Overclock.net › Forums › AMD › AMD Motherboards › ROG Crosshair VI overclocking thread
New Posts  All Forums:Forum Nav:

ROG Crosshair VI overclocking thread - Page 860

post #8591 of 17379
Quote:
Originally Posted by Reikoji View Post

weird that plain balanced plan could actually increase performance in gaming, tho AMD said its core parking settings was a cause of reduced performance. maybe in a select few games, on a good day.

I think it has to do with balanced and core parking. It parks half the cores, theoretically making it a 4/8 cpu. Problem with high performance is how randomly windows will send instructions from one ccx to another (and therefore introducing latency and possibly even making it worse) but again it is VERY dependent more so on the game then actually on windows (even though they really should do some tweaking with Ryzen and the hp mode).

Heres actual proof in gaming (although its only one, its still there) that it could matter in a game.

i5 4690k/r7 1700x/r7 1700x 3.8ghz, 2933mhz(hp)/balanced
Grid Autosport.
Min 86.98 / 69.90 / 82.35 / 82.733
Avg 111.84 / 95.30 / 107.15 / 107.68
Max 156.24 / 132.85 / 139.34 / 142.75

Tomb Raider is the same story in 2 of the 3 benchmarks, its within a margin of error though but grid isn't.

Rise of the Tomb Raider (DX12)

Mountain Peak
Min 47.58 / 45.76 / 48.81 / 46.20
Avg 84.44 / 68.61 / 84.60 / 82.99
Max 126.17 / 119.18 / 138.08 / 145.87

Syria
Min 18.75 / 37.06 / 35.25 / 32.07
Avg 64.71 / 71.00 / 64.55 / 64.98
Max 77.93 / 97.79 / 82.97 / 91.23

Geothermal Valley
Min 34.96 / 30.33 / 41.99 / 42.41
Avg 58.27 / 63.29 / 58.64 / 58.79
Max 79.48 / 79.69 / 73.78 / 74.90
Overall 69.38 / 67.43 / 69.54 / 69.14
Edited by bluej511 - 4/11/17 at 9:05pm
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
post #8592 of 17379
Quote:
Originally Posted by bluej511 View Post

I think it has to do with balanced and core parking. It parks half the cores, theoretically making it a 4/8 cpu. Problem with high performance is how randomly windows will send instructions from one ccx to another (and therefore introducing latency and possibly even making it worse) but again it is VERY dependent more so on the game then actually on windows (even though they really should do some tweaking with Ryzen and the hp mode).

Balanced at stock actually parks every core. The minimum cores to unpark starts at 0% and thats why you consistantly see a drop in single core performance with every run.

Doing multiple runs, when you have the minimum above 0%, at any reasonable level, the processor seems like it tries and spool up with a parked core instead of an unparked one, and instead of going for a core that isn't parked it will continue with that one, and you will see the score jump around a lot before it finally stabilizes, giving random results. 100% and 0% unparked cores gives near the same score every time.

Does ryzen's balanced power plan allow core parking in it?
post #8593 of 17379
Quote:
Originally Posted by Reikoji View Post

Balanced at stock actually parks every core. The minimum cores to unpark starts at 0% and thats why you consistantly see a drop in single core performance with every run.

Doing multiple runs, when you have the minimum above 0%, at any reasonable level, the processor seems like it tries and spool up with a parked core instead of an unparked one, and instead of going for a core that isn't parked it will continue with that one, and you will see the score jump around a lot before it finally stabilizes, giving random results. 100% and 0% unparked cores gives near the same score every time.

Does ryzen's balanced power plan allow core parking in it?

If i watch hwinfo while gaming, core 0-3 gets utilized and 4-7 gets ZERO and i mean ZERO percent. So its clearly favoring 0-3 in balanced, that could work in most games that aren't optimized for more then 4 cores. As you saw above in Grid theres some improvement. I haven't downloaded my other games yet.

In siege the averages between hp and ryzen mode were identical, the differences? in min and max (which are far more important then just avg alone) so there is definitely something going on.

I may try siege with all 3 power modes and see whats up.

P.S. I just checked balanced power mode, since with your help i was able to unlock the setting in power option, and the min is actually 10% not 0% so correction for you. I may try 50% core parking in hp mode and see what it does but thats for a different day.
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
post #8594 of 17379
Quote:
Originally Posted by bluej511 View Post

If i watch hwinfo while gaming, core 0-3 gets utilized and 4-7 gets ZERO and i mean ZERO percent. So its clearly favoring 0-3 in balanced, that could work in most games that aren't optimized for more then 4 cores. As you saw above in Grid theres some improvement. I haven't downloaded my other games yet.

In siege the averages between hp and ryzen mode were identical, the differences? in min and max (which are far more important then just avg alone) so there is definitely something going on.

