Overclock.net › Forums › AMD › AMD - General › VRM on the new AM4 motherboards
New Posts  All Forums:Forum Nav:

VRM on the new AM4 motherboards - Page 95

post #941 of 1994
You want 5-10 years warranty on mobo? It comes with 3 years, which is 1 year over 2 year minimum in EU.
MSI Titanium, again 3 years.

Overall the x370 boards are being milked IMHO compared to Zx70 boards especially.
post #942 of 1994
Quote:
Originally Posted by TheOldTechGuru View Post

He also assumes only 300 Khz switching even though the mosfet is rated for 1 Mhz.

For MSI this was very good. biggrin.gif
In modern buck converters with switching frequencies (under load) >300kHz the switching losses are often higher than the losses in the on-states of the FETs.
NexFETs and PowIRstages are aiming for decreasing these high switching losses with standard MOSFETs like the NIKO parts on the XPower.
Quote:
Originally Posted by TheOldTechGuru View Post

He does note that running the two PK632BA's in parallel yields 1/2 the resistance but stops short of following through on that - 1.5 milliohms, which puts the effective resistance down to the level of the best (NextFET rated at 1.2) but a current capacity of 78 amps each - yielding nearly the effective efficiency with a higher package current capacity.

Forget about on-state losses and the on-resistance, switching losses are what makes a multiphase buck converter efficient or inefficient nowadays.

Switching losses are what "kills" the efficency of the VRMs of the XPower since there are six phases with three FETs each of the CPU VCC.
Also their Rise- and Fall-Times aren't as good as e.g. the Rise- and Fall-Times of the CSD87350s. Even when they are driven @12V VGS.

When this discusisson came up I just did the calculation of C6H vs. XPower for FET losses @50°C @1,4V @110A and the XPower was much more inefficient just caused by the switching losses.
Unfortunately I can't share the calcualtion now because I'll not have access to my computer for the next days.
Quote:
Originally Posted by Zhany View Post

Buildzoid is using worst case scenario for temperatures and it makes sense as this way you can see what kind of current you can drive in the worst possible case scenario without the VRM dying, the 50A rating per mosfet on the NIKOs is at TCase temperature of 25C at TCase 100C that rating drops down to 31.5A at 125C that amperage rating is going to drop even lower.

But 125°C are really unrealistic.
Also they'd have to upscale the on-resistance for temperature too when calcuating losses of a buck converter.
0.5% more on-resistance for 1°C more.
Quote:
Originally Posted by Zhany View Post

As for the switching frequency he is assuming what is typically used on Motherboards by most manufactures, as stated in the video he doesn't know for sure if that is the switching frequency since he doesn't have an actual sample board to test with, so he is letting the audience know this is what my assumption is.

But 300kHz really is a pretty low frequency for controllers, drivers, doublers and FETs which usually are able to handle frequencies up to 1MHz or 1.5MHz.
Of courses switching losses can be minimized by that but for hardcore overclocking a manufaturer should better run the VRMs at their maximum switching frequency since the transient response performance is the best in this case.
post #943 of 1994
Thanks br0da, this additional information is helpful (at least to me).

I knew this is all a lot more complicated than it might seem, at least for those of us lacking any formal education in this area. From what I read, the capacitors and chokes can also have an impact in that when they are well-designed they can mitigate some of the workload of the mosfets, reducing the strain upstream (again, I suspect I am way over-simplifying this).
post #944 of 1994
Thanks broda for this list:
https://www.hardwareluxx.de/community/f12/am4-mainboard-vrm-liste-1155146.html

That should be very helpful.

Hopefully the X390 rumors are true and hopefully we will see something soon from AMD.






Quote:
Originally Posted by DADDYDC650 View Post

Taichi PCB breakdown

https://youtu.be/_ic5d5qAhBU


Pretty much what we expected. RAM VRM too is very good (that's gravy on the good VRM so to speak). Probably the best of the X370 boards, VRM wise anyways.

The BIOS clearly needs work though - the LLC is very immature. I think it is best to wait a few months before buying right now.

