[XFastest] Puts different boards through VRM heat test, which will burn first?

[sin0822]

Here is the Link, and I posted a summary below: http://www.xfastest.com/thread-69873-1-1.html




Platform Setup: The CPU voltage as 1.4V, load line to maximum.

ECS: Due to no turbo boost function, it's only ran as 3.2GHz，so highest mosfet temperature only 54.3ﾟC, and CPU voltage only 1.34V.
MSI: Due to no turbo boost function, it's only ran as 3.2GHz，so highest mosfet temperature was 65.2ﾟC
ASROCK:The guy said Asrock's CPU voltage is reliable(ran as 1.4v), However their CPU frequency start from 3.5GHz, than drop to 3.2GHz, at last down to 1.2GHz, and their highest mosfet temperature was 89.8ﾟC
ASUS: CPU voltage was 1.44V, higher than 1.4v, and CPU ran as 3.5GHz in whole test, the highest mosfet temperature was 85.1ﾟC
GIGABYTE: BIOS version was F7, CPU voltage was 1.392V，Lower than 1.4V（Due to load line set up）, and CPU ran as 3.5GHz in whole test, the highest mosfet temperature was 83.9ﾟC

Very interesting results if I might say so myself. Excellent effort XFastest!

[eXa]

Wait! No one burned up? Are they still going... or??? Up the voltage? confused...

[biohead]

judging from those pictures they're measuring the heatsink temperature without consistent probe placement. so what exactly do these results mean?

[xdan]

Not sure what that quy measures or how it tests, but the guys from lab501 tested Asrock X79 Extreme4 and they run 3930K at 5ghz 1.5v stable.
Nu throttle, no burn. With a Corsair H80 cpu cooler.
http://lab501.ro/placi-de-baza/asroc...reme4-review/4

[sin0822]

I think on purpose they didn't put airflow over the VRMs. I think they were trying to burn up a board again, so show that some new BIOS was throttling, but in fact the GB didn't throttle.

These tests are to see how the boards handle the environment they first put that other UD3 in, so see what happens. The boards SHOULD throttle down when they reach their VRM limit. The whole point is seeing how high one can set that limit.

[Kallenator]

This is actually good people, if such a test although a more refined one (with an IR camera as well imo) would become mainstream when reviewing enthusiast mainboards then the companies would pay more attention!

[sin0822]

I 100% agree, I think Tin did a test of the UD7 Vs M4E remember that? He used his IR camera and took shots. Pretty interesting stuff.

of course then you get into the mess of which board has higher switching frequency, which board has a more aggressive PWM scheme, and which board has more phases, which infact would affect the temperature.

Of course it is interesting to put up too boards and compare none the less. I do however agree that IR camera is the best way to do it. Of course if most reviewers don't have a scope how can you expect them to have an IR camera of any value? The FLIR ones are like $1000.

[STEvil]

switching frequency does almost nothing for pwm temperature. Only way to bring temperatures down significantly is to use more fets.

[Particle]

switching frequency does almost nothing for pwm temperature. Only way to bring temperatures down significantly is to use more fets.

That isn't really accurate. As switching frequency goes up, potential output quality increases but at the expense of efficiency if everything else stays the same. As efficiency drops, VRM temp will increase when the load remains constant.

[Hondacity]

who has the hottest heatsinks win?

[Kallenator]

I 100% agree, I think Tin did a test of the UD7 Vs M4E remember that? He used his IR camera and took shots. Pretty interesting stuff.

of course then you get into the mess of which board has higher switching frequency, which board has a more aggressive PWM scheme, and which board has more phases, which infact would affect the temperature.

Of course it is interesting to put up too boards and compare none the less. I do however agree that IR camera is the best way to do it. Of course if most reviewers don't have a scope how can you expect them to have an IR camera of any value? The FLIR ones are like $1000.

I don't remember Tin's test. Have a link? =)

IR cameras are indeed expensive, but should be able to get a second hand one on the cheap though if one is able to find?

Scopes wont be a problem for us much longer:
http://www.kickstarter.com/projects/...l?ref=category

[sin0822]

switching frequency does almost nothing for pwm temperature. Only way to bring temperatures down significantly is to use more fets.

It does increase the temperature if raised and lower it is lowered. How can you increase the frequency at which the VRm operates at and have no repercussions, and expect the switching losses to what, disappear?

Why don't we quote Raja's R4E OC Guide?:

CPU Voltage Frequency: Sets the switching frequency of the power FETs supplying processor Vcore. Lower switching frequencies lead to a higher VRM efficiency (small power saving) and lower VRM operating temperatures. Setting a higher switching frequency aids transient response (the recovery of voltage to the applied level after a load condition) – at the expense of heat.

From what i saw for the IR meter when i googled retail is like 1000+ so second hand maybe $500? Still pretty expensive.

here is his comparision: http://www.xtremesystems.org/forums/...lutions-review

Um I think they measured MOSFET temperatures, Maybe they put a probe next to the FET or between the heatsink and one of the FETs. I don't really know what they did, but they said MOSFET temp.

What is that you linked?

[STEvil]

What I was getting at is that manufacturers are using increased switching frequencies to make things look more stable voltage wise, which increases VRM temperatures and reduces efficiency, as you guys said. The fet transient states (voltage spikes basically) get pretty ugly with the larger switching speeds as well, which means more smoothing capacitors required.

