What are you looking for in a Motherboard Review?

[agent_x007]

@ir_cow
I'm asking to only do a single lower voltage test on boards that will throttle under auto settings. If you want to bypass "golden sample" issue, just decrease all boost frequencies to a lower and fixed value (like ~4,00GHz (P-cores, 3GHz for E-cores)).
I'm pretty sure all new "usual" x86 CPUs made will be able to do it at 1.1V (silicon lottery be damned). To be clear : I am NOT asking you to push your sample "to the max" on lower voltage. Just pick a ~10% lower voltage point and adjust frequency down until You think most CPUs should be able to achieve.

 

[ir_cow]

@agent_x007 I'm not following this completely. What will knowing the lowest voltage for a particular CPU accomplish?

 

[Mussels]

@agent_x007 I'm not following this completely. What will knowing the lowest voltage for a particular CPU accomplish?

It'd show you the board that lies about it's stock voltage the most, by using aggressive LLC?

 

[agent_x007]

@agent_x007 I'm not following this completely. What will knowing the lowest voltage for a particular CPU accomplish?

What's with going to lowest voltage all of the sudden ?
I want to know, how much this "quick fix" can help users of such boards.
Example : 7950X/13700k, set 1.1V @ 4GHz (I'm guessing since I can't know what stock voltage on AM5 or Z790 stuff is), check if board doesn't throttle.
If it does - it's beyond saving, if it doesn't - it may help some to "fix" this issue temporarely.

 

[ir_cow]

Quick fix for for lower CPU temperatures under load is what you're looking for?

 

[agent_x007]

Not CPU - it's VRM. This is MB testing thread after all.
Also not only temps, but getting performance to not drop as well.
Lower CPU temp is just a byproduct in this case.
"Validation of "quick fix" for such cases" - writing it differently.

 

[Mussels]

but voltage doesnt equal the same wattage if you use a different CPU

That's why wattage is the better measurement

 

[ir_cow]

Not CPU - it's VRM. This is MB testing thread after all.
Also not only temps, but getting performance to not drop as well.
Lower CPU temp is just a byproduct in this case.
"Validation of "quick fix" for such cases" - writing it differently.

Oh okay I think I understand now you want to see what the vrm temperature would be with the lowest amount of voltage/ wattage without losing any performance versus a stock CPU. Am I correct?

 

[Ferrum Master]

@ir_cow
I'm asking to only do a single lower voltage test on boards that will throttle under auto settings. If you want to bypass "golden sample" issue, just decrease all boost frequencies to a lower and fixed value (like ~4,00GHz (P-cores, 3GHz for E-cores)).
I'm pretty sure all new "usual" x86 CPUs made will be able to do it at 1.1V (silicon lottery be damned). To be clear : I am NOT asking you to push your sample "to the max" on lower voltage. Just pick a ~10% lower voltage point and adjust frequency down until You think most CPUs should be able to achieve.

All you would test is VRM performance at certain point. It is not linear, it is a a curve and the curve depends on the amount of phases and complexity of the VRM internal gating mechanisms, that are usually left unimplemented as makers are way too lazy to experiment with stability for the sake of few W of POWER. The point is, at different steps you can expect vastly different results, also depending of the BIOS vendor not only the VRM IC. Only places where proper VRM's are doing their job are mostly laptops, because there less current (in Amperes) matters because of battery life and limited heat dissipation capacity.
Basically if it consumes more at idle it ain't always the case it will do that more at medium or high loads and vice versa, there are too many variables.

I hope you all got the idea about my usual whine, you could at least try to start using proper physics terms in reviews and not to submit to dark ages of English slang, where bloody nobody here speaks correctly using SI system, it is plague in online communities. There are no wattage, amperage, there is power and current as there is Faradage or Kilogramage or whatever, there is temperature and weight.

Please look, if someone would mention some smartarse online dictionary showing that those words are real, you can shove them to the dictionary maintainer in the name of that respected physicist who created the unit.

