn't make any difference - if they can handle 125W they can power a 14900K for a gaming only system
Intel Core i9-14900K Raptor Lake Tested at Power Limits Down to 35 W - Minimum FPS | TechPowerUp

Excuse the ugly MS paint, this is only to express the concept without a solid wall of text.
Once you know the wattage a motherboards VRM's can sustain, is TPU's existing testing enough to compare to? If not, why not?
Knowing the wattage a board can sustain lets us know what CPU's run on it problem free, can we simply compare to the measured power consumption of compatible CPUs?
Fictional example below

Intel 12/13/14th gen would be included only, as the compatible CPUs.

With this fictional example above, it'd be easy to judge that as enough for any gaming system - even with a 14900k.
Anyone wanting unlocked power limits instead of stock for multi-threaded workloads, would want to look elsewhere or actively cool those VRMs.

VRM efficiency is part of the reviews, I've already found some waste 20% and up as heat meaning the higher the wattage the worse they do - VRM efficiency is easy to rate and compare separately to temperatures.

Question 1 to you guys:
Is synthetic testing of the VRM's to find a motherboards power limits enough to compare to the existing CPU benchmarks?
If not, what *exactly* do you think should be tested and why - program/game examples needed, and a reason why they need re-testing.


RAM testing is simple since it's not about overclocking.
This is all about the out of the box experience here, not manual RAM overclocking. I've got plenty fast DDR5 to test their limits with.
Once those limits are found, however...

Question 2: AIDA64 is an obvious inclusion, what RAM Bandwidth/latency sensitive games/programs matter the most to you?
With the manual effort of swapping the RAM sticks and recovering from failed boots and such, this really wastes a ton of time so simpler is better


Question 3: What parts of the *chipset* matters to you, for focused testing?
Intel and AMD both include more and more of current systems within the CPU itself from integrated graphics to SATA ports and PCI-E lanes direct to CPU's and M.2 slots - these are not part of the motherboard.
Modern chipsets are almost just PCI-E switch to spread the bandwidth between lower priority components, and it seems that very few people care about things like networking and onboard audio since it's easy to find reviews on those specific components elsewhere, or outright replace them.
I do intend to test USB 3.2/4.0 performance with external NVME drives where possible, but USB 4.0 is so new they're hard to get access to at this time.
NVME testing will be done to compare CPU lanes vs chipset to look for a bottleneck.

 

[unwind-protect]

Testing that Thunderbolt, if present, actually works.

 

[AusWolf]

Not really an answer to the detailed questions, but more to the thread title: what matters to me is...

1. Board layout,
2. Build quality, with special emphasis on VRM quality, temperatures, durability, etc.,
3. BIOS quality, usability, menu layout, quirks, etc.

Performance differences are miniscule, and overclocking is basically dead these days, so I don't care about these.

 

[ir_cow]

I'm always for maximum abuse. MB vendors on the Intel side are sneaky. Even though the 14900K CPU might have a TDP of 250, it will automatically be set to 4096 many of times. Meaning unlimited power to boost to the Turbo limit or hit the temperature limit. Whichever is first.

When vendors have the BIOS auto select your power limit based on cooling, it can lead to problems. You can always manually change that, but a CPU pulling 250 vs 340-380 is a bit different on the VRM. Low end MBs just can't handle that passively. Worth a test I think.

 

[Frick]

As an introduction I would like a deep dive-ish article into why motherboards matter, beyond the obvious stuff (layout, VRM's and features). Personally I just go for the cheapest one that looks like I want (PCIe 1x on top).

 

[Shrek]

I agree about 'don't bother with overclocking'

However, DDR5 has some error correction and it would good to know if that mechanism ever gets triggered, although this may not be a motherboard thing.

VRM limits are a nice thing, but not sure how this can be tested.

 

[unwind-protect]

However, DDR5 has some error correction and it would good to know if that mechanism ever gets triggered, although this may not be a motherboard thing.