Buildzoid has a video on this:


But it's a software issue, so we'll have to wait and see. In a few months we'll know.





Quote:
Originally Posted by DADDYDC650 View Post

Overpriced VRM design on the MSI Titanium? Ouch!


Pretty much a repeat of the things I've said this whole time.

Oh and capacity wise, Buildzoid believes that the VRM can only take about half as much (252A on X370 XPower vs 480A on X370 Taichi). Not to mention, at 252A, as Buildzoid highlights at around 12:30, it's much less efficient (64W of waste heat, assuming 1.4V and 252A @ 125C). While MSI's heatsink is probably the best design, the VRM is just not what you expect on a flagship grade board.

There's no way to fix this one, save with a Revision 2.0 hardware release.





Quote:
Originally Posted by TheOldTechGuru View Post

I keep hearing how inefficient this design is and yet it performs better than the Gigabyte design he touts in comparison...and is only a *hair* behind the CH6. The wattage measurements were taken *at the wall*, not with software. Still waiting on an explanation. I know Chew's response, but *at the wall*...so I don't want to hear excuses about software or calibration...

http://www.kitguru.net/components/motherboard/luke-hill/msi-x370-xpower-gaming-titanium-motherboard-review/10/
Warning: Spoiler! (Click to show)

By the way, my question is serious. Somebody must be missing something, and I think it is in everyone's interest to understand what it is...

As linked previously , even the KitGuru link, which you've linked calls MSI out for their poor VRM.


http://www.kitguru.net/components/motherboard/luke-hill/msi-x370-xpower-gaming-titanium-motherboard-review/3/?PageSpeed=noscript

Quote:
The 6+4-phase power delivery system will be split as six phases for the CPU and four for the SOC voltage rail. MSI uses an International Rectifiers IR35201 PWM controller which can manage up to eight phases. The presence of 3+2 (CPU+SOC) International Rectifiers IR3598 dual/doubler MOSFET drivers shows that MSI is leveraging phase doubling to drive the MOSFETs.

For the CPU power delivery section, six NIKOS PowerPAK PK616BA and twelve NIKOS PowerPAK PK632BA are combined. The SOC section gets four PK616BA MOSFETs and four PK632BA.

MSI’s power delivery system seems a little light for a flagship, overclocking-geared design. ASRock and ASUS offer higher total phase counts on their competitors while also using efficient Texas Instrument NexFET power blocks (MOSFETs).

That was a review you linked.

As for the reason why it's only slightly higher at load is because Ryzen at load is not drawing near the limits. If it were, the Nikos would be in very serious trouble unless additional cooling (ex: waterblocks) was supplied. That's a testament to AMD"s power efficient architecture. That does not mean however that exonerates the MSI board. Flagship boards mean flagship VRMs, not VRMs fit for a board half its price. Yet another possibility is that there are other components on the motherboard that are drawing power (ROG boards do tend to have a lot of "stuff" on them from experience). We would need a very thorough breakdown of what specific components are using power in that test scenario.










Quote:
Originally Posted by br0da View Post

For MSI this was very good. biggrin.gif
In modern buck converters with switching frequencies (under load) >300kHz the switching losses are often higher than the losses in the on-states of the FETs.
NexFETs and PowIRstages are aiming for decreasing these high switching losses with standard MOSFETs like the NIKO parts on the XPower.
Forget about on-state losses and the on-resistance, switching losses are what makes a multiphase buck converter efficient or inefficient nowadays.

Switching losses are what "kills" the efficency of the VRMs of the XPower since there are six phases with three FETs each of the CPU VCC.
Also their Rise- and Fall-Times aren't as good as e.g. the Rise- and Fall-Times of the CSD87350s. Even when they are driven @12V VGS.

