X670E Motherboard: Are VRMs all that different?

[The Von Matrices]

I'm looking into various X670E motherboards for a new AM5 system. Even the cheapest motherboards have a dozen power phases, which, to me, seems like overkill for a socket whose hottest chip has a 170W TDP. But when I read reviews of the various motherboards, these reviewers imply that these 14 phase VRMs on the ~$300 motherboards are only adequate for running processors at stock speed and you need to buy $500-$1000 motherboards to run even with a mild overclock.

To me this seems absurd that reviewers are saying that a 14-phase VRM is too basic. Am I missing something here? Would spending a lot more on a motherboard actually make a difference in overclocking for a system that will never see LN2? I'd appreciate the insight.

 

[oxrufiioxo]

Although any of the X670E boards will run any Ryzen CPU no problem the phase count is irrelevant it's more about the quality of the components used in the VRM. Most boards run two power stages in each phase there are not very many real 14 phase boards. Even the overkill X670E hero is really a 9 phase design.

 

[tabascosauz]

I'm looking into various X670E motherboards for a new AM5 system. Even the cheapest motherboards have a dozen power phases, which, to me, seems like overkill for a socket whose hottest chip has a 170W TDP. But when I read reviews of the various motherboards, these reviewers imply that these 14 phase VRMs on the ~$300 motherboards are only adequate for running processors at stock speed and you need to buy $500-$1000 motherboards to run even with a mild overclock.

To me this seems absurd that reviewers are saying that a 14-phase VRM is too basic. Am I missing something here? Would spending a lot more on a motherboard actually make a difference in overclocking for a system that will never see LN2? I'd appreciate the insight.

There are a lot of factors that influence actual VRM thermal performance, that it can only be reliably determined on a case-by-case review basis..................that said, whoever they are those reviewers are idiots. Even with DrMOS parts (really low bar given prevalence of cheap 50A DrMOS), any 14-phase VRM will not struggle with anything.

The "true" phase count doesn't matter when it comes to current handling capacity. Power in a parallel 7 phase is still going through 14 power stages. Even the cheapest X670 boards (Asus Prime, ASRock Pro-RS) are what, 14 x 50A? 12 x 60A?

On AM4 (assuming a decent sized heatsink and no weird anomalies in power plane design, etc.), an 8 x 50A Vcore will handle any AM4 CPU, PBO or not, with plenty of headroom to spare.

200W+ draw is only new as stock power draw Ryzen. Intel CPUs have been doing 200W+ for generations, since 9th gen you can push them way harder than Ryzen. Go look at any midrange or higher Z490/Z590 board, the VRMs on most of those already intrude into overkill territory.

Besides, exactly what sort of overclock are you looking to do on a stock 230W part.........? The 230W CPU parts are already maxed out and way into inefficient land, people are only keeping or reducing the power and using CO to improve performance. Assuming you're after a 7950X, you will struggle far sooner with the 7950X's own heat density when OCing past 230W than you will run into VRM problems.

 

[MachineLearning]

For pure CPU OC I sincerely don't think there is a significant difference, barring some theoretical major mistake these new VRMs are excellent (albeit overkill).

The biggest difference between AM5 mobos for OC will almost definitely be memory topology and 2DPC vs 4DPC. Especially since it's hard to tell how far these boards will be able to push the DDR5 of tomorrow and its wicked high frequencies.

 

[oxrufiioxo]

For pure CPU OC I sincerely don't think there is a significant difference, barring some theoretical major mistake these new VRMs are excellent (albeit overkill).

The biggest difference between AM5 mobos for OC will almost definitely be memory topology and 2DPC vs 4DPC. Especially since it's hard to tell how far these boards will be able to push the DDR5 of tomorrow and its wicked high frequencies.

I agree, the only reason to spend more on a board is if you know for sure it offers better memory support. Assuming all other needs/wants are met.