I may try siege with all 3 power modes and see whats up.

P.S. I just checked balanced power mode, since with your help i was able to unlock the setting in power option, and the min is actually 10% not 0% so correction for you. I may try 50% core parking in hp mode and see what it does but thats for a different day.

Ah yea thats right. But 10% of 8 might as well be 0 :3 how would it explain the reduced single core performance if its only utilizing 0-3? what if 0-3 happen to be one or more the parked ones?

hmm you may be right about 50%. At that amount it doesn't seem like 0-3 are selected for parking. Seems like a good number to start, but anything below 50% looks like it will have a performance hit.
Edited by Reikoji - 4/11/17 at 9:33pm
post #8595 of 17379
Quote:
Originally Posted by MNMadman View Post

Got two crap overclocker 1700 chips here ... but they both can do 3200 all day every day on 0081 BIOS. Good trade I guess...

First 1700 I had. 3.85GHz stable at 1.381v, RAM at 3200 14-14-14-14-34-1T at 1.35v and 1.35v DRAM Boot. SoC at 0.95v.


New 1700. 3.85GHz stable at 1.362v, RAM at 3200 14-14-14-14-34-1T at 1.35v and 1.35v DRAM Boot. SoC at 0.969v.


Both boards and CPUs do the several-second pauses under stress tests, so it does seem to be a software/driver issue of some sort. Still need to try a clean install of Win10 Creator.

same here
realbench stress test with playing vid ... then sometimes pause and wind fast but pass test
clean install either
post #8596 of 17379
Quote:
Originally Posted by Gadfly View Post

Not really a board limitation from what I have observed. So I guess this is a good time to introduce my theory about memory clocks. I have two CH6's, two 1800x, and one 1700 that belong to a buddy of mine. We have 4 ram kits, all G.skill, 1ea 2x8 3200 Cas 14, 2ea 2x8 3600 Cas 16, and 1ea 2x8 4266 C19.


Here is my initial observation:

On both boards all 4 memory kits will run 3200 CL14 without any issues on both 1800x. On both boards, and all 4 memory kits, the 1700 struggles to hit over 2933 without some serious SOC and DRAM voltage increases.. So 4 memory kits, 2 motherboards, the memory results are the same for each CPU no matter how you mix and match them.

So from our testing it appears the CPU is the deciding factor. Which after some thinking and reviewing of AMD's materials, started to make ALOT of sense to me. I am in the opinion that what we are seeing is limitations in the data fabric (or Infinity Fabric depending on who you ask). In talking to a few people I have also began to notice a correlation between max stable OC and memory speeds. It appears the CPU's that can hit 4.0ghz on lower Vcores also seem to be the CPU's that are able to hit higher memory speeds.

I think a big part of this architecture that no one has been talking about is that the data fabric runs at
DRAM clock (half DDR). This fabric is a series of interconnects traced onto the CPU die, and just like all traces on a die, some are cleaner than others. Contaminants, thickness, slight narrowing of the trace, etc. all impact how well electrons move down the trace, and as a result effect what the maximum frequency it can sustain.

Just like with CPU overclocking ability, this is a lottery. On cleaner CPU's we are seeing higher clock speeds, and on those cleaner dies the fabric is cleaner and is able to run at higher frequencies; where on the lower binned CPU's, those that are a little dirtier, the clocks are lower, and the fabric is unable to support higher memory clocks.

So I believe the lottery on Ryzen is not only for CPU clock, but for memory clock as well. We know (or at least strongly suspect) that the higher end CPU's get the better bins, which is why a much greater percentage of 1800x CPU's can hit 4.0Ghz at lower voltages than the 1700x or 1700's. If my theory is correct, it might also suggest that the lower end CPU's are going to struggle to hit the higher DRAM clocks as well.

I would love some feedback on what you guys think.

My first 1700x did not have an issue hitting 3200 14-14-14-14-34-1T, 1.35v, SOC on auto 1.11v no cold boot issues or ram boot up issues. Stable at default or low clock speeds. CPU had a hard time OCing above 3.8ghz even with 1.45v, just did not matter. Second 1700x, same with ram, no issue, had it up to 3500mhz with BCLK. 2nd 1700x can easily hit 4ghz OC.

So for ram it maybe independent of OC ability of the cpu as well.
Destroyer
(13 items)
 
  
Reply
Destroyer
(13 items)
 
  
Reply
post #8597 of 17379
Quote:
Originally Posted by bluej511 View Post

I think it has to do with balanced and core parking. It parks half the cores, theoretically making it a 4/8 cpu. Problem with high performance is how randomly windows will send instructions from one ccx to another (and therefore introducing latency and possibly even making it worse) but again it is VERY dependent more so on the game then actually on windows (even though they really should do some tweaking with Ryzen and the hp mode).