The DDR5 built-in ECC cannot be reported. The errors never leave the module. You still need real ECC memory to know whether one of your memory modules is going bad.

 

[Lost_Wanderer]

Layout, CPU cooler-graphics card compatibility As in which orientation(s) for big/wide air coolers like the D15 or U14 and whether or not it will overhang the first full PCI-E slot.


I know of people that won't do enough homework.

 

[HD64G]

Connectivity, RAM stability, UEFI layout and potential, warranty, VRM and power delivery.

 

[ir_cow]

VRM limits are a nice thing, but not sure how this can be tested.

Same as before, list the parts, do the math and jam 340~ into the CPU while measuring the VRM temps.

 

[evernessince]

I feel like TPU's motherboard reviews are already a bit skinny in regards to charts but if I had to cut somewhere it'd be the gaming section for sure. No need to test at both Medium and High for the motherboard or that many games. I'd merge performance into a single page and just have 10 or total applications with 2 to 3 of them being games tested at a single setting. I do wish there was a bit more emphasis on word economy, as in reduction of the amount of words. You can often say more with less IMO. The contents and packaging page doesn't need a description at all, just pictures and a list of inclusions is fine.

Probably the biggest time sink is anything memory related. I do believe that having stock and sweet spot is important but anything outside of that IMO is teiriary and can be done in a separate memory OC MB roundup.

What parts of the *chipset* matters to you, for focused testing?

This is a tough one because it really varies on the board. The focus of a budget board is going to be different that an entusiast one. For budget boards it's more important that the core features work whereas if a PCIe M.2 slot gets lower bandwidth it's not the end of the world and a sacrifice that usually has to be made in that segment.

Really it's about thinking who is going to buy the board and then tailoring that review to those people so they have the information they need. For example, for a work oreinted motherboard you could probably go without OC memory tests completely and just report VRM thermal at stock. Not as comprehensive of course but you are putting resources where they are most likely to be useful.

A lot of time goes into the photos and getting the article itself looking good.
Things take a lot longer than you'd assume there - those graphs and charts take manual time and effort although I am working with W1zz on automating more of that.
He's a machine, but I personally don't speak HTML as a second language - just assume that however long you think these things take, at least triple it.

Does TPU not have a WYSIWYG system for publishing content?

 

[chrcoluk]

I agree with removing testing duplication (games, app benches etc.), I actually think there is very little need for performance testing on boards.

Here is what I would like to see, this is in no particular order.

1 - Testing of BIOS functionality to see how well the BIOS actually works, things misbehaving, or not working at all, tjmax in spec, voltage fed to CPU vs other boards, that sort of thing. Perhaps overkill to test every single option, but at least try to test the ones most likely to be tinkered with.

2 - Basic load stress tests, so this would cover testing VRM's, but no need to run game benches and the like, you testing how hot they get and if system is stable.

3 - Testing the process of install and removing things with the board in a case, how accessible are the fan headers, the M.2 ports, the layout, is it toolless, quality of the heat spreaders, that sort of thing.

4 - Detailed analysis of connectivity on the board, PCIE lane allocation, SATA ports. M.2, PCIe slots, USB ports, onboard power/reset, bios flashback etc.

5 - Test if the board is causing a clock wall on RAM, especially 4 dimm slots.

All this is far more important to me than if it gives an extra couple of frames in games.

I think this would make me a very happy recipient of the reviews, and with no performance benches should save you considerable time (they all going to perform practically the same unless they have serious deficiencies which would be found out in the stress, bios and ram testing). Hope you read this post. Thanks.

 

[adilazimdegilx]

1. VRM temps and throttling testing is important. Especially on midrange and budget motherboards. On AM5 side there are bad mobos and some budget ones have trouble running higher TDP CPUs. On highend though, I guess you can only compare temps and try to throttle them with high TDP loads. HWUs and GNs motherboard tests are usually sufficient, maybe you can use them as guide too (about what to and how to test)

2. AIDA is very obvious although testing RAM on actual programs and games are hard. HWU does good job including and comparing RAMs time to time. But it is individual tests for many games so it's very hard to do on large scale. I dont think motherboards OC capabilities on RAMs are really important. They are more or less the same and it is mostly dependent on the RAM kits.

