cancel
Showing results for 
Search instead for 
Did you mean: 

How fast ? ;-)

root
Associate II
Posted on November 16, 2011 at 08:29

Hello,

Just wondering ... did somebody already tried to overclock an STM32 (F2 series for example) ?

What max clock can we achieve before the chip blows up (with correct flash wait state) ?

Thomas.
14 REPLIES 14
domen23
Associate II
Posted on November 21, 2011 at 11:15

We played with stm32f1 and external clock. I think 132 MHz was the limit with 12 MHz crystal. When it failed, it just didn't work, and when it worked it worked for days without noticeable problems.

emalund
Associate III
Posted on November 21, 2011 at 14:59

DO NOT OVERCLOCK

if you overclock

it may fail if the temperature gors up a couple of degerees

if yout power supply drop a few millvilts

if you replace the chip

if ...

if ...

Erik

tchoutchawn
Associate
Posted on October 26, 2013 at 02:48

Fastest I could go on a stm32F407VG was 272MHz, but the main PLL has to run at 544MHz to acheive this frequency because of the mandatory /2 divider on the PLLP. 

jj2
Associate II
Posted on October 26, 2013 at 22:49

Agree w/Erik's guidance and add:

a) future revisions, same MCU, and/or even different ''lots'' may not ''maintain'' that, ''beyond spec'' performance or operation.

b) simple MCU aging may also fail to ''maintain'' earlier performance ''gains.''

c) operation of most ICs - ''outside'' specified conditions - usually voids the warranty.

d) such ''outside specified condition'' operation vastly expands your (and your firm's) exposure to any/all legal liabilities - even those not directly/exactly related to such excess.

Can it be done - to some degree - sometimes yes.  But - is it wise...?

Risk-Reward dictates that you fully & properly evaluate any/all consequences which may result from such choice... 
frankmeyer9
Associate II
Posted on October 27, 2013 at 10:56

Albeit I agree to both of you in the commercial point of view, your stated concerns are most probably of minute interest to the crash-kid like ''get it as fast as it can'' experimenter.

I believe they doing it for fun, on their own risk, and with no serious commercial idea in mind ...

...and why not ?

Tuttle.Darrell
Associate II
Posted on October 28, 2013 at 13:06

All the usual warnings apply etc  🙂 ... YMMV

When developing a board using an STM32F407, the eval board I had used an STM32F207. I had a compiler switch so I could test some code on the eval board, or compile for the full 168Mhz for the STM32F407. At one point, I accidentally compiled for the STM32F407 and loaded an ran the test code on the STM32F207 eval board. It ran! I ran it all day. On further investigation, I compared the PWR control register in the reference manuals for both processors. The STM32F4 has a ''VOS'' (regulator voltage scaling) bit required to be set (bit 14) for running full speed. This bit is listed as ''Reserved'' for the STM32F2, but (for my eval board processor at least) apparently has the ''VOS'' functionality.

Posted on October 28, 2013 at 14:53

apparently has the ''VOS'' functionality.

That, or it's inconsequential?

F2 has some issues with a critical path in the ART/Prefetch implementation, it's caught a few GNU/GCC users.
Tips, buy me a coffee, or three.. PayPal Venmo Up vote any posts that you find helpful, it shows what's working..
Tuttle.Darrell
Associate II
Posted on October 28, 2013 at 15:35

>

That, or it's inconsequential?

After stumbling onto this, I tried it both ways and the undocumented ''VOS'' needed to be set for 168Mhz to work. Note there were other documented registers to change for 168Mhz (PLL, external SRAM timing, and I don't remember what else), but all of these were also set to the required values since I accidentally compiled for STM32F4 at 168Mhz).

emalund
Associate III
Posted on October 28, 2013 at 15:37

Albeit I agree to both of you in the commercial point of view, your stated concerns are most probably of minute interest to the crash-kid like ''get it as fast as it can'' experimenter.

 

I believe they doing it for fun, on their own risk, and with no serious commercial idea in mind ...

 

 

...and why not ?

 

because then, in a few weeks we get to process a mile long thread ''xxx worked fine, now yyy does not'' and after two hundered suggestions the OP finally states ''xxx worked firn with overclocking''

Erik