Be careful when recommending B560 motherboards to novice builders (HWUB)

[Valantar]

HWUnboxed just posted a pretty interesting video on how OOB performance varies across B560 boards with 65W 11th-gen Intel CPUs. While all of is is within Intel spec, performance in sustained all-core workload varied by over 40% on the 11700 and over 30% on the 11400F. Gaming performance was more even, but still varied by double digit percentages. The gist of it is that higher end (~$200) B560 boards run the chips without active power limits OOB, while cheaper boards enforce them strictly - which again introduces variable performance due to variations in voltage tuning etc. Even between the cheaper B560 boards with enforced power limits there were notable performance differences. These boards do allow for disabling power limits, though for two of the three boards tested this resulted in VRM power throttling on the 11700, causing intermittent hard throttling (below spec, 800MHz for one, 2GHz for the other). That's going to give a juddery and terrible experience, and alleviating it requires adding more VRM cooling (if at all possible).

So, given just how excellent the value proposition of this platform is overall, it's definitely worth ensuring that buyers know what they're getting into. Most CPU reviews even of the 11400 is likely done on unlocked Z590 platforms, so users can potentially see significantly lower performance than expected.

They promised a larger round-up of B560 boards coming up focusing on this, which should provide a pretty decent starting point for making recommendations.

 

[Zyll Goliat]

Yeah....That's not good at all Intel

 

[W1zzard]

The gist of it is that higher end (~$200) B560 boards run the chips without active power limits OOB, while cheaper boards enforce them strictly

Surprised that this is news to anyone

 

[Valantar]

Surprised that this is news to anyone

Whether it's news or not is one thing, the fact that it brings 30-40% performance differences between boards with nominally similar features is definitely not normal. Intel's TDP calculations really haven't withstood the change from 4c8t to 8c16t or higher very well at all. That's the main change after all, as the new 65W CPUs have dramatically lower base clocks than previous lower core count chips, leading to major performance drops for boards that don't allow them to turbo indefinitely. It seems that as more time passes, the range of what is "in spec" just keeps increasing. As they point out in the video, a 7700K had a 17% delta between base and boost clocks, while the 11700K has a 96% delta between base and turbo. If the range of possible "in-spec" performance is literally 1x-2x, that spec is pretty meaningless.

 

[W1zzard]

is definitely not normal

This IS definitely normal for Intel. Check my non-K CPU reviews over the last generations. What you want is the most power-limit castrated CPU at best pricing, and then give it unlimited power limit. No need for OC at all

The underlying reason for those lame TDPs is that some (shitty) OEM systems are designed to only handle 65 W in terms of cooling and power.

 

[Valantar]

This IS definitely normal for Intel. Check my non-K CPU reviews over the last generations. What you want is the most power-limit castrated CPU at best pricing, and then give it unlimited power limit. No need for OC at all

While it's true that you've been covering it (since 10th gen it seems - there are no 9th gen 65W parts reviewed on TPU that I can find, and the i5-8500 review makes no mention of adjusting power limits from what I can tell), but your i5-10400F review shows a 3.5% uplift in Cinebench MT and a 2.1% increase in gaming at 720p - including a 3% BCLK adjustment. So it hardly matches the increases seen here, that's for sure. The 10500 bumps that to 7% in CB MT, but again ... that's not 30, let alone 40.

One relevant question here though is whether your CB testing is a single run, an average or several, or a single run after a given warm-up time of looping runs. If it's just a single run that would go some way towards showing why your tests show relatively minimal changes compared to HWUB's results.

The underlying reason for those lame TDPs is that some (shitty) OEM systems are designed to only handle 65 W in terms of cooling and power.

As for this ... that's a consequence, not a reason. The reason for the 65W TDP is history and PR. Sure, OEM expectations based on this history is a large part of why it's maintained, but if Intel changed their TDPs, OEMs would have no choice but to adjust accordingly (and there would likely be cTDP-down modes for SFF OEM desktops). IMO the main reason for Intel not changing TDPs to keep sensible base-to-boost ratios for their chips is that this would look really embarassing for them compared to Ryzen. They were likely holding out on making changes for a few years in hopes that 10nm and 7nm would pan out, but when that didn't happen they've just stuck with it to not lose face. Pinning Intel's design cop-outs on their OEM partners sounds like a misrepresentation of the power balance in those relations.