3. Honestly, I dont really care about performance of outside ports. NVME and PCI slots just work within specs. Number of ports and types are important but I dont think they require testing.

 

[Mussels]

I'm always for maximum abuse. MB vendors on the Intel side are sneaky. Even though the 14900K CPU might have a TDP of 250, it will automatically be set to 4096 many of times. Meaning unlimited power to boost to the Turbo limit or hit the temperature limit. Whichever is first.

When vendors have the BIOS auto select your power limit based on cooling, it can lead to problems. You can always manually change that, but a CPU pulling 250 vs 340-380 is a bit different on the VRM. Low end MBs just can't handle that passively. Worth a test I think.

I definitely care about the stock settings, that was often broken on AM4 boards and resulted in their confusing reputation

On every AM4 board i've tested from gigabyte over the years, disabling PBO actually disabled the limits - the opposite to what users think and results in instant 100c temps and throttling/crashing.
Every other brand, would cap you within the expected PBO values

 

[Belfaborac]

1 - Testing of BIOS functionality to see how well the BIOS actually works, things misbehaving, or not working at all, tjmax in spec, voltage fed to CPU vs other boards, that sort of thing. Perhaps overkill to test every single option, but at least try to test the ones most likely to be tinkered with.

2 - Basic load stress tests, so this would cover testing VRM's, but no need to run game benches and the like, you testing how hot they get and if system is stable.

3 - Testing the process of install and removing things with the board in a case, how accessible are the fan headers, the M.2 ports, the layout, is it toolless, quality of the heat spreaders, that sort of thing.

4 - Detailed analysis of connectivity on the board, PCIE lane allocation, SATA ports. M.2, PCIe slots, USB ports, onboard power/reset, bios flashback etc.

5 - Test if the board is causing a clock wall on RAM, especially 4 dimm slots.

^This, and stuff like this:

MB vendors on the Intel side are sneaky. Even though the 14900K CPU might have a TDP of 250, it will automatically be set to 4096 many of times. Meaning unlimited power to boost to the Turbo limit or hit the temperature limit. Whichever is first.

When vendors have the BIOS auto select your power limit based on cooling, it can lead to problems. You can always manually change that, but a CPU pulling 250 vs 340-380 is a bit different on the VRM. Low end MBs just can't handle that passively. Worth a test I think.

On every AM4 board i've tested from gigabyte over the years, disabling PBO actually disabled the limits - the opposite to what users think and results in instant 100c temps and throttling/crashing.
Every other brand, would cap you within the expected PBO values

 

[lemonadesoda]

STABILITY.

POWER CONSUMPTION.

I’d like to see some form of mb stability test and voltage regulator test with under/over voltage and noise on the power rails. I do understand that it’s the PSU job to get most of that work done. But I’d like to know how robust the voltage regulator circuitry is on the mainboard. So adjustable voltage PSU needed and some noise.

premium manufacturers like to brag about their capacitors and VRMs. Nobody tests it.

If you dont know how to test it, or don’t have the tools, then at least have a review section on it, and for the moment, list the components used, incl. total capacitor capacity and other relevant metrics. Need ti talk with an expert, I’m not that person

 

[Outback Bronze]

I'm TPU's latest motherboard reviewer.

Congrats! Look forward to reading your reviews. Wish you well

 

[Assimilator]

Not answering your questions directly, but I want to see a stronger focus on connectivity and PCIe lane allocation.