 

[ir_cow]

From my personal experience you can have lots of power stages and still have a bad experience with overclocking and just stock operations. The amount of power stages only helps up to a certain point, after which is more marketing than anything else. More power stages does help with the overall VRM temperature as the load is distributed among the many. But this can also be a problem because let say (making this up here). 105 A best efficiency is at 30% load. You have 16 of them (1680 A in total) but 31.5A is optimal for them. That would be 504A. If the CPU is pulling 160A at its peak from the turbo. Its still only 31%, which well below the limit.

What happens here is if you are a manufacturer, putting effort already into more PCB layers, better memory support (ie better traces and more copper) and 16 phases, might has well go to 20 or 22. I don't think the cost from them goes up much and it looks better on paper. If you took those 14x 50A and did the 30% on those are well you are left with 210A, which is good enough. The problem is now the MB is missing a ton of features, lower PCB layer and poor memory support. Oh don't forget for whatever reason, the less power stages you have, the shitter the heatsink is. It will hit 100c easily.

 

[The Von Matrices]

Thank you so much everyone for the info!

I don't expect overclocking to actually yield much gain but I like the ability to tinker just to satisfy my curiosity. I didn't think about memory overclocking when looking at the motherboards but it would make sense that the more expensive boards could have better signals. I'll have to look for that more closely. Sounds to me like I'll be happy with something towards the lower end of the price range (if you can call $300-$400 "low end").

Thanks again!

 

[Kapone33]

Thank you so much everyone for the info!

I don't expect overclocking to actually yield much gain but I like the ability to tinker just to satisfy my curiosity. I didn't think about memory overclocking when looking at the motherboards but it would make sense that the more expensive boards could have better signals. I'll have to look for that more closely. Sounds to me like I'll be happy with something towards the lower end of the price range (if you can call $300-$400 "low end").

Thanks again!

In my experience MSI current MBs are the best for mesing around with Memory OC. They have a Mem Try it function in the BIOS that has like 40 profiles and will even boot most of them into Windows but you have to monitor for WHEA errors.

 

[ir_cow]

I forgot to mention while that example I showed was for all power stages active, they don't have to be all on at once. ASUS has a option (at least on HERO and above) to have power stages all on, active based on current load and based on tempature. So for example, with 20 stages, maybe only 10 are on at once. You would have to take a oscilloscope to really see whats going on.

One of the many things that make the whole VRM "overkill" argument hard to make black and white.

 

[tabascosauz]

I don't expect overclocking to actually yield much gain but I like the ability to tinker just to satisfy my curiosity. I didn't think about memory overclocking when looking at the motherboards but it would make sense that the more expensive boards could have better signals. I'll have to look for that more closely. Sounds to me like I'll be happy with something towards the lower end of the price range (if you can call $300-$400 "low end").

That part about memory OC has always been true - e.g. if you wanted to let loose your 5700G on pure frequency with some good Micron or Hynix, you'd choose a 2DIMM board with a good topology on a 6- or 8-layer PCB, and not a cheap whichever 4-layer 4DIMM board. There's just more importance on signal integrity now with higher DDR5 speeds, and board makers are beefing up the layer count across the board due to both DDR5 and new PCIe versions.

I forgot to mention while that example I showed was for all power stages active, they don't have to be all on at once. ASUS has a option (at least on HERO and above) to have power stages all on, active based on current load and based on tempature. So for example, with 20 stages, maybe only 10 are on at once. You would have to take a oscilloscope to really see whats going on.

One of the many things that make the whole VRM "overkill" argument hard to make black and white.

Is this a new Z690 or X670 feature? I remember seeing something related to Phase Control in the Digi+ section of my Impact, but never bothered since it's a parallel 4-phase Vcore lol and still a IR35201

I remember hearing a bunch of times back when smarter high phase count controllers started appearing (ie. Infineon XDPE) that they are smart enough to adjust phase control on the fly for better efficiency. Because running 14 phases with TDA21490 at idle would just be idiotic. But my Unify-X doesn't seem to allow user control. Or maybe I need to go back and snoop in the BIOS.

 

[ir_cow]

@tabascosauz I've seen in on the ASUS X570 Hero. Not sure about anything before that.