 

[newtekie1]

This IS definitely normal for Intel. Check my non-K CPU reviews over the last generations. What you want is the most power-limit castrated CPU at best pricing, and then give it unlimited power limit. No need for OC at all

The underlying reason for those lame TDPs is that some (shitty) OEM systems are designed to only handle 65 W in terms of cooling and power.

It isn't like this is only an Intel problem. There are some B550 motherboards out there that can't handle a fully loaded 5800X either.

I think every major manufacturer is guilty of putting out an absolute garbage VRM budget B550/B560 motherboard. We've moved into the era where even if you are just going to run the system "at stock" spending, even a little bit, more on the motherboard can make a difference.

 

[Chomiq]

News:

 

[Valantar]

It isn't like this is only an Intel problem. There are some B550 motherboards out there that can't handle a fully loaded 5800X either.

I think every major manufacturer is guilty of putting out an absolute garbage VRM budget B550/B560 motherboard. We've moved into the era where even if you are just going to run the system "at stock" spending, even a little bit, more on the motherboard can make a difference.

What's a "fully loaded" 5800X? I've never seen AMD motherboards deviate from stock power limits the way Intel boards often do, at least at stock. But then I haven't been reading that many motherboard reviews. A 5800X boosts up to ...138W or something? Not quite the 144 of a 5900X and 5950X at least. Are there boards that can't handle this stock boost and will throttle due to VRM thermals?

News:

This is a rather different angle though. Those videos cover K SKUs and Z-series motherboards. This is about supposedly locked-down non-K 65W SKUs on non-OC chipset boards. One should be able to expect relatively consistent performance given these things, but ... well, apparently not.

 

[newtekie1]

What's a "fully loaded" 5800X? I've never seen AMD motherboards deviate from stock power limits the way Intel boards often do, at least at stock. But then I haven't been reading that many motherboard reviews. A 5800X boosts up to ...138W or something? Not quite the 144 of a 5900X and 5950X at least. Are there boards that can't handle this stock boost and will throttle due to VRM thermals?

Simple, you pop a 5800X in, turn on PBO, and load it up fully with whatever load testing program you prefer(Prime95, OCCT, Linpack, etc.). Yes, it will power throttle in some B550 boards.

I'm not sure where you are getting your power numbers, but a 5800X with PBO OFF(which is AFAIK how the chips are tested here at TPU) will pull 160w. And a 2700X is rated at 105w and will pull 195w(that's more of a deviation than the 10900K and 11900K). The 5950x is also rated at 105w but will pull 195w as well(again more of a deviation than Intel). Sorry, AMD isn't squeaky clean here.

If you put any of those processors in a AsRock B550M-HDV or an MSI B550M-A Pro, for examples, you're going to see VRM throttling.

The fact is, it doesn't matter what platform you are on, if you buy a shit budget board and put a processor that pulls 150w+ under load, you're going to have a bad time. The days of "the motherboard doesn't make a performance difference" are long gone.

Plus you can get a decent H570 motherboard for $130-140, and B560 boards start at about $110(the "budget" AsRock B560 they talk about in the original video is $115). I honestly struggle to see why anyone would even consider B560 when the leap in price to H570 boards is so small. And really, the same applies to B550 and X570. Any decent B550 is going to cost you $110-115, and a good X570 is $140. I don't see a point in saving $20-30 on a part that is the central nervous system of the computer by going down an entire product tier.

 

[tabascosauz]

Simple, you pop a 5800X in, turn on PBO, and load it up fully with whatever load testing program you prefer(Prime95, OCCT, Linpack, etc.). Yes, it will power throttle in some B550 boards.