Lack of USB4 and/or Thunderbolt 4 in nearly 2024 is something that should be called out both in the review body and as a negative in the conclusion. I really don't care if USB4/TB4 add-on chips don't exist, or there isn't enough supply, or whatever - all we ever hear from these companies are excuses, if you start dinging them repeatedly and pointedly it will hopefully light a fire under their collective lazy asses. Similarly we're still seeing boards come with only a single rear USB-C port, often not even capable of DP Alt mode, which is just inexcusable rubbish in this day and age.

This stupidity whereby mechanically x16 slots - that are only capable of x4 or even x2 - are placed on boards, should be called out as the deliberately deceptive bullshit that it is. If you're offering x4 or x2, USE A MECHANICALLY x4 SLOT.

I'd like the platform vendors (AMD and Intel) to be called out for the deliberately arbitrary and market-segmenting bullshit of permitting certain PCIe lanes to be used only for certain things. The prime example is AMD requiring 4 of its AM5 platform's PCIe 5 lanes to be dedicated to an NVMe drive, in a market where almost nobody owns or even wants to own a PCIe 5 SSD. I was under the impression that the whole point of making more PCIe lanes available was to allow motherboard designers to come up with a plethora of different models that satisfy the needs of as many different groups of consumers as possible; forcing certain lanes to be used for certain functions achieves the exact opposite. If all 24 lanes of PCIe 5 could be allocated to PCIe slots, that would enable boards with an x16 and x8 slot, which would be extremely useful for those of us with PCIe NVMe cards by allowing 4x NVMe SSDs in the x16 slot and a GPU in the x8 slot (yes I know you'd lose some GPU performance but HAVING THE OPTION TO CHOOSE TO DO THAT is the point).

And finally, I'd like the general lack of innovation in the motherboard space to be called out, hard. It doesn't matter if a board has 7000A chokes and 20 million power stages and can handle 1.21 jigawatts of power through the CPU, NOBODY CARES BECAUSE EVERY OTHER BOARD CAN DO THE SAME. So board manufacturers need to stop marketing this irrelevant bullshit as if it's an important factor, because it isn't. Remember when Asrock did DDR2 and DDR3 memory slots together on their boards? THAT was innovation, THAT is the type of thing that should be highlighted, not this penis-measuring contest of who can build the most unnecessary VRM.

 

[AusWolf]

This stupidity whereby mechanically x16 slots - that are only capable of x4 or even x2 - are placed on boards, should be called out as the deliberately deceptive bullshit that it is. If you're offering x4 or x2, USE A MECHANICALLY x4 SLOT.

Not necessarily. There are cards that use a physical x16 slot, but are only wired in with 4 lanes. A physical x4 slot would prevent you from using such cards.

As for USB 4.0, I've yet to see a single board that supports it.

Other than that, I agree.

 

[Assimilator]

Not necessarily. There are cards that use a physical x16 slot, but are only wired in with 4 lanes. A physical x4 slot would prevent you from using such cards.

The number of cards that use mechanical x16 with electrical x4 is far smaller than the number of cards that use mechanical + electrical x4. And open-ended x4 slots were developed for precisely the first use-case.

 

[A Computer Guy]

The DDR5 built-in ECC cannot be reported. The errors never leave the module. You still need real ECC memory to know whether one of your memory modules is going bad.

Someone replied about this very topic somewhere in the forums and showed there was a way to get additional information from DDR5. I'm going to check if I book marked that info.
(edit)
Well this is what I found https://www.techpowerup.com/forums/threads/memtest64-or-memtest86.305648/post-4973542 following the link to https://www.memtest86.com/tech_DIMM_Decoding.html and it seems to me Memtest86 is just calculating the chip that had an error rather than grabbing some data from the module so I think it's still fair to say the DDR5 on-chip ECC cannot be reported.

Not necessarily. There are cards that use a physical x16 slot, but are only wired in with 4 lanes. A physical x4 slot would prevent you from using such cards.

There is a flipside too such as open ended pci-e slots. Theoretically you can fit a longer card in the slot but other motherboard components or heatsinks are in the way negating the ability to actually fit a longer slot card. Example: I've see ASRock do this alot on their boards.

