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 119

post #1181 of 1994
Not all 100%s are created equal.
post #1182 of 1994
Quote:
Originally Posted by virpz View Post

I can get 100% utilization with both tests but obviously they do not run the same on Ryzen... In my experience the OCCT in the before mentioned settings ( pick your time, that was just an example ) is putting my Ryzen to sweat a little more if compared to prime95, no matter FFT sizes I set.

I've been using prime95 for way more than a decade now but I sure can still learn some tricks about it. Mind to teach me ? tongue.gif

Nyet, I poke fun of ten minute test. Length of reply was to other user asking about my temps after blender for 30.
Best prime? Probably blend custom 90% ram allocated, staggered start to eliminate synchronous load reduction.would add some variety to watching per core power useage. Small ffts exclusively tested skips out on ram testing. Not entirely but eh it does not push any large cache arch enough. Not just ryzen. My way I get some specificity in errors returned that can point toward ram vs cpu and retest with other software, settings accordingly. I'm open to other general tests though, think my notes have a couple edge cases that took longer runs to show instability and of course to make it even I will flip it around and try ... whatever the recommended flavor was. Tablet sucks for reviewing. Is there anything it is especially good at finding fast? I mean if it turns up more specificity like ram needing a touch more love in ten minutes that'd be awesome.
Edited by yendor - 4/27/17 at 4:55pm
post #1183 of 1994
Quote:
Originally Posted by KarathKasun View Post

Power is a function of amps and volts. So the two are directly related given that we know the voltage range for the chips because current times voltage equals power.

Ryzen does not seem to be able to scale with voltage to the point where the VRM will be the main limiting point. Very high 3ghz OC's are possible with the eight core chips on four phase (4+2) boards without much issue.
Except vrms itself don't care for voltage.


VRMs can be limiting in high OCs not because of current rating, but rather heat output. If you have a poor 3+2 mobo without any heatsinks on the vrms, you might not hit the current limit of the vrms, but you could cook them because of insufficient cooling.
BigBox
(16 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5 3570K  Asus P8Z77-V MSI RX480 Gaming X 8GB Corsair Vengeance 
Hard DriveHard DriveOptical DriveCooling
Seagate 2.5" Intel SSD 330 LiteON @ OptiArc DVD-RW Noctuanh-U12P push/pull 
OSMonitorKeyboardPower
Windows 10 64bit AOC I2360PHU Logitech Internet Pro EVGA SuprNova G2 650W 
CaseMouseMouse PadAudio
Antec Three Hundred MS Wireless Optical Mouse 2.0A OEM Onboard 
  hide details  
Reply
BigBox
(16 items)
 
  
CPUMotherboardGraphicsRAM
Intel Core i5 3570K  Asus P8Z77-V MSI RX480 Gaming X 8GB Corsair Vengeance 
Hard DriveHard DriveOptical DriveCooling
Seagate 2.5" Intel SSD 330 LiteON @ OptiArc DVD-RW Noctuanh-U12P push/pull 
OSMonitorKeyboardPower
Windows 10 64bit AOC I2360PHU Logitech Internet Pro EVGA SuprNova G2 650W 
CaseMouseMouse PadAudio
Antec Three Hundred MS Wireless Optical Mouse 2.0A OEM Onboard 
  hide details  
Reply
post #1184 of 1994
Quote:
Originally Posted by hojnikb View Post

Except vrms itself don't care for voltage.


VRMs can be limiting in high OCs not because of current rating, but rather heat output. If you have a poor 3+2 mobo without any heatsinks on the vrms, you might not hit the current limit of the vrms, but you could cook them because of insufficient cooling.

Its a 3+2 with a heatsink on the VCore phases (this item). And the 3 phases are sufficiently efficient to push ~70-80A while only dumping ~15w of heat, which should be perfectly fine with even a tiny heatsink. The only real problem of a 3 phase board is that the power delivery will be noisier, requiring slightly higher voltages to hit the same speeds that a true 6+ phase board would need.

To put that current capability in perspective... the R7 8 core pulls around 100A (~140w) @ 4ghz, R5 1600 and its X variant pull around 70-80A (~105w) @ 4ghz, and R5 1400/1500x pull 55-65A (~84w) @ 4ghz. With an R5 1400/1500x, a bare 4 phase VRM would be fine for even chip destroying voltages and overclocks.

Ryzen simply does not scale far enough with voltage for VRM's to be a problem, it is the polar opposite of Bulldozer. Likely due to the heritage of its 14nm process.