I'm not sure where you are getting your power numbers, but a 5800X with PBO OFF(which is AFAIK how the chips are tested here at TPU) will pull 160w. And a 2700X is rated at 105w and will pull 195w(that's more of a deviation than the 10900K and 11900K). The 5950x is also rated at 105w but will pull 195w as well(again more of a deviation than Intel). Sorry, AMD isn't squeaky clean here.

If you put any of those processors in a AsRock B550M-HDV or an MSI B550M-A Pro, for examples, you're going to see VRM throttling.

The fact is, it doesn't matter what platform you are on, if you buy a shit budget board and put a processor that pulls 150w+ under load, you're going to have a bad time. The days of "the motherboard doesn't make a performance difference" are long gone.

Plus you can get a decent H570 motherboard for $130-140, and B560 boards start at about $110(the "budget" AsRock B560 they talk about in the original video is $115). I honestly struggle to see why anyone would even consider B560 when the leap in price to H570 boards is so small. And really, the same applies to B550 and X570. Any decent B550 is going to cost you $110-115, and a good X570 is $140. I don't see a point in saving $20-30 on a part that is the central nervous system of the computer by going down an entire product tier.

No, none of those CPUs pull over 150W stock. Fine text in the upper right corner of any of the TPU power graphs - Whole System, not CPU. None of those Vermeer chips are pulling beyond ~145W worst case under stock settings. 142W PPT is 142W PPT. Sometimes you can account for a ~1-2W deviation either way depending on the board, but PPT/TDC/EDC is always the law until you set a static OC.

That said, I agree that none of this 11th gen power limit stuff is new and is a little overblown.

The good boards are good boards that would be recommended, regardless of what platform it's on.
The three boards that suck, well, suck, and wouldn't be recommended anyways, regardless of what platform it's on.

 

[80-watt Hamster]

Plus you can get a decent H570 motherboard for $130-140, and B560 boards start at about $110(the "budget" AsRock B560 they talk about in the original video is $115). I honestly struggle to see why anyone would even consider B560 when the leap in price to H570 boards is so small.

Is it even statistically likely that an arbitrary H570 board will have better power delivery than an arbitrary B560? HU's been doing quite a bit on the VRM side lately, and one takeaway seems to be that robust power and board price are not directly correlated.

Edit: spelling

 

[Frick]

Surprised that this is news to anyone

It's news to me for sure. Historically, and up to quite recently apparently, any motherboard has been fine for "avarage" use, beyond that it was about features and overclocking capabilities. Boards not being able to actually fully handle supported CPUs is ... pretty bad, IMO.

The good boards are good boards that would be recommended, regardless of what platform it's on.
The three boards that suck, well, suck, and wouldn't be recommended anyways, regardless of what platform it's on.

If you lose double digit performance because you bought "a sucky motherboard" I would absolutely argue the specs for the base level should change. In the past you could pair a high end CPU with a low end motherboard and the CPU would perform as it should. This is the normalcy to strive for.

 

[X71200]

There's a military green PCB Tomahawk on that budget Intel end, but it's been out of stock lately. Costs around 130, and seems to have a decent VRM.

Historically, there have been some boards that sucked awfully even at stock with certain CPUs. This has happened quite a bit with old MSI boards and the Piledriver platform. They had crap VRM and the CPUs sucked power to no end, ended up with even popping the VRM. There was a huge thread on OCN some 10 years ago when Piledriver was still relevant about this. You could still dig and find specs about those Bulldozer platforms, not that it's of any relevancy today.

 

[W1zzard]

It's news to me for sure. Historically, and up to quite recently apparently, any motherboard has been fine for "avarage" use, beyond that it was about features and overclocking capabilities. Boards not being able to actually fully handle supported CPUs is ... pretty bad, IMO.

It is still like that. The difference is that some boards run CPUs out of spec, at default settings

 

[Valantar]

It is still like that. The difference is that some boards run CPUs out of spec, at default settings

You're kind of missing a major point here though: the span of what constitutes 'in spec' has grown massively over the past 3-4 generations. Which in turn opens the door for this quasi-overclocking through unlocked power limits, but also means that a chip can run at anything from, say, 2.5 to 4.6GHz (no, I didn't bother looking up precise numbers rn) and everything will still be 'in spec'. And Intel is in turn encouraging this through not enforcing their power limit specs beyond ensuring the baseline is met. MCE on high end SKUs and Z-series boards was already confusing, this is now propagating down the stack.