When this discusisson came up I just did the calculation of C6H vs. XPower for FET losses @50°C @1,4V @110A and the XPower was much more inefficient just caused by the switching losses.
Unfortunately I can't share the calcualtion now because I'll not have access to my computer for the next days.
But 125°C are really unrealistic.
Also they'd have to upscale the on-resistance for temperature too when calcuating losses of a buck converter.
0.5% more on-resistance for 1°C more.
But 300kHz really is a pretty low frequency for controllers, drivers, doublers and FETs which usually are able to handle frequencies up to 1MHz or 1.5MHz.
Of courses switching losses can be minimized by that but for hardcore overclocking a manufaturer should better run the VRMs at their maximum switching frequency since the transient response performance is the best in this case.



Buildzoid usually uses the worst case situations. Agree that 125C is way, way higher than normal, but it's a matter of making sure your hardware can perform .... in the worst case that will happen.

There are a few extreme cases where I have heard of this happening:

http://www.legitreviews.com/xfx-radeon-r9-290-double-dissipation-video-card-review_138612/12



Keep in mind that this was playing Crysis 3. Furmark or OCCT would easily push this to 150C and cause system shutdown. XFX did correct this for the 390 DD, but it's alarming that this even happened to begin with.



That said, 125C is used consistently (ex: Buildzoid uses it for his GPU calculations too to prove that the VRMs are adequate or inadequate in the case of reference boards and a few downgraded PCBs). It does provide an apples to apples comparison.


The problem here is that:
  • For 300 USD, MSI is offering a 6 stage NIKOS (2 low side, 1 high side)
  • For 200 USD, Asrock is offering a 12 stage (doubled 6 really) CSD87350 configuration
  • Asrock also offers 4 stages of CSD87350 for the SOC versus 2 on the MSI of Nikos and Asrock offers a very solid RAM VRM

From where I am standing, it is very difficult to justify the MSI board from a value point of view.
Edited by CrazyElf - 4/17/17 at 7:26pm
Trooper Typhoon
(20 items)
 
  
CPUMotherboardGraphicsGraphics
5960X X99A XPower MSI r9 290X Lightning  MSI r9 290X Lightning 
RAMHard DriveHard DriveHard Drive
G.Skill Trident Z 32 Gb Samsung SM843T 960 GB Western Digital Caviar Black 2Tb Samsung 850 Pro 
Hard DriveOptical DriveCoolingCooling
Samsung SV843 960 GB LG WH14NS40 Cryorig R1 Ultimate 9x Gentle Typhoon 1850 rpm on case 
OSMonitorKeyboardPower
Windows 7 Pro x64 Korean 27" 2560x1440 Ducky Legend with Vortex PBT Doubleshot Backlit... EVGA 1300W G2 
CaseMouseAudioOther
Cooler Master Storm Trooper Logitech G502 Proteus Asus Xonar Essence STX Lamptron Fanatic Fan Controller  
  hide details  
Reply
Trooper Typhoon
(20 items)
 
  
CPUMotherboardGraphicsGraphics
5960X X99A XPower MSI r9 290X Lightning  MSI r9 290X Lightning 
RAMHard DriveHard DriveHard Drive
G.Skill Trident Z 32 Gb Samsung SM843T 960 GB Western Digital Caviar Black 2Tb Samsung 850 Pro 
Hard DriveOptical DriveCoolingCooling
Samsung SV843 960 GB LG WH14NS40 Cryorig R1 Ultimate 9x Gentle Typhoon 1850 rpm on case 
OSMonitorKeyboardPower
Windows 7 Pro x64 Korean 27" 2560x1440 Ducky Legend with Vortex PBT Doubleshot Backlit... EVGA 1300W G2 
CaseMouseAudioOther
Cooler Master Storm Trooper Logitech G502 Proteus Asus Xonar Essence STX Lamptron Fanatic Fan Controller  
  hide details  
Reply
post #945 of 1994
Quote:
Originally Posted by TheOldTechGuru View Post

From what I read, the capacitors and chokes can also have an impact in that when they are well-designed they can mitigate some of the workload of the mosfets, reducing the strain upstream (again, I suspect I am way over-simplifying this).

They aren't able to mitigate some of the MOSFETs workload but of course they are affecting the overall efficiency of a VRM.
Unfortunately usually there is no public datasheet for the chokes such as the datashets for the FETs so the saturation current and the inductance are the only informations we've got. Nobody knows how the power loss rises with the current flowing through the coil.