Also, higher voltages require more on time in the FETs. Which can massively increase temps, meaning that they absolutely respond to desired output voltage. Power is a measure of the whole FET throughput situation and is a much better way to look at real world capability than simply looking at amperage IMHO.
Edited by KarathKasun - 4/28/17 at 1:25am
μRyzen
(12 items)
 
Mini Box
(4 items)
 
 
CPUMotherboardGraphicsRAM
Ryzen R5 1400 MSI B350M Gaming Pro Zotac GTX 670 4GB G.SKILL FORTIS Series 8GB (2 x 4GB) 
Hard DriveCoolingOSOS
WD Green 3tb Wraith Stealth Windows 10 Debian 8.7 
MonitorKeyboardPowerMouse
ViewSonic VX-2257-8 Chinese backlit mechanical Kingwin 850w Chinese laser optical 
CPUMotherboardGraphicsRAM
Athlon 5350 Asus AM1I-A EVGA GTX 750 Ti SC 2x4GB DDR 3 1333 
  hide details  
Reply
μRyzen
(12 items)
 
Mini Box
(4 items)
 
 
CPUMotherboardGraphicsRAM
Ryzen R5 1400 MSI B350M Gaming Pro Zotac GTX 670 4GB G.SKILL FORTIS Series 8GB (2 x 4GB) 
Hard DriveCoolingOSOS
WD Green 3tb Wraith Stealth Windows 10 Debian 8.7 
MonitorKeyboardPowerMouse
ViewSonic VX-2257-8 Chinese backlit mechanical Kingwin 850w Chinese laser optical 
CPUMotherboardGraphicsRAM
Athlon 5350 Asus AM1I-A EVGA GTX 750 Ti SC 2x4GB DDR 3 1333 
  hide details  
Reply
post #1185 of 1994
I use prime because 90% of stability on amd is and has been directly related to imc since integrated imc was first introduced. Amds imc has always been rather finicky compared to intel.

Using tests that stress core just dont exhibit the issues im looking for.

Blend-'->custom runs iterations not possible to hit under any of the presets. 85%-90% ram allocated is a must...or your wasting your time.
The other advantage is various load on/off. On/off load can find gremlins that sustained load can not. That itself effects boards more than cpu and llc but...board not stable...cpu not stable...on/off heavy load is also more real world. Example idle solidwork task...idle game...etc etc..

For extra measure tossing some 3d in once your finally stable is a good idea. Especially if you understand what soc actually does and controls on the chip.
Edited by chew* - 4/28/17 at 4:12am
post #1186 of 1994
It is finicky but once you find the ideal stable timings for the ram then it's golden, performance and stability is way way up there and enjoyable. Yeah the IMC is picky on AMD but it's a strong performer when the timings/ speed and modules are dialed in just right. Samsung memory seems to provide the highest stability and performance, headache free operation.

Sometimes Auto is ideal, auto timings... Compare to your modules timings but also compare with Auto timings... Auto is actually the fastest as long as the clock/ memory multi is just right.

For increasing the overclocked stability further is to turn up the voltages except the memory and the southbridge a couple ticks... This helps a lot.

The one picky thing about the ram is even know it passes a stress test or passes 5 hours of MemTest86, it could easily still be unstable for applications. That's when I found that Kingston was the issue. With Samsung everything works great. As far as I know Corsair are possibly the only manufacturer using Samsung modules, correct me if I'm wrong.
Edited by chris89 - 4/28/17 at 5:35am
post #1187 of 1994
everyone is using Samsung, gskill,geil,corsair,crucial.

there is more to the memory tuning than just the 5 timings.

If you have ref clock you have access to sub timing tuning.

This is why I have been testing ref clock 120.

so far every board at 2666 strap is faster with 3200 than 3200 strap.

prime95 when used right picks up those instabilities instantly.

it drops threads its IMC/memory...it crashes/bsod its core
post #1188 of 1994
Quote:
Originally Posted by bardacuda View Post

Exactly.

@chew*

I watched the video but it looked like you were measuring the back of the chokes or caps. If your IR thermometer is anything like mine, it actually measures the temp about 3/4" "below" where the laser is, but you had the unit rotated to the left, so I think it was measuring the temp to the right of the laser.

Maybe I just find it hard to believe the heatsink could be that ineffective. I could see if there were fets on the low side that didn't have a sink/airflow on them but that wasn't the case for that board as far as I could tell. I really think the sensor might be reporting choke temps.