International System of Units - Wikipedia

en.wikipedia.org

 

[AusWolf]

All you would test is VRM performance at certain point. It is not linear, it is a a curve and the curve depends on the amount of phases and complexity of the VRM internal gating mechanisms, that are usually left unimplemented as makers are way too lazy to experiment with stability for the sake of few W of POWER. The point is, at different steps you can expect vastly different results, also depending of the BIOS vendor not only the VRM IC. Only places where proper VRM's are doing their job are mostly laptops, because there less current (in Amperes) matters because of battery life and limited heat dissipation capacity.
Basically if it consumes more at idle it ain't always the case it will do that more at medium or high loads and vice versa, there are too many variables.

I hope you all got the idea about my usual whine, you could at least try to start using proper physics terms in reviews and not to submit to dark ages of English slang, where bloody nobody here speaks correctly using SI system, it is plague in online communities. There are no wattage, amperage, there is power and current as there is Faradage or Kilogramage or whatever, there is temperature and weight.

Please look, if someone would mention some smartarse online dictionary showing that those words are real, you can shove them to the dictionary maintainer in the name of that respected physicist who created the unit.

International System of Units - Wikipedia

en.wikipedia.org

If you don't have enough wattage for your CPU, you upgrade your computer's motherboardage and/or power supplyage. Easy.

 

[Ferrum Master]

If you don't have enough wattage for your CPU, you upgrade your computer's motherboardage and/or power supplyage. Easy.

I think he actually talks about anomalies, vdroop, throttle, bad design induced errors.

There should not be modern motherboards not capable of running a CPU model for its respective socket. If not the motherboard should not boot and have a blacklist. OC is a different topic.

On the paper it always looks nice, but the reality always differs, that's sure, that's what he actually tries to say imho. There are usually quirks with each board/CPU combo actually making it work normally. Often it even needs BIOS fixes as some options in reality actually do nothing and do not work. Reviewers do the job on a really fast manner and often don't indulge themselves into the bits if we talk about long overclocking stability and actually what happens in the unguaranteed OC modes with each motherboard. It simply would take weeks for each board.

ofc we would like to see max RAM speeds for each config. Single, dual, then double the ranks, CPU OC... yet there are usually stuff that ends up ignored, like some fan headers do not really support fan stop(yet the option is present in BIOS, F* GB), limited range etc and those things matter and are often overlooked.

 

[AusWolf]

I think he actually talks about anomalies, vdroop, throttle, bad design induced errors.

There should not be modern motherboards not capable of running a CPU model for its respective socket. If not the motherboard should not boot and have a blacklist. OC is a different topic.

On the paper it always looks nice, but the reality always differs, that's sure, that's what he actually tries to say imho. There are usually quirks with each board/CPU combo actually making it work normally. Often it even needs BIOS fixes as some options in reality actually do nothing and do not work. Reviewers do the job on a really fast manner and often don't indulge themselves into the bits if we talk about long overclocking stability and actually what happens in the unguaranteed OC modes with each motherboard. It simply would take weeks for each board.

ofc we would like to see max RAM speeds for each config. Single, dual, then double the ranks, CPU OC... yet there are usually stuff that ends up ignored, like some fan headers do not really support fan stop(yet the option is present in BIOS, F* GB), limited range etc and those things matter and are often overlooked.

True. Though I was just being sarcastic about the use of improper terms found in modern reviews (which bugs me too, by the way).

 

[Ferrum Master]

True. Though I was just being sarcastic about the use of improper terms found in modern reviews (which bugs me too, by the way).

The sad story is, if you complain, the other party ie reviewer rejects it and acts like nothing happened. Basically rejects and whole fundamental physics theory and teachings, I do not know, it is beyond rude.

 

[AusWolf]

The sad story is, if you complain, the other party ie reviewer rejects it and acts like nothing happened. Basically rejects and whole fundamental physics theory and teachings, I do not know, it is beyond rude.

It's probably governed by the same unwritten rule that says that if X product is 5% behind Y product in performance, then you'll have to post the review with a large "THIS IS SH!*" clickbait banner.

 

[agent_x007]

Oh okay I think I understand now you want to see what the vrm temperature would be with the lowest amount of voltage/ wattage without losing any performance versus a stock CPU. Am I correct?