More fets is really the only way to decrease VRM temperatures, with or without the higher switching speed.

[Raja@ASUS]

The fet transient states (voltage spikes basically) get pretty ugly with the larger switching speeds as well, which means more smoothing capacitors required.

This isn't true, there are a whole host of factors that determine the amount of bulk capacitance a given VRM needs. Buck controllers and FETs that support higher switching freqeuncies (provided the implementation is optimized for the task at hand) are generally less reliant on large amounts of bulk capacitance to meet ripple and overshoot tolerance margins - otherwise there is no real reason to use them.

[STEvil]

Which is why I made a general statement about them. When used within their limitations things are fine, but when they get out of range it turns ugly, and getting a single fet out of range isnt very hard, I mean that is what caused so many Z7S-WS deaths in the end, right?

Which reminds me I should PM gary, your email server isnt letting my emails through again I think..

[BeepBeep2]

What I was getting at is that manufacturers are using increased switching frequencies to make things look more stable voltage wise, which increases VRM temperatures and reduces efficiency, as you guys said. The fet transient states (voltage spikes basically) get pretty ugly with the larger switching speeds as well, which means more smoothing capacitors required.

More fets is really the only way to decrease VRM temperatures, with or without the higher switching speed.

I thought it was the opposite...

Load Line Calibration being used to make voltages more stable but increasing the amount of transient overshoot, increased switching speeds reduce that transient overshoot when coupled with aggressive Load Line Calibration at the expense of efficiency and heat.

We all know the Z7S-WS was an extremely poorly designed board...and I can understand why you'd hate on ASUS for it however that is just one product.

[Raja@ASUS]

Which is why I made a general statement about them. When used within their limitations things are fine, but when they get out of range it turns ugly, and getting a single fet out of range isnt very hard, I mean that is what caused so many Z7S-WS deaths in the end, right?
.

This has more to do with OCP and OTP than the number of capacitors used, so the "general" statement angle isn't correct either. Regardless of switching frequncy, if a FET or VRM is used out of range things can get ugly if OCP or other safety features don't kick in.

[STEvil]

And thats the issue, its easier to push them out of spec now.

[Raja@ASUS]

And thats the issue, its easier to push them out of spec now.

Which has not relation to capacitance at all. And seeing this thread is about X79, we had OTP and OCP implemented properly on our boards from the get-go, so you cannot push them "out of spec" to the point of failure.

[STEvil]

Which has not relation to capacitance at all. And seeing this thread is about X79, we had OTP and OCP implemented properly on our boards from the get-go, so you cannot push them "out of spec" to the point of failure.

Not enough capacitance on either side of the fet could allow for out of spec operation dependent on the implementation method and BIOS settings. As was stated earlier though, the capacitance issue is more of a general issue and was not meant to be the "end all be all" statement nor to point to it being the largest of the issues or the cause of the fets entering cascade, only that due to the implementation methods used it is an issue.

You monitor each fet separately? I highly doubt that.

Do you even know how the monitoring functions of the drivers and fets work? If one fet enters a cascade scenario there is almost no method of reliably saving it from damage, let alone outright failure.

[bamtan2]

you can see from these results, and other recent performance and power results, that today's high-end asus boards really are high-end. they are good engineering. they have a premium price, but they back it up with premium engineering. am I right?

[STEvil]

you can see from these results, and other recent performance and power results, that today's high-end asus boards really are high-end. they are good engineering. they have a premium price, but they back it up with premium engineering. am I right?

I'd say they're decent, about average on the high end, though you're the only one to bring up current Asus build quality directly so far.. for some reason.

[Copie]

I'd say they're decent, about average on the high end, though you're the only one to bring up current Asus build quality directly so far.. for some reason.

The R4E is getting nothing but huge praises from owners and reviewers alike, i can see via your sig that your not keen on Asus, but even still they have their X79 range almost perfect.

[Kallenator]

From what i saw for the IR meter when i googled retail is like 1000+ so second hand maybe $500? Still pretty expensive.

here is his comparision: http://www.xtremesystems.org/forums/...lutions-review

Um I think they measured MOSFET temperatures, Maybe they put a probe next to the FET or between the heatsink and one of the FETs. I don't really know what they did, but they said MOSFET temp.

What is that you linked?

Quite the sum of money that, problematic. Did a little more searching as well, found a second hand one for 3000USD :p

Wow, Tin really knows his way around!

The link includes a USB oscilloscope, if you watch the short video you can see a bit of its capture software, it looks very easy to use and makes nice graphs to read.
And it is extremely cheap, 20USD if I haven't misunderstood anything, and that is silly money for a proper scope.

[STEvil]

The R4E is getting nothing but huge praises from owners and reviewers alike, i can see via your sig that your not keen on Asus, but even still they have their X79 range almost perfect.

Thats 1 board which has had no 3rd party thermal tests performed of its VRM yet.. (to points of failure/protection I should say, which isnt an extreme issue on most top end boards anyways).

I'm sure it'd do fine though.

EDIT

For those who didnt look close enough, xfastest's tests were done with a simple thermal temperature probe taped to each VRM heatsink.

EDIT 2

That Gigabyte board really made some ugly waves in TiN's tests. The fet drivers looked a bit toasty too.