AC68 C1 Revision (RT-AC1900) - Overclock gives strange results - Underclocks from 1Ghz to 100Mhz!!

[Zimtu]

Hi,

I have a new ASUS router RT-AC1900. From the FCCID and the label on the box, it seems to be a rebadged AC68U H/W Ver C1 with the C0 cpu stepping.

Out of the box the CPU speed was 1Ghz with 1998 Bogomips. Since it is hardware version C1, I upgraded to Merlin 59 successfully, which supports AC68 H/W ver C1.

When I tried to overclock through init-start/services-stop script to 1200 Mhz, I got a strange result - nvram value of clkfreq would stick to 1200,666 after reboot, but Bogomips went down to 199!!! I also checked with openvpn speed aes-128-cbc. Earlier, at 1Ghz, it was approx. 36Mb/sec. Now it is 3.6Mb/sec!!

CFE version shows 1.1.1.2

So, it seems overclocking the C1 h/w version of AC68U results in CPU getting underclocked from 1Ghz to 100Mhz!!

I repeated the same with Merlin 57 with same results.

This is getting to be interesting!

A few questions.

1. Does anybody have an explanation for the above underclocking?
2. Any suggestions on how to overclock HW version C1?
3. Anybody have any success in overclocking AC68U C1 version or the AC1900 C1 version? Please share your experiences.

 

[Zimtu]

Here is an update after further research - the above underclocking behavior to 100Mhz, happens only when reboot command is used at command line. When power is switched off and switched on, with init scripts set to clkfreq=1200,666 the CPU remains locked in at 1Ghz but the NVRAM values show 1200. Next, if power is switched off and switched on, with init scripts set to clkfreq=1400,666 the CPU goes to 100Mhz but the NVRAM values show 1400! The CPU true speed was measured using openssl.

This is quite bizarre. Anybody have any comments? Are there any other tests anybody would like to suggest on this C1 version of AC68?

 

[Zimtu]

No comments to share on above? Has anybody seen a CPU underclock to 100Mhz in any ASUS router, when NVRAM value of clkfreq is changed?

 

[chris.at]

Found this thread because I was running into the same issue with the new revision E1. Any new findings on this?