Well, problematic boards will have lower than stock performance...
I want to have a way to either recover lost performance from throttling (by offseting CPU Voltage by 1-10% below stock), or set "custom" frequency/voltage throttle point (previously mentioned fixed 1.1V 4GHz example), that better makes use of limited power budget VRM can provide.

Either way, don't do it if you can't or simply don't want to - this is only a suggestion.

 

[Count von Schwalbe]

Well, problematic boards will have lower than stock performance...
I want to have a way to either recover lost performance from throttling (by offseting CPU Voltage by 1-10% below stock), or set "custom" frequency/voltage throttle point (previously mentioned fixed 1.1V 4GHz example), that better makes use of limited power budget VRM can provide.

Either way, don't do it if you can't or simply don't want to - this is only a suggestion.

As in, if there are issues at stock with a cheap motherboard, can it be fixed by undervolting?

 

[agent_x007]

Yes, with option for max. boost frequency limit on top of undervolting (as for example decreasing all turbo ratios to all core limit max.). Don't know how bad VRMs we will see...

 

[JustBenching]

Reviewing a motherboard is HARD. The only thing you can safely measure is how good the vrm implementation is.

Personally im more interested in memory support, especially now with ddr5. Can a cheap 200$ mothetboard support 6600+ frequencies?

 

[Ferrum Master]

Reviewing a motherboard is HARD. The only thing you can safely measure is how good the vrm implementation is.

Personally im more interested in memory support, especially now with ddr5. Can a cheap 200$ mothetboard support 6600+ frequencies?

It doesn't mean the reviewer gets those frequencies, you will get it also and vice versa. It depends on silicon. How the CPU IMC pairs with memory. They all have variation the IMC and memory modules.

Basically being lucky is involved too much in the process.

 

[JustBenching]

It doesn't mean the reviewer gets those frequencies, you will get it also and vice versa. It depends on silicon. How the CPU IMC pairs with memory. They all have variation the IMC and memory modules.

Basically being lucky is involved too much in the process.

It goes without saying that youll use a cpu and dimms that can actually do it, and then its up to the motherboard, lol.

 

[ir_cow]

Well, problematic boards will have lower than stock performance...
I want to have a way to either recover lost performance from throttling (by offseting CPU Voltage by 1-10% below stock), or set "custom" frequency/voltage throttle point (previously mentioned fixed 1.1V 4GHz example), that better makes use of limited power budget VRM can provide.

Either way, don't do it if you can't or simply don't want to - this is only a suggestion.

I'm neither for or against it. Just trying to figure out what the request was at the core.

It goes without saying that youll use a cpu and dimms that can actually do it, and then its up to the motherboard, lol.

Yep. First you need to validate the CPU IMC can actually handle it and that is the hard part. For example so far I've only managed to get DDR5-6933 stable. This required a OC motherboard as well. I wish it was that easy, but the SA, VDDQ_TX and IMC (VDD2) voltages are not valid on different motherboards. It will get you close, but each one is slightly different. Essentially for every MB you need to start from the beginning and work your way up in freq and voltage.

After that is done you can roughly say what the maximum freq supported is for X generation CPU currently. The flaw in that method is we know from previous generations of CPUs in the same socket, is that the new always have better memory support. What was the max before is not for the next CPU. That is what makes it complicated. You have "max freq" auto XMP, max freq manual voltages and you have the unknown of the future. Should I explain everything above for each review, so just say this is the max, even though it isn't the absolute?

Some MBs are really easy though. If I can't boot above DDR5-6000 and the CPU can do DDR5-6933, chances are the next generation CPU will not fair any better.

Lastly, I personally stick to "safe voltages". Meaning lets say I could jam 1.65 V into some memory and possibly make DDR5-7000+ stable. That isn't safe per the information on the datasheet. Same goes for the CPU voltages. For me DDR5-6933 is the soft limit. Can't go higher without breaking some sort of safety limit.

Now I could see reporting the max bootable memory freq even if its not stable. That is about 7200 MT/s right now without pushing to hard into the no no voltage zone. However this could be misleading. Stable in windows and bootable are two very different things.