The power loss of the capacitors aren't a big deal nowadays, their resistance is such low that they don't heat up a lot.
post #946 of 1994
Sometimes misinformation is worse than the lack of information...

I ask for 1.406 i get 1.41 @ socket loaded on my dmm which rounds up due to lack of fine grain...

I fail to see the problem here?

https://youtu.be/xuXIT_CHdcs

If i measure @ caps i get innacurate data....solution...do not measure where your not supposed to measure....
Edited by chew* - 4/17/17 at 5:05pm
post #947 of 1994
Quote:
Originally Posted by CrazyElf View Post



sad-smiley-002.gif
post #948 of 1994
Quote:
Originally Posted by CrazyElf View Post

Thanks broda for this list:
https://www.hardwareluxx.de/community/f12/am4-mainboard-vrm-liste-1155146.html



Oh and capacity wise, Buildzoid believes that the VRM can only take about half as much (252A on X370 XPower vs 480A on X370 Taichi). Not to mention, at 252A, as Buildzoid highlights at around 12:30, it's much less efficient (64W of waste heat, assuming 1.4V and 252A @ 125C). While MSI's heatsink is probably the best design, the VRM is just not what you expect on a flagship grade board.

Ah, but the spec sheet for the CSD87350 shows it is limited to only 10 amps at 110c so his comparisons at 125c are extremely faulty. So in a "worse case scenario", the Taichi is actually worse than the MSI. Sure the CSD part performs better up until 90c, but he is the one setting the criteria. I would argue that the 125c condition is a faulty comparison point in either case, but if he is going to use it he needs to be consistent and not arbitrarily favor one part over another ignoring it's actual performance at *his* reference point.
post #949 of 1994
You should change your name to SaltyTitaniumGuru. Give it up already! rolleyes.gif
Gaming PC
(17 items)
 
  
CPUMotherboardGraphicsRAM
AMD RYZEN R7 1800X @4Ghz Asrock x370 Fatal1ty Professional Nvidia GTX 1080 G.SKILL TridentZ RGB Series 16GB (2 x 8GB) 3200... 
Hard DriveHard DriveHard DriveCooling
Samsung 840 Pro 256GB SSD Samsung 850 Pro 512GB SSD Samsung 850 EVO 1TB SSD Corsair H100i 
OSMonitorKeyboardPower
Microsoft Windows 10 Pro 64-bit Samsung 49KS8000 Corsair K70 RGB Corsair AX1200i 
CaseMouseMouse PadAudio
Cooler Master Mastercase Maker 5  Corsair M65 RGB Custom SF 49ERS mouse pad Philips X1 headphones with V-MODA BoomPro mic 
Audio
Sound BlasterX Katana 
  hide details  
Reply
Gaming PC
(17 items)
 
  
CPUMotherboardGraphicsRAM
AMD RYZEN R7 1800X @4Ghz Asrock x370 Fatal1ty Professional Nvidia GTX 1080 G.SKILL TridentZ RGB Series 16GB (2 x 8GB) 3200... 
Hard DriveHard DriveHard DriveCooling
Samsung 840 Pro 256GB SSD Samsung 850 Pro 512GB SSD Samsung 850 EVO 1TB SSD Corsair H100i 
OSMonitorKeyboardPower
Microsoft Windows 10 Pro 64-bit Samsung 49KS8000 Corsair K70 RGB Corsair AX1200i 
CaseMouseMouse PadAudio
Cooler Master Mastercase Maker 5  Corsair M65 RGB Custom SF 49ERS mouse pad Philips X1 headphones with V-MODA BoomPro mic 
Audio
Sound BlasterX Katana 
  hide details  
Reply
post #950 of 1994
Go to the english forums they said. Discussions are filled with much more arguments there they said. rolleyes.giftongue.gif
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD - General
Overclock.net › Forums › AMD › AMD - General › VRM on the new AM4 motherboards