Overclocking RT-AC68U

[horus1963]

I am trying to over clock my RT-AC68U to 1200,800 but I am having a problem. I am using the following commands, as I found on this forum:

nvram set clkfreq=1200,800
nvram commit && reboot

The problem is that the settings will not survive the reboot. After reboot, nvram get clkfreq will give 800,533. After a lot of experimenting, I found that if I change the CPU frequency to 1200 Mhz (1000 works also) but not change the RAM frequency, then the setting will stick. No RAM frequency value other than 533 works (I have tried 666 and 800)

My router has CFE version 1.0.11 and firmware version 3.0.0.4.374.39 (Merlin)

I do not know what I am doing wrong or if I am missing something.

Thanks in advance for your help.

 

[wifi_reef]

with the BL 1.0.1.1

To get RT-AC68U overclocked follow steps:

From the default FW:

1. check the nvram values:

nvram get clkfreq

2. in web interface go to the:
Administration --> Performance Tuning
- choose LED button Behavior as OverClock: Press the button to turn on overclock, release the button to turn off.
- press Apply

3. After restart of service PRESS BUTTON on the back of the unit and wait for restart.

4. check the nvram values (it is still the same 800,533):

nvram get clkfreq

5. NOW enter desired values (for example):

nvram set clkfreq=1000,800
nvram commit && reboot

6. After restart verify again entering:

nvram get clkfreq


In the web interface you will find under Tools / sysinfo proper value for the CPU Frequency.

And yes it works ... I'm getting around 65 MB/s transfer with USB 3.0 when wired to the PC.

 

[horus1963]

Thank you very much! I was not aware that I needed steps 2 & 3. Now everything is happily clocked at 1200,800.

Thanks again.

 

[wifi_reef]

In theory it should work without "pushing the button" however it does not. End there is one more thing ... be careful with settings and monitor temperature ! RMerlin does not recommend this workaround because you may loose the way back if something goes wrong with OC !

It is good to understand what OC will give you. It affects greater WAN to LAN through-output (suitable only for just a few people around the world) and better Samba performance (this is interesting for more people). OC also consumes more energy as one can easy predict.

 

[horus1963]

Will Do. I will keep a watchful eye on the temps. I also ordered a fan for it which should be here any day now. Should help lower the temperature and keep it in check.

Thanks a lot.

 

[Sky1111]

In theory it should work without "pushing the button" however it does not. End there is one more thing ... be careful with settings and monitor temperature ! RMerlin does not recommend this workaround because you may loose the way back if something goes wrong with OC !

It is good to understand what OC will give you. It affects greater WAN to LAN through-output (suitable only for just a few people around the world) and better Samba performance (this is interesting for more people). OC also consumes more energy as one can easy predict.

Not only that... OpenVPN client performance benefits quite noticeably from overclocking; LAN transfers seem to be faster too...

 

[horus1963]

Just wanted to give an update.

When I followed the above steps initially, I noticed that the nvram value of 1200,800 did stick and that the system info reported that the processor frequency was 1200 and I thought all was well. This was until I checked Bogomips.

The value I got was 1998.84. Reading more in this forum I noted that this is the value for the 1000 processor frequency. So basically, pressing the button made the router ignore the nvram value of clkfreq on reboot and set the router to 1000,533. Since it ignored the nvram value it did not reset it back to 800,533. An nvram get clkfreq will still show 1200,800 in this case.

If you release the button, then the router will read the nvram value on reboot, reject it and reset it back to 800,533.

So I was back to square one . I can change the clkfreq for any valid processor value (I tried both 1000 and 1200) but I can't change the memory frequency (I tried 666 and 800).

Setting the nvram value to 1200,533 will stick (without the over clock button pressed) and will yield a Bogomips of 2398.61 which is consistent with 1200 processor speed.

I have reread all the posts related to overclocking but have yet to figure out why I can't change the memory frequency. May be I am missing something or maybe the memory frequency of my AC68U just can't be changed .

 

[TonyH]

Hmmm,
Just look at cpu load B4 and after OC'ing. In some cases minimal benefit. If one is running
features requiring cpu power yes, otherwise no need to bother. High temp. is no good for
cpu or memory(unstable shortens life) Regardless most will replace their routers B4 they die.

 

[wifi_reef]

Just wanted to give an update.

The value I got was 1998.84. Reading more in this forum I noted that this is the value for the 1000 processor frequency. So basically, pressing the button made the router ignore the nvram value of clkfreq on reboot and set the router to 1000,533. Since it ignored the nvram value it did not reset it back to 800,533. An nvram get clkfreq will still show 1200,800 in this case.

If you release the button, then the router will read the nvram value on reboot, reject it and reset it back to 800,533.

So I was back to square one . I can change the clkfreq for any valid processor value (I tried both 1000 and 1200) but I can't change the memory frequency (I tried 666 and 800).