If you noticed parts of the vrm run hotter. It seems that the reference point for temps is...peak hottest.

I noticed similar on taichi....parts of the vrm run hotter. Its not even all the way around albeit cooler there are hotter areas.

Taichi was hotter near the doubler where the soc/vcore heatsinks meet. 2 of the vcore fets actually sit under soc heatsink i believe...

K7 hotter completely opposite at the very end of vcore vrm sink.

My biggest concern since i run on a bench is some user with poor airflow in a case with no active cooling on vrm..

I always have to keep this is mind versus how i test.

I have yet to pop a fet on any board but at this point if i did on ln2 i could 90% nail which fet popped while benching on ln2 before removing sink to verify.

This is normal procedure for me.

I hammer boards extensively on air prior to cold testing to see if they are worth my time.

If i have to rma something it needs to be pre cold...once i prep board rma is not an option.. only a person of poor character rmas a board they ran on ln2...

I have connections and can ask...but im honest...hey i ran on ln2 it died....can you replace it? Yes/no if no go buy another.

Many times answer has been yes but write this on rma slip....

Currently the only board worth the time cold is ch6.

Taichi and k7 have to many bios issues.

Hardware wise i am not to concerned...both should hold up....current bios on both would make benching infuriating however and a waste of ln2.

When you see the board under my phase change unit....its ready...
Edited by chew* - 4/28/17 at 6:25am
post #1189 of 1994
Quote:
Originally Posted by chew* View Post

If you noticed parts of the vrm run hotter. It seems that the reference point for temps is...peak hottest.

I noticed similar on taichi....parts of the vrm run hotter. Its not even all the way around albeit cooler there are hotter areas.

Taichi was hotter near the doubler where the soc/vcore heatsinks meet. 2 of the vcore fets actually sit under soc heatsink i believe...

K7 hotter completely opposite at the very end of vcore vrm sink.

My biggest concern since i run on a bench is some user with poor airflow in a case with no active cooling on vrm..

I always have to keep this is mind versus how i test.

Is Taichi vrm area hotter than the K7?
PURPLEHAZE
(13 items)
 
  
CPUMotherboardGraphicsRAM
AMD Ryzen R7 1700 Asrock X370 Taichi Palit GTX 1080 Super JetStream 16GB KFA2 HOF 3600MHz 
Hard DriveCoolingOSMonitor
2x Crucial MX300 750GB Corsair H110i GT Windows 10 Pro 64 bit Iiyama G-Master Red Eagle GB2488HSU-B2 
KeyboardPowerCaseMouse
Corsair K70 Lux MX Red RGB Corsair RM750x Phanteks Enthoo Pro M Logitech MX-Master 
  hide details  
Reply
PURPLEHAZE
(13 items)
 
  
CPUMotherboardGraphicsRAM
AMD Ryzen R7 1700 Asrock X370 Taichi Palit GTX 1080 Super JetStream 16GB KFA2 HOF 3600MHz 
Hard DriveCoolingOSMonitor
2x Crucial MX300 750GB Corsair H110i GT Windows 10 Pro 64 bit Iiyama G-Master Red Eagle GB2488HSU-B2 
KeyboardPowerCaseMouse
Corsair K70 Lux MX Red RGB Corsair RM750x Phanteks Enthoo Pro M Logitech MX-Master 
  hide details  
Reply
post #1190 of 1994
Quote:
Originally Posted by chew* View Post

everyone is using Samsung, gskill,geil,corsair,crucial.

there is more to the memory tuning than just the 5 timings.

If you have ref clock you have access to sub timing tuning.

This is why I have been testing ref clock 120.

so far every board at 2666 strap is faster with 3200 than 3200 strap.

prime95 when used right picks up those instabilities instantly.

it drops threads its IMC/memory...it crashes/bsod its core

Yeah? That's saying the IMC isn't meant to run that fast but can be done by increasing the Northbridge/ CPU NB/ CPU voltages a slight bit. I notice this on Opteron 1389, it's IMC is really loving 1333 and nothing higher but can go up near 1500Mhz and can boot at 1600Mhz. Just out of it's range of compatibility. Though to get these memory speeds stable needs extra voltage on the CPU IMC and Northbridge/ Southbridge.

Basically to even boot 1600Mhz on a 1333 IMC is to using timings from 1-3 multi's higher on your desired memory multi, make sense? Needs looser timings than the speed is said to work at is the only way it can handle higher speeds.
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