Of course it's also a major change from the 4c8t generations where base clock was typically not something you actually saw in practice, with most chips running within the given power limit while still exceeding base clocks noticeably. This is of course down to Intel (impressively, but still) stretching the usefulness of their 14nm process to stay competitive against much more efficient architectures and nodes.

Simple, you pop a 5800X in, turn on PBO, and load it up fully with whatever load testing program you prefer(Prime95, OCCT, Linpack, etc.). Yes, it will power throttle in some B550 boards.

So, in other words, you enable an auto OC and a in a power virus. You see how that is no longer stock, right? Enabling PBO isn't stock. It's an available option in BIOS. Hardly comparable to the issue at hand here.

As for your power numbers cited here, they're pure nonsense. PPT is strictly enforced unless settings are changed manually, and constitutes a hard package power likmit. Either show a source demonstrating otherwise or stop making misleading comparisons, please.

 

[W1zzard]

You're kind of missing a major point here though: the span of what constitutes 'in spec' has grown massively over the past 3-4 generations. Which in turn opens the door for this quasi-overclocking through unlocked power limits, but also means that a chip can run at anything from, say, 2.5 to 4.6GHz (no, I didn't bother looking up precise numbers rn) and everything will still be 'in spec'. And Intel is in turn encouraging this through not enforcing their power limit specs beyond ensuring the baseline is met. MCE on high end SKUs and Z-series boards was already confusing, this is now propagating down the stack.

You are 100% correct. I still think it's really bad to expect (want?) out of spec operation out of the box, from any product that you buy. What does that even do to warranty? Legally you press the power switch and the warranty is gone

 

[Valantar]

You are 100% correct. I still think it's really bad to expect (want?) out of spec operation out of the box, from any product that you buy. What does that even do to warranty?

Yeah, it's a complete mess. Defining usable warranty terms within a strictly controlled multi-component system is difficult enough. When specs instead become vague guidelines that nobody follows, terms like 'stock operation' become utterly meaningless. Of course this was the same with MCE on Z-series boards previously, but at least that was somewhat limited to high end OC SKUs and enthusiast users. Now it's becoming the de facto standard. All the while it completely erodes the applicability of benchmarks and makes any expectation of a given level of performance come with a huge asterisk attached.

 

[Mussels]

Surprised that this is news to anyone

the amount of people i've shown this video who dont believe its real...

"intel is the budget champion now!" (if you're fine with a 50% performance loss on cheap boards)

 

[asdkj1740]

i dont really get it. whats the problem?

 

[Mussels]

i dont really get it. whats the problem?

Did you watch the video?
Intels specs are all over the place, boards that follow intels guidelines are upto 50% slower than ones that blow the limits out, and throw extra wattage to the CPU (from 65W to 250W, i think was the worst)(
loosely:
Cheap board: 65W, 50% slower
high end board with SAME CPU: 250W, 50% faster.

So when you buy an intel chip, do you really get the same product you see reviewed, if they're using top end boards and you're not?

 

[W1zzard]

do you really get the same product you see reviewed, if they're using top end boards and you're not?

that's why my reviews show "default" at true Intel stock settings, and "max power limit", too.

 

[asdkj1740]

Did you watch the video?
Intels specs are all over the place, boards that follow intels guidelines are upto 50% slower than ones that blow the limits out, and throw extra wattage to the CPU (from 65W to 250W, i think was the worst)(
loosely:
Cheap board: 65W, 50% slower
high end board with SAME CPU: 250W, 50% faster.

