RIVEBE + 4960X + Windows 8.1 vs Linux - OC not sticking

[Vegim]

Here's a quick summary of what's going on: I have a 4.7GHz overclock that works fine and primed for 48 hours in Windows 8.1. However, using any flavor of Linux, the overclock does not "register". CPU reporting tools also verify this, along with benchmark tests.

Also, I've cross posted this at another forum I frequent so my apologies if your seeing this twice. 🙂

System Specs

• Motherboard: Rampage IV Extreme Black Edition bios 0701
  
• CPU: Intel i7 4960X
  
• Memory: G.Skill Trident 32GB Kit F3-1600C7Q-32GTX
  

There's a custom water loop and the machine is powered by SSDs.

The Overclock

Extreme Tweaker


 
 BCLK 100
 
 CPU Core Ratio:  47 synced across all cores
 
 DRAM Frequency:  1600Mhz
 
 VCore: 1.4v
 
 VTT CPU Voltage:  1.1
 
 2nd VTTCPUY Voltage:  1.10387
 
 CPU VCCSA Voltage:  Manual, 1.1v
 
 DRAM Voltage:  Stock @ 1.5v per memory req.
 
 CPU PLL Voltage:  1.8v

DIGI+ Power Control


 
 CPU Load-line Calibration Ultra High
 
 CPU Current Capability:  140%
 
 CPU Voltage Frequency:  Manual, 350KHz
 
 VRM Over Temperature Protection: Disabled
 
 CPU Power Duty Control:  Extreme
 
 Vcore MOS volt. Control:  7v
 
 Current Inrush Inertia:  Auto
 
 CPU VCORE Boot Up Voltage:  1.4v
 
 CPU VSA
 
 VCCSA Load-line Calibration:  Extreme
 
 VCCSA Current Capability:  130%
 
 VCCSA Fixed Frequency:  550
 
 VCCSA Boot Up Voltage:  1.1v

HT disabled, EIST disabled, CPU C states disabled
All extra stuff disabled. Sound, ASMedia controllers, USB3, etc. Only thing left on was Intel NIC PXE boot

Windows 8.1 64bit

CPU-Z was reading a strong 4.7GHz overclock @ 1.4v (in software, did not multimeter it). Prime95 for 48 hours, everything good. 4.8GHz crashed at 1.5v (2 hours in) even so I thought 4.7 @ 1.4 is damn good.

The system is rock solid in Windows. Everything works as expected.

Even did some SuperPi 1M runs for fun - 4.7GHz hit 7.937s while 3.6GHz (stock) hit 10.343s.

So everything is dandy in Windows

Any flavor / kernel of Linux

The overclock does not seem to "hit". No matter what I do, I am always seeing 3.6GHz. In any flavor of Linux. Ubuntu, Xubuntu, Fedora, Debian (all desktop/server .. either or, your pick).

I've tried two tools to tell me CPU frequency, one is i7z and the other is i-nex. I prefer i7z just because it's what I know and have used for a long time.

Here's the i7z output on start:

i7z DEBUG: i7z version: svn-r93-(27-MAY-2013)
i7z DEBUG: Found Intel Processor
i7z DEBUG:    Stepping 4
i7z DEBUG:    Model e
i7z DEBUG:    Family 6
i7z DEBUG:    Processor Type 0
i7z DEBUG:    Extended Model 3
i7z DEBUG: msr = Model Specific Register
i7z DEBUG: detected a newer model of ivy bridge processor

And here's the i7z output when it's showing output:

Cpu speed from cpuinfo 3599.00Mhz
cpuinfo might be wrong if cpufreq is enabled. To guess correctly try estimating via tsc
Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 3599 MHz
  CPU Multiplier 36x || Bus clock frequency (BCLK) 99.97 MHz