...trimmed quote....

An issue with reporting VRM temperatures is this might vary from case to case. Poor airflow cases will suffer more so I think if you report VRM temps you need to describe the conditions for the temp measurement such as "when measured on an open-air test bench with no air-flow or with air-flow using fan at xxx rpm's 8cm away".

 

[evernessince]

I'd like the platform vendors (AMD and Intel) to be called out for the deliberately arbitrary and market-segmenting bullshit of permitting certain PCIe lanes to be used only for certain things. The prime example is AMD requiring 4 of its AM5 platform's PCIe 5 lanes to be dedicated to an NVMe drive, in a market where almost nobody owns or even wants to own a PCIe 5 SSD. I was under the impression that the whole point of making more PCIe lanes available was to allow motherboard designers to come up with a plethora of different models that satisfy the needs of as many different groups of consumers as possible; forcing certain lanes to be used for certain functions achieves the exact opposite. If all 24 lanes of PCIe 5 could be allocated to PCIe slots, that would enable boards with an x16 and x8 slot, which would be extremely useful for those of us with PCIe NVMe cards by allowing 4x NVMe SSDs in the x16 slot and a GPU in the x8 slot (yes I know you'd lose some GPU performance but HAVING THE OPTION TO CHOOSE TO DO THAT is the point).

I'm with you on creativity but the scenario you are describing is very very very niche. The person has to first use up all their motherboard's NVMe slots, which can go as high as 5 on AM5. Then they have to need an additional 4 NVMe drives that they want to each run at full speed simultaneously to justify need an 16x NVMe adapter. If they aren't maxing them all out at once there are x4 and x8 cards that can dynamically allocate all the bandwidth to any one SSD, thus getting full speed on any of the NVMe drives even in an x4 slot. If the person just wants size or endurance, enterprise class drives can be adapted to any M.2 or PCIe slot (so long as it isn't x1) with capacities of up to 30TB. If the person wants additional high speed m.2 slots, they can get that in an x8 or x4 adapater, up to 8 in fact on some of the very expensive $2,000+ highpoint cards. There are really not many people laying out that kind of money just for the NVMe card though.

 

[Assimilator]

I'm with you on creativity but the scenario you are describing is very very very niche. The person has to first use up all their motherboard's NVMe slots, which can go as high as 5 on AM5. Then they have to need an additional 4 NVMe drives that they want to each run at full speed simultaneously to justify need an 16x NVMe adapter. If they aren't maxing them all out at once there are x4 and x8 cards that can dynamically allocate all the bandwidth to any one SSD, thus getting full speed on any of the NVMe drives even in an x4 slot. If the person just wants size or endurance, enterprise class drives can be adapted to any M.2 or PCIe slot (so long as it isn't x1) with capacities of up to 30TB. If the person wants additional high speed m.2 slots, they can get that in an x8 or x4 adapater, up to 8 in fact on some of the very expensive $2,000+ highpoint cards. There are really not many people laying out that kind of money just for the NVMe card though.

By that argument, discrete GPUs are niche.

 

[dragontamer5788]

The number of cards that use mechanical x16 with electrical x4 is far smaller than the number of cards that use mechanical + electrical x4. And open-ended x4 slots were developed for precisely the first use-case.

Almost all GPUs use mechanical x16 but support electrical x4, x8, and x16.

The x16 mechanical is designed to be compatible with x1, x2, x4, x8, and x16 mechanical. And IIRC, x4 mechanical is designed to be compatible with x1 and x2 as well. As such, we only ever see x1, x4, and x16 mechanical despite the existence of x2 and x8.

For maximum compatibility, everything should be a mechanical x16 slot and then documented elsewhere whether its x1 or x16. But that's not cost-efficient.

 

[evernessince]

By that argument, discrete GPUs are niche.

You didn't really read what I wrote if that's your conclusion. If you are going to reply at least explain your logic.