So when you buy an intel chip, do you really get the same product you see reviewed, if they're using top end boards and you're not?

from what i have heard, all asus z490s are set to be 125w by default. and they got tremendous criticism then, that's why they have changed since z590.

there are plenty top end shit mobos (expensive as well) in the recent intel platforms:
z370 gigabyte gaming 7 (actually the whole z370 lineup messed up with thermal pad's thickness)
z390 msi ace (12 phases 1H+1L discret mosfets with poor heatsink and no chokes cooling and poor pcb cooling design)
z490 asus proart (looks good mosfet components but fucked up all over 100c)
z590 gigabyte aorus pro ax (great mosfet components with good looking heatsink but still fucked up)

to your question:
no, because intel chips cant work without mobo.
if you want your intel chip bought to be running at its full potential, you better check the power consumption of it then find a mobo that can provide such levels of power, if prime95 avx512 is all you want.
if what you guys want is a cheap board but 0% slower, then good luck my friend.

high end mobo, what is that, msi x570?

 

[newtekie1]

Is it even statistically likely that an arbitrary H570 board will have better power delivery than an arbitrary B560? HU's been doing quite a bit on the VRM side lately, and one takeaway seems to be that robust power and board price are not directly correlated.

Edit: spelling

It does seem like even the cheap H570 boards have decent enough VRMs compared to the cheap B560 boards.

No, none of those CPUs pull over 150W stock. Fine text in the upper right corner of any of the TPU power graphs - Whole System, not CPU. None of those Vermeer chips are pulling beyond ~145W worst case under stock settings. 142W PPT is 142W PPT. Sometimes you can account for a ~1-2W deviation either way depending on the board, but PPT/TDC/EDC is always the law until you set a static OC.

That said, I agree that none of this 11th gen power limit stuff is new and is a little overblown.

The whole system for the 5950X is 195w, it's going over 150w. And even if you normalize for the rest of the system, it's going more out of spec than the Intel chips are. Same for the 2700X. PPT/TDC/EDC are all adjustable, it's what PBO, AMD's own feature built into the processor, changes. So with PBO enabled 142w PPT limit is raised.

So, in other words, you enable an auto OC and a in a power virus. You see how that is no longer stock, right? Enabling PBO isn't stock. It's an available option in BIOS. Hardly comparable to the issue at hand here.

As for your power numbers cited here, they're pure nonsense. PPT is strictly enforced unless settings are changed manually, and constitutes a hard package power likmit. Either show a source demonstrating otherwise or stop making misleading comparisons, please.

PBO is an AMD technology. It's where they get their advertised performance numbers from. It isn't a shady overclock put in place by the board manufacturers. That's like saying enabling Thermal Velocity Boost on Intel's side is an auto overclock that takes the processors out of stock. It just isn't true, AMD built PBO into the processors and reported performance numbers with in enabled. Hell they thoughted it as a major feature during the 3000 series launch. It also does NOT overclock the processor. It does not push the clock speeds beyond the advertised speeds, what it does is allow the processor to boost more cores for longer, putting more load on the VRMs. It does this specifically by raising those PPT limits(along with a few other power related limits). This whole thread is basically complaining that motherboard manufacturers on the Intel side are shipping boards with essentially Intel's PBO enabled by default and some don't(and I have to wonder if AMD motherboard manufacturers do this sometimes too).

If you want the proof, just go look at the TPU reviews. You'll see 105w rated processors consuming way more than 105w, as I pointed out. But back to the motherboards and VRM throttling, when you buy a B550 motherboard, and it only has a 4-pin CPU power connector that is rated for 75w, you really have to scratch your head and ask "is this board going to really be able to reliably deliver double that for long periods of time?" And I can already answer that, No.

"intel is the budget champion now!" (if you're fine with a 50% performance loss on cheap boards)

It still applies, like I said, the "cheap" boards aren't really that much cheaper anyway. It's like $20 difference.

 

[GorbazTheDragon]

It does seem like even the cheap H570 boards have decent enough VRMs compared to the cheap B560 boards.

H570s are pretty thin on the ground compared to B560s, overall selection is less but realistically you can just lump all of them into the same category...

In terms of VRMs the only ones I would really trust are ones with powerstages, and maybe on a stretch the MSI ones with heatsinks... As long as Asus and GB are using 4C10N/4C06N setups their boards with discrete mosfets are pretty much on the avoid list for anything more than a quad core (same goes for Asrock and their random rubbish mosfets).