1. RAM clk does not affect on bogoMIPS numbers - this refers only to CPU status (pls read about in wiki for more details) - generally it shows how many empty instruction can execute CPU in one second doing nothing.

2. Regarding above instruction - it seems to be the only way to change RAM and CPU clk at the time in rt-ac68u. Also in this way you can go with CPU to 1000 MHz for now - because of Merlin's restriction.

One can always ask RMerlin to allow 1000 and 1200 choice for CPU ... in future, 40 edition ;-)

 

[horus1963]

I understand that bogoMIPS is only relevant to the processor speed. I just used it to show that even though the clkfreq was set to 1200,800 when the over clock button is pressed the processor speed was set to 1000 regardless of the nvram setting.

The other point is that I can't change the memory frequency and I do not know of a way to check its value other than reading the nvram value.

Like I mentioned, may be it can't be changed.

 

[RMerlin]

One can always ask RMerlin to allow 1000 and 1200 choice for CPU ... in future, 40 edition ;-)

I have no intention of supporting overclocking at the webui level (beside Asus's own Turbo button on the AC68U, of course). That'd be a support nightmare for me to deal with.

 

[gschlact]

current USB 3.0 SDD HDD performance on Asus ac68u

I just wanted to provide an update for USB 3.0 transfer speed performance on the Asus AC68u. I know I had trouble finding performance numbers / expectations online and wanted to provide this info.

Test Conditions:
Stock Firmware 3.0.0.4.376_1663 + Tweaks
nvram set clkfreq 1200,800
Reducing USB 3.0 interference=Disabled
wl_txpower=500
Wired Lan 1gbps wired to AC68u hardware vA2.
passive PNY 240GB SSD USB 3.0 wired 1gbps lan to AC68
passive WD 1TB HDD usb 3.0 wired 1gbps lan to ac68
Windows 7 prof. SP1
file transfer ~1.3GB single movie file

RESULTS:
reported: cat /proc/dmu/temperature --> 82 C (at end of xfers)
command line: TOP shows still about 20% idle on CPU.

passive SSD:
Read = ~65MB/sec (win reports ~590mbps in Task Mgr)
Write= ~55MB/sec (win reports ~550mbps)

passive WD Passport HDD:
Read = ~63MB/sec (win reports ~570mbps in Task Mgr)
Write= ~51MB/sec (win reports ~460mbps)

COMMENTS:
---cable quality made a difference and I observed 20% drop off with thinner cables on both drive types (HDD, SDD)
---for all transfers for the first 1-2 seconds I observed a burst (reported by Win) of about 50% higher transfer speed which then settled down to the numbers shown above. I assume this is some caching.
---UNFORTUNATELY, this same SDD gives me 150MB/100MB/sec when attached to my lenovo USB 3.
---Performance varied a bit if I didn't stretch the 20" USB3 cable moving SDD away from ac68
--- I saw somewhere that performance may be related to the USB 3 power supplied by the port for the drive. ANy comments? Help?
--- can someone run this test with a powered USB 3.0 Drive? Longer cable?

HOW are we going to get closer to the Laptops direct USB connection of 150/100MB/sec???????????

-Guy
gschlact

 

[RMerlin]

HOW are we going to get closer to the Laptops direct USB connection of 150/100MB/sec???????????

-Guy
gschlact

You won't. First, you are comparing a 1 GHz low-power ARM CPU with a dual/quad multi-gigahertz CPU. Also, your laptop motherboard's internal bus can handle multiple gigabits of data transfer between the sub-components, while the router's SoC probably has like one or two single PCI-E lanes to transfer everything (not just USB data). And finally, you won't be transferring 1500+ Mbits over a 1000 Mbits LAN, aside from the overhead added by SMB itself.

 

[gschlact]

RMerlin,
I appreciate your response. However, I contend there is even more performance to yield from the ac68 router.

While 60MB/sec is still pretty impressive and >2x of USB2, while transferring and getting this performance, there was still CPU and Ram headroom resources. On the Lenovo, the limitation was the HDD at about 100MB/sec and the Flash on the SDD at 150MB/sec. DDR800 even at 1byte writes should can provide the (1Byte/8bits) * 800M writes/sec = 100MB/sec. I would think however that the flash is at least 2-4x this in width- So the DDR isn't the limitation.

REQUEST:
It would be nice for someone to test a setup with a) longer USB 3.0 cable for extra distance interference protection and b) powered USB 3.0 hub for the HDD so that the ac68 can utilize full speed spec without power (amperage) drain from the passive HDD. At least with this test, we can gain more confidence in the actual bottleneck- CPU as you contend, or USB 3.0 controller power limitation of passive HDD load.

-Guy

 

[RMerlin]

RMerlin,
I appreciate your response. However, I contend there is even more performance to yield from the ac68 router.

I'm not saying it cannot be improved, simply that it will never get anywhere close to the numbers you are seeing with a computer, and when accessing the device locally rather than over a gigabit network.

