ADSP overclocking

By mistake the 25MHz rated ADSP-2185L was put to run at 50MHz clock. It
was working absolutely normal. No excessive heat. No glitches. The
mismatch in the speed grade was noticed only by a chance.
I was really surprised that the ADSP can take 2x overclocking so easy.
Does that mean AD is using the same silicon for all speed grades?
    

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Vladimir Vassilevsky wrote:
> 
> By mistake the 25MHz rated ADSP-2185L was put to run at 50MHz clock. It
> was working absolutely normal. No excessive heat. No glitches. The
> mismatch in the speed grade was noticed only by a chance.
> I was really surprised that the ADSP can take 2x overclocking so easy.
> Does that mean AD is using the same silicon for all speed grades?
> 
> 
> Vladimir Vassilevsky
> 
> DSP and Mixed Signal Design Consultant
> 
> http://www.abvolt.com

If DSP is tested in the same manner as regular micros, the parts are
sorted by rated speed. It is possible that good yields at the higher 
speeds mean some parts that actually rate higher are binned to the 
lower speed to meet marketing requirements. 

I'm guessing, but this sort of thing might work reliably over low
temperature and on specific parts, but for production purposes, 
doing this en general  would subsume risk. You may also encounter 
odd, hard-to-reproduce problems -  the usual overclocking stuff. 

Please note - I'm guessing here, applying things that apply 
on a widely known basis to non-DSP. 

--
Les Cargill

This is not from experience, but intuitively, there should be some
instruction (or possibly sequence of instructions) that is the worst case
for timing.  If you never use those instructions, you can probably get
better performance than the chip is rated at.  Of course, more that likely
you have no way of knowing what those worst case instructions are...

"Les Cargill" <lcargill@worldnet.att.net> wrote in message
news:3F6247A1.2D84AA2B@worldnet.att.net...
> Vladimir Vassilevsky wrote:
> >
> > By mistake the 25MHz rated ADSP-2185L was put to run at 50MHz clock. It
> > was working absolutely normal. No excessive heat. No glitches. The
> > mismatch in the speed grade was noticed only by a chance.
> > I was really surprised that the ADSP can take 2x overclocking so easy.
> > Does that mean AD is using the same silicon for all speed grades?
> >
> >
> > Vladimir Vassilevsky
> >
> > DSP and Mixed Signal Design Consultant
> >
> > http://www.abvolt.com
>
> If DSP is tested in the same manner as regular micros, the parts are
> sorted by rated speed. It is possible that good yields at the higher
> speeds mean some parts that actually rate higher are binned to the
> lower speed to meet marketing requirements.
>
> I'm guessing, but this sort of thing might work reliably over low
> temperature and on specific parts, but for production purposes,
> doing this en general  would subsume risk. You may also encounter
> odd, hard-to-reproduce problems -  the usual overclocking stuff.
>
> Please note - I'm guessing here, applying things that apply
> on a widely known basis to non-DSP.
>
> --
> Les Cargill

Jon Harris wrote:
> 
> This is not from experience, but intuitively, there should be some
> instruction (or possibly sequence of instructions) that is the worst case
> for timing.

In the case of ADSP-21xx, a pack of fetch-multiply-accumulate commands
creates the very maximum load for the power supply. 

> If you never use those instructions, you can probably get
> better performance than the chip is rated at.

There are the external bus related issues too. 


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com