Heres actual proof in gaming (although its only one, its still there) that it could matter in a game.

i5 4690k/r7 1700x/r7 1700x 3.8ghz, 2933mhz(hp)/balanced
Grid Autosport.
Min 86.98 / 69.90 / 82.35 / 82.733
Avg 111.84 / 95.30 / 107.15 / 107.68
Max 156.24 / 132.85 / 139.34 / 142.75

Tomb Raider is the same story in 2 of the 3 benchmarks, its within a margin of error though but grid isn't.

Rise of the Tomb Raider (DX12)

Mountain Peak
Min 47.58 / 45.76 / 48.81 / 46.20
Avg 84.44 / 68.61 / 84.60 / 82.99
Max 126.17 / 119.18 / 138.08 / 145.87

Syria
Min 18.75 / 37.06 / 35.25 / 32.07
Avg 64.71 / 71.00 / 64.55 / 64.98
Max 77.93 / 97.79 / 82.97 / 91.23

Geothermal Valley
Min 34.96 / 30.33 / 41.99 / 42.41
Avg 58.27 / 63.29 / 58.64 / 58.79
Max 79.48 / 79.69 / 73.78 / 74.90
Overall 69.38 / 67.43 / 69.54 / 69.14

was the balanced with 10% min unparked or was it changed to 50%?
post #8598 of 17379
Quote:
Originally Posted by Gadfly View Post

New issue:

Cpu ratios and vcore will not save in bios.

I can save the ratios and core voltage and reboot, board will post but will be at default clock speed and voltage.

Anyone seen this before?

I tried reflashing the BIOS to several different versions, clearing cmos, removing batter for 10 min with power unplugged, USB flashback, etc.

Motherboard is dead. I have a second CH6, swapped over all the same components on my "test bench" and everything fired up first try. All my boot, post, and bios issues are resolved.

Guess I will see if Fry's will let me exchange it, or I might have to RMA it, but I hear that takes weeks.
post #8599 of 17379
Quote:
Originally Posted by Reikoji View Post

was the balanced with 10% min unparked or was it changed to 50%?

These are a week old unchanged so its just balanced to 10% stock balanced power option.

i believe mnmadman or wtv his name is has tested with 0-50-100% core parking but not sure what his results were, it was only synthetic benchmarks if i recall.

Sounds like 50% might be the sweet spot for Ryzen but again totally dependent on the application.
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
R7 1700X asus crosshair 6 Sapphire R9 390 Nitro CMK16GX4M2B3200C16 
Hard DriveHard DriveCoolingCooling
Kingston HyperX 3K Seagate ST2000DM006 EKWB Supremacy Evo EK-XRES 100 Revo D5 - Acetal 
CoolingCoolingCoolingCooling
Alphacool GPX r9 390 m01 alphacool xt45 240 Alphacool xt45 360 Alphacool D5 PWM 
OSMonitorKeyboardPower
Windows 10 LG 29UM68 Corsair Strafe Corsair RM1000 
CaseMouseMouse Pad
Thermaltake X5 Corsair M65 Pro RGB Razer Goliathus Speed 
  hide details  
Reply
post #8600 of 17379
I've been working on tuning the offset voltage on a 4 GHz P-state overclock. Started at .26v based on calculation, but the result looked a bit higher than necessary. After bumping downward in several steps, .23v failed stress testing, and a bump to .234v proved stable.

That generates 1.119V at idle (1550MHz or so) such as while typing this and under stress testing 1.40V. Did not not change LLC from my manual overclock. Using the Realbench Stress test CPU Tctl hit a peak of 77.8 C and an average of 72.6 C.

No problems came up with the memory at 3200MHz, 14-14-14-14-34. I wound up using the Ryzen Power Plan - Maximum processor frequency left at 0 MHz, Minimum processor state at 5%, Maximum processor state at 100%. Don't know if that is optimal or not. I do notice that after booting in to Windows, the CPU runs at 4GHz for perhaps two minutes before down-clocking. After that it seems to ramp up and down as expected.

It seems doubtful I can wring much more from a memory performance perspective until we get higher divisors - maybe a bit tighter timings, have not tried that. Unless some future magic allows higher CPU frequency without raising temperatures, that won't increase for me. Might squeeze a .001 or .002 decrease in CPU volts offset, if I'm lucky.

This is intended to be my "daily do everything, set and forget" overclock setting. For such a new platform, I really am very happy so far. Think I'll play with my pretty Aura effects for a bit. biggrin.gif

New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: AMD Motherboards
Overclock.net › Forums › AMD › AMD Motherboards › ROG Crosshair VI overclocking thread