Socket [0] - [physical cores=6, logical cores=6, max online cores ever=6]
  TURBO ENABLED on 6 Cores, Hyper Threading OFF
  Max Frequency without considering Turbo 3698.97 MHz (99.97 x [37])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4/5/6 Cores is  47x/47x/47x/47x/47x/47x
  Real Current Frequency 3599.00 MHz [99.97 x 36.00] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %  Temp	  VCore
        Core 1 [0]:	  3599.00 (36.00x)	 100	   0	   0	   0    29	1.1259
        Core 2 [1]:	  3599.00 (36.00x)	 100	   0	   0	   0    34	1.1259
        Core 3 [2]:	  3599.00 (36.00x)	 100	   0	   0	   0    33	1.1259
        Core 4 [3]:	  3599.00 (36.00x)	 100	   0	   0	   0    37	1.1259
        Core 5 [4]:	  3599.00 (36.00x)	 100	   0	   0	   0    33	1.1259
        Core 6 [5]:	  3599.00 (36.00x)	 100	   0	   0	   0    32	1.1259

C0 = Processor running without halting
C1 = Processor running with halts (States >C0 are power saver modes with cores idling)
C3 = Cores running with PLL turned off and core cache turned off
C6, C7 = Everything in C3 + core state saved to last level cache, C7 is deeper than C6
  Above values in table are in percentage over the last 1 sec
[core-id] refers to core-id number in /proc/cpuinfo
'Garbage Values' message printed when garbage values are read
  Ctrl+C to exit

So - I know what you're thinking. The reporting tools are probably wrong because this CPU is newer than when the code was last updated, blah blah blah. Yes, I thought of this as well and covered my tracks. So in Windows I did multiple tests to check performance. I could tell the OC was real and I was seeing that 25% performance gain.

TURBO ENABLED on 6 Cores, Hyper Threading OFF
Max Frequency without considering Turbo 3698.97 MHz (99.97 x [37])
Max TURBO Multiplier (if Enabled) with 1/2/3/4/5/6 Cores is 47x/47x/47x/47x/47x/47x

So it sees the TURBO multiplier is 47 and that TURBO is enabled, but for some reason I'm not getting the performance.

Benchmarks

I've used sysbench to test the OC to maybe see if the reporting tools were wrong like I mentioned earlier. Having the BIOS multi set @ 36 or 47 yielded the same results

Sysbench 3.6GHz


sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.10:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          23.5980s
    total number of events:              10000
    total time taken by event execution: 23.5970
    per-request statistics:
         min:                                  2.33ms
         avg:                                  2.36ms
         max:                                  9.59ms
         approx.  95 percentile:               2.35ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   23.5970/0.00

Sysbench 4.7GHz

sysbench --test=cpu --cpu-max-prime=20000 run
sysbench 0.4.10:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 1

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000


Test execution summary:
    total time:                          23.5980s
    total number of events:              10000
    total time taken by event execution: 23.5970
    per-request statistics:
         min:                                  2.33ms
         avg:                                  2.36ms
         max:                                  9.59ms
         approx.  95 percentile:               2.35ms

Threads fairness:
    events (avg/stddev):           10000.0000/0.00
    execution time (avg/stddev):   23.5970/0.00

Now the real kicker: This exact same "OS / Software / Setup" with a RIVE + 3970X reports correctly on high overclocks with any type of overclock - multi or bclk

Is it possible that there's something that ASUS did in the BIOS (most likely on accident) to not allow the overclock to "register" in Linux?

I'm at a bit of a loss, if anyone has any ideas, please share.

[Vegim]

Bringing some more information to the table. If I use a CPU strap, the frequency is reporting correctly (however I don't know why it says the multiplier is @ 37x .. when it's at 36x.

This makes more sense now because I could never get a CPU strap based overclock working (where as with the 3970X, all day...)

So if the OC is not taking in the OS, then the voltage would be wrong or auto as well. So a 4.5GHz overclock would crash immediately due to being undervolted, which is what it does.