Most likely, the next performance boost will come when Asus can get the new SDK7 code from Broadcom to work correctly. Broadcom improved the USB stack quite significantly in this new SDK. Initial tests on an AC87 showed a performance gain of up to 20% in some specific scenarios (will probably be less in the AC68/AC56 cases).

 

[dvohwinkel]

Just wanted to add some info.. I did a nvram get clkfreq on a new RT-AC68U hardware rev B1 and it returned 1000,800. The interesting thing to me is the now default 800Mhz memory clock. I am hoping to try 1200,800 after some extensive testing, at stock, on Merlin's awesome firmware.

 

[gschlact]

current USB 3.0 SDD HDD performance on Asus ac68u

I just wanted to provide an update for USB 3.0 transfer speed performance on the Asus AC68u. I know I had trouble finding performance numbers / expectations online and wanted to provide this info.

Test Conditions:
Stock Firmware 3.0.0.4.376_1663 + Tweaks
nvram set clkfreq 1200,800
Reducing USB 3.0 interference=Disabled
wl_txpower=500
Wired Lan 1gbps wired to AC68u hardware vA2.
passive PNY 240GB SSD USB 3.0 wired 1gbps lan to AC68
passive WD 1TB HDD usb 3.0 wired 1gbps lan to ac68
Windows 7 prof. SP1
file transfer ~1.3GB single movie file

RESULTS:
reported: cat /proc/dmu/temperature --> 82 C (at end of xfers)
command line: TOP shows still about 20% idle on CPU.

passive SSD:
Read = ~65MB/sec (win reports ~590mbps in Task Mgr)
Write= ~55MB/sec (win reports ~550mbps)

passive WD Passport HDD:
Read = ~63MB/sec (win reports ~570mbps in Task Mgr)
Write= ~51MB/sec (win reports ~460mbps)

COMMENTS:
---cable quality made a difference and I observed 20% drop off with thinner cables on both drive types (HDD, SDD)
---for all transfers for the first 1-2 seconds I observed a burst (reported by Win) of about 50% higher transfer speed which then settled down to the numbers shown above. I assume this is some caching.
---UNFORTUNATELY, this same SDD gives me 150MB/100MB/sec when attached to my lenovo USB 3.
---Performance varied a bit if I didn't stretch the 20" USB3 cable moving SDD away from ac68
--- I saw somewhere that performance may be related to the USB 3 power supplied by the port for the drive. ANy comments? Help?
--- can someone run this test with a powered USB 3.0 Drive? Longer cable?

HOW are we going to get closer to the Laptops direct USB connection of 150/100MB/sec???????????

-Guy
gschlact

Skip to 2016-July, and Firmware 380.59
I am having trouble reobtaining this performance.
I now on USB3 only getting 7MB-14Mb reading from the attached USB3 (over 300Mbps link)

I am having trouble keeping the higher values after reboot for:
--> nvram set clkfreq=1200,800
-->nvram commit && reboot

Also, having trouble with keeping Tx power higher and country code settings..

Any help would be appreciated.

 

[mjp93]

Skip to 2016-July, and Firmware 380.59
I am having trouble reobtaining this performance.
I now on USB3 only getting 7MB-14Mb reading from the attached USB3 (over 300Mbps link)

I am having trouble keeping the higher values after reboot for:
--> nvram set clkfreq=1200,800
-->nvram commit && reboot

Also, having trouble with keeping Tx power higher and country code settings..

Any help would be appreciated.

I m getting these speeds om stock frequency on 380.59
I tried to overclock but after reboot the cpu is still at 1Ghz.

SAMBA:
Wired: upto 15MB/s(R) and 8MB/s(W)
Wireless: upto 10MB/s(R) and 7MB(W)

FTP:
Wired: upto 37-39MB/s(R/W)
Wireless: upto 17MB/s(R/W)

Did u checked on FTP?

 

[sfx2000]

You won't. First, you are comparing a 1 GHz low-power ARM CPU with a dual/quad multi-gigahertz CPU. Also, your laptop motherboard's internal bus can handle multiple gigabits of data transfer between the sub-components, while the router's SoC probably has like one or two single PCI-E lanes to transfer everything (not just USB data). And finally, you won't be transferring 1500+ Mbits over a 1000 Mbits LAN, aside from the overhead added by SMB itself.

The Router is not a NAS - so one should set appropriate expectations - it's a checkbox and convenience feature to share some data across the LAN.

As @RMerlin mentions above - there's only so much that a dual-core ARM can do - and it's priority is routing packets to/from the LAN to the WAN.

 

[mjp93]

The Router is not a NAS - so one should set appropriate expectations - it's a checkbox and convenience feature to share some data across the LAN.

As @RMerlin mentions above - there's only so much that a dual-core ARM can do - and it's priority is routing packets to/from the LAN to the WAN.

Yes bro, I m not targeting to achieve NAS speeds but just what a normal users will get. Hence I was expecting that I would atleast get around 50s megabit.