[Beaglebone] overclocking success

[Uncle Joe]

Fellow high performance computing enthusiasts,

I just overclocked my dog from 720MHz to 1152 MHz (only MPU PLL) !
Take that you lazy dog! Get those ARMs moving! Crunch those numbers
like you mean it.

Strangely, increasing the TPS65217's DCDC3 voltage from 1.1V to 1.3V
didn't add more overclocking room. I just realized OMAP3 is an
overclocker's dream. So many PLLs and clock dividers to play with.
I've also doubled the clock to the SGX graphics unit to 400MHz without
problem. Too bad I don't have the HDMI cape so that I can make a video
showing how much faster the graphics go. Next, I will attempt to
overclock the DDR2 memory from 266 MhZ to its 400MHz spec.

Has anyone pushed their board beyond 1.2 GHz? Anyways, I'd better
hide. Gerald Coley is going to come after me with a shotgun for
abusing his pet like he did to that cheating PCB vendor. Yikes!


-----------------------before - 720 MHz------------------------------

LINPACK benchmark, Single precision.
Machine precision: 6 digits.
Array size 200 X 200.
Average rolled and unrolled performance:

Reps Time(s) DGEFA DGESL OVERHEAD KFLOPS
----------------------------------------------------
16 0.79 87.34% 5.06% 7.59% 30100.461
32 1.60 88.13% 3.75% 8.12% 29895.688
64 3.18 88.68% 2.52% 8.81% 30308.049
128 6.37 88.70% 2.51% 8.79% 30255.896
256 12.73 88.53% 2.59% 8.88% 30308.035
512 25.47 89.05% 2.20% 8.76% 30255.877
1024 50.93 88.65% 2.61% 8.74% 30255.879

--------------------------after - 1152
MHz--------------------------------

LINPACK benchmark, Single precision.
Machine precision: 6 digits.
Array size 200 X 200.
Average rolled and unrolled performance:

Reps Time(s) DGEFA DGESL OVERHEAD KFLOPS
----------------------------------------------------
32 0.99 94.95% 1.01% 4.04% 46259.645
64 1.96 90.31% 2.04% 7.65% 48559.863
128 3.94 90.10% 3.30% 6.60% 47768.129
256 7.86 89.69% 2.93% 7.38% 48293.012
512 15.73 89.96% 2.48% 7.57% 48359.477
1024 31.45 89.79% 2.80% 7.41% 48293.090
2048 62.89 89.89% 2.61% 7.51% 48351.172


(synthetic test I made, measuring only raw arithmetic throughput)

--------------------------before - 720
MHz-------------------------------
1.15017G int adds/s
0.359117G int multiplies/s
0.102649G float adds/s
2.87421G neon int32 adds/s
0.718913G neon int32 multiplies/s
1.43731G neon float adds/s
1.43813G neon float multiplies/s
1.43747G neon float madds/s

--------------------------after - 1152
MhZ-----------------------------------
1.84078G int adds/s
0.574921G int multiplies/s
0.164397G float adds/s
4.60351G neon int32 adds/s
1.15064G neon int32 multiplies/s
2.30213G neon float adds/s
2.2995G neon float multiplies/s
2.30176G neon float madds/s

[Gerald Coley]

Here is my take,

 

You are certainly free to stretch the board to its extremes. It is not reasonable to expect all boards to be able to do this, so if others want to try, that is fine, but do not request an RMA saying it does not work at 1.1GHZ!

 

As for this board, please watch it closely. If it hollers "Uncle" please scale it back! :^)

 

DVI-D Capes are shipping, so you should be able to get one.

 

Gerald

 

--
To join: http://beagleboard.org/discuss
To unsubscribe from this group, send email to:
beagleboard...@googlegroups.com
Frequently asked questions: http://beagleboard.org/faq

--

Gerald

 

ger...@beagleboard.org

http://beagleboard.org/

http://circuitco.com/support/

[Koen Kooi]

Op 1 mrt. 2012, om 12:55 heeft Uncle Joe het volgende geschreven:

> Fellow high performance computing enthusiasts,
>
> I just overclocked my dog from 720MHz to 1152 MHz (only MPU PLL) !
> Take that you lazy dog! Get those ARMs moving! Crunch those numbers
> like you mean it.
>
> Strangely, increasing the TPS65217's DCDC3 voltage from 1.1V to 1.3V
> didn't add more overclocking room. I just realized OMAP3 is an
> overclocker's dream. So many PLLs and clock dividers to play with.
> I've also doubled the clock to the SGX graphics unit to 400MHz without
> problem.

Any instructions on how to overclock the SGX? I have some applications that can use a nice boost and I have enough beaglebones to sacrifice one to the overclocking gods.

regards,

Koen

[Brian Hutchinson]

I can see where this is heading .... visions of YouTube videos of dead
BeagleBones being buried in their tin boxes or set adrift with lighter
fluid.

[Uncle Joe]

Well, I've purposely not put any directions here for fear of people
damaging their boards, unless you think it's good to share
overclocking instructions. Koen, I overclocked the SGX by just
decreasing the CM_DIV_M4_DPLL_CORE divider. There's no need to put the
PLLs in bypass mode and wait for them to lock. The USB and Ethernet
functions also share the same clock, and they seemed OK after the
overclock. I'll email you the code that I used.

My overclocking attempt was more out of need than curiosity/greed. My
program for controlling a Scorbot 3 needs all the cycles it can get to
do feature extraction from the ASUS Xtion Pro RGB/depth camera. I'm
playing around with the PRUSS to handle motor encoder interrupts -
despite having to use the high level assembly code. I though about
buying a board with more cores like PandaBoard or TrimSlice, but I
didn't think they had enough raw power to justify the complexity of
integrating them with the BeagleBone. Quad core Tegra 3 makes my mouth
water, but I know it will probably be too locked down to build a
community around. I can't wait until ARM A15 based Beagles come out.

One other possibility I'd love to try is to stack multiple BeagleBones
while maintaining a shared, but non-uniform memory, so I can use
OpenMP. Does anyone know if this is possible?

thanks,
Yale

On Mar 1, 10:14 am, Brian Hutchinson <b.hutch...@gmail.com> wrote:

[Charlie Fullerton]

+1 LOL

[kw90...@gmail.com]

Hello there Joe!

I've been looking everywhere trying to find where exactly I can modify the MPU PLL. 

If you could please point us in the right direction that would be great. Thank you.