 

[Mussels]

Yeah you're not doing LN2 testing here, so overclocking should have some preset goals


For example if you know your CPU can do 5.2GHz on one board, you can find out how easy/hard it was to reach that on alternative boards (especially useful with say, ITX vs ATX boards), and see if there was large voltage or wattage differences.

"X board had more stable voltages so the same OC used 20W less" is valuable, but only something you can determine by using the same clock speeds between boards instead of hunting for max overclocks.

 

[Solid State Brain]

If reviews were to introduce overclocking-focused topics as discussed above, I would add oscilloscope readings from the VRM during transients in a standard high load at the various LLC and PWM settings.

See a comprehensive example here: https://elmorlabs.com/2019-09-05/vrm-load-line-visualized/

Summary

The captures below show the output voltage transient behavior when loaded with about 70 A for ~150 μs. The LLC1 capture illustrates ideal load-line behavior. As the load-line value decreases (higher level), the line flattens and the under/overshoot spikes at start and finish become more pronounced. The lowest voltage point at the beginning of the load transient does not improve much. In this case, using a Load-Line Level of above 3 seems questionable. The load voltage would increase meaning higher power consumption, but the worst case lowest voltage would stay the same. Additionally the over-shoot after load release increases. Increasing the VRM Switching Frequency from 500 KHz to 800 KHz shows very minor improvements.

 

[Ferrum Master]

If reviews were to introduce overclocking-focused topics as discussed above, I would add oscilloscope readings from the VRM during transients in a standard high load at the various LLC and PWM settings.

Those dudes should invest in better probes. Still spending half a grand on a test setup just for reviews doesn't make sense. Well getting a oscilloscope rarely does, I have those, but I got mine cheap as a sell out from bankrupted service location to cover their bills, usually just use ones your job provides and those at least sometimes pay off... but in reality never, they are just there because of the manufacturer requirements to give service authorizations.

 

[JustBenching]

I'm neither for or against it. Just trying to figure out what the request was at the core.


Yep. First you need to validate the CPU IMC can actually handle it and that is the hard part. For example so far I've only managed to get DDR5-6933 stable. This required a OC motherboard as well. I wish it was that easy, but the SA, VDDQ_TX and IMC (VDD2) voltages are not valid on different motherboards. It will get you close, but each one is slightly different. Essentially for every MB you need to start from the beginning and work your way up in freq and voltage.

After that is done you can roughly say what the maximum freq supported is for X generation CPU currently. The flaw in that method is we know from previous generations of CPUs in the same socket, is that the new always have better memory support. What was the max before is not for the next CPU. That is what makes it complicated. You have "max freq" auto XMP, max freq manual voltages and you have the unknown of the future. Should I explain everything above for each review, so just say this is the max, even though it isn't the absolute?

Some MBs are really easy though. If I can't boot above DDR5-6000 and the CPU can do DDR5-6933, chances are the next generation CPU will not fair any better.

Lastly, I personally stick to "safe voltages". Meaning lets say I could jam 1.65 V into some memory and possibly make DDR5-7000+ stable. That isn't safe per the information on the datasheet. Same goes for the CPU voltages. For me DDR5-6933 is the soft limit. Can't go higher without breaking some sort of safety limit.

Now I could see reporting the max bootable memory freq even if its not stable. That is about 7200 MT/s right now without pushing to hard into the no no voltage zone. However this could be misleading. Stable in windows and bootable are two very different things.

Yeah, I was thinking something along the lines of max bootable "stable" frequency, meaning it gets to desktop and maybe can run a single CBR23. If a mobo boots at 6600 for example. chances are you can make it stable at least at 6400. Or even having an xmp kit of 6600 and checking if it works at XMP. The voltages should be left to auto, since that's what most people would do / expect to do when they just try to run XMP.

Especially with ddr5 this is especially important, cause there are lots of complaints about people not being able to boot over 6k in most 4dim motherboards (and ofcourse the November batch of apex :O )

Personally I got an apex (which sucked), and now eventually a unify X to make sure 7k+ dims actually work, cause nobody really tests fast dimms on midrange motherboards.