Cpu speed from cpuinfo 4500.00Mhz
cpuinfo might be wrong if cpufreq is enabled. To guess correctly try estimating via tsc
Linux's inbuilt cpu_khz code emulated now
True Frequency (without accounting Turbo) 4500 MHz
  CPU Multiplier 36x || Bus clock frequency (BCLK) 125.00 MHz

Socket [0] - [physical cores=6, logical cores=6, max online cores ever=6]
  TURBO ENABLED on 6 Cores, Hyper Threading OFF
  Max Frequency without considering Turbo 4625.00 MHz (125.00 x [37])
  Max TURBO Multiplier (if Enabled) with 1/2/3/4/5/6 Cores is  37x/37x/37x/37x/37x/37x
  Real Current Frequency 4499.00 MHz [125.00 x 35.99] (Max of below)
        Core [core-id]  :Actual Freq (Mult.)	  C0%   Halt(C1)%  C3 %   C6 %  Temp	  VCore
        Core 1 [0]:	  4499.00 (35.99x)	 100	   0	   0	   0    36	1.1208
        Core 2 [1]:	  4499.00 (35.99x)	 100	   0	   0	   0    39	1.1208
        Core 3 [2]:	  4499.00 (35.99x)	 100	   0	   0	   0    37	1.1208
        Core 4 [3]:	  4499.00 (35.99x)	 100	   0	   0	   0    41	1.1208
        Core 5 [4]:	  4499.00 (35.99x)	 100	   0	   0	   0    38	1.1208
        Core 6 [5]:	  4499.00 (35.99x)	 100	   0	   0	   0    37	1.1208

C0 = Processor running without halting
C1 = Processor running with halts (States >C0 are power saver modes with cores idling)
C3 = Cores running with PLL turned off and core cache turned off
C6, C7 = Everything in C3 + core state saved to last level cache, C7 is deeper than C6
  Above values in table are in percentage over the last 1 sec
[core-id] refers to core-id number in /proc/cpuinfo
'Garbage Values' message printed when garbage values are read
  Ctrl+C to exit

[HiVizMan]

I am not sure it is possible to actually OC for some Linux, I do recall quite some time ago seeing a similar thread about OCing not working on the RIVE board when it first came out. Will try and find the thread and see how it turned out.

[Vegim]

HiVizMan wrote:
I am not sure it is possible to actually OC for some Linux, I do recall quite some time ago seeing a similar thread about OCing not working on the RIVE board when it first came out. Will try and find the thread and see how it turned out.

I've personally OC'd and had prime stable boxes with the original RIVE + 3970X .. 4.8GHz-5.1GHz. I had a custom BIOS from ASUS at one point to address an issue with TSC, but that was fixed after a certain BIOS.

This should indeed work. There is something off here with the overclocking being recognized in Linux.

Why is BCLK working but multiplier not working?

[Nate152]

Hello Vegim,

I remember reading something about this too, try updating the kernel for Linux and see if that solves your issue.

[Vegim]

Nate152 wrote:
Hello Vegim,

I remember reading something about this too, try updating the kernel for Linux and see if that solves your issue.

This is something I have not tried yet but was curious if it would work. I know for a fact whatever the default kernel for Xubuntu 14.04, that did the same behavior.

I'll try updating the kernel tomorrow to see if that does the trick.

[Nate152]

OK, great ! Let us know how it goes.

[Vegim]

Nate152 wrote:
OK, great ! Let us know how it goes.

Testing, testing, testing...

I'll cut down to the chase. EIST is the culprit. Xubuntu 14.04 works fine (3.13.x kernel) with EIST set to Enabled or Auto.

However, older kernels with either EIST enabled or disabled still don't work.

I'm still pointing fingers at the board. The red boards with 3970X work flawlessly. I'm currently getting a red board together with a BIOS that supports the 4960X and am going to perform these tests.

[Vegim]

Nate152 wrote:
OK, great ! Let us know how it goes.

Just updating this in case someone ends up running in to the same problem. It is indeed a bug in the BIOS that isn't going to get fixed for the Black Board. It is recommended if you're running Linux that you leave EIST enabled or auto.