Hi all, I've a project in which I'm very short of DSP time. The idea is to preselect DSPs and to overclock them. EMIF stays at 100 MHz. I want to hear your opinion on this. How much overclocking can be possible? How many DSPs would be good/bad? What about aging? How to do the preselection? Is there a test program out there? What parts of the DSP are to be tested and how? Are there some other caveats I have not yet thought of? TIA Gustl |
6713 overclocking
Started by ●January 19, 2005
Reply by ●January 22, 20052005-01-22
Gustl,
I would not recommend over clocking in a product that goes to
customers.
Are you using a 300 Mhz part?? If the part could go something like
320 or 330 Mhz reliably, TI would probably advertise it at that
speed.
The easiest way to improve system performance would be to use a static or
psuedo-static external RAM.
If your key performance involves external memory, you may want to
empirically test for harmonic problems between the CPU and external memory
accesses. You may actually improve system performance by slowing down the
EMIF clock. Conversely, over clocking by a few Mhz could actually slow
down performance.
If you plan on 'over clocking', I would suggest...
Use a PCB design that properly supports the inner thermal pad. Use a
heatsink on the part. Keep the cabinet cool - heat is your enemy.
Use a dedicated clean core supply that is variable in .05 or .1
increments. Start with the supply at 1.45v.
Proper characterization of an over clocked product will take lots of time
and testing. I know of no tools/programs to ease this. I would
suggest instrumenting your app to verify as much information as possible [you
can build a 'test' or 'production' version of the
code]. Run up the clock to the point of failure - I would expect the
failure to occur on some sort of internal I/O or memory access [this is a
guess]. Try this a few times to see if there is a type of failure that you
can look for.
Run your test version of software at an elevated clock rate and temperature
for an extended period of time. Followup by running with only high clock
and normal temperature and then normal clock and high temperature.
You may want to equip the product with a 'diagnostic mode' that
runs at an elevated clock freq.
Once again, I would not 'put my name on a product' that runs
outside of the manufacture's specs. depending on what the product
does, there could also be some form of legal liability from 'premature
failures'.
good luck,
mikedunn
Bernhard 'Gustl' Bauer <g...@quantec.de> wrote:
|
Reply by ●March 1, 20052005-03-01
Overclocking DSP Clock is not a very good idea, and is not adviced by the manufacturer. The Core do work for significant amount of time at room temperature however at Elevated Temp we had seen spurious behaviour. -Anurag Darbari > >Hi all, > >I've a project in which I'm very short of DSP time. The idea is to >preselect DSPs and to overclock them. EMIF stays at 100 MHz. I want to >hear your opinion on this. > >How much overclocking can be possible? > >How many DSPs would be good/bad? > >What about aging? > >How to do the preselection? Is there a test program out there? What >parts of the DSP are to be tested and how? > >Are there some other caveats I have not yet thought of? > >TIA > >Gustl > > |
Reply by ●March 2, 20052005-03-02
Hi anurag123, > Overclocking DSP Clock is not a very good idea, and is not > adviced by the manufacturer. The Core do work for significant amount > of time at room temperature however at Elevated Temp we had seen > spurious behaviour. While I do agree on your reply, just as a notice I've been using a 500Mhz DM642 oc to 600Mhz without any problem, working full time on MPEG4 compression/decompression. This has been tested on a batch of 20 boards. I've also tested a 720Mhz DM642 oc to 792Mhz without any significant problem. Notice that using an hairdrier and a termomether I've tested them at 109 Celsius degrees. :) This of course does not mean oc is a safe procedure. -- Best regards, Reply to: Marco Braga Insignis Technologies |
Reply by ●March 2, 20052005-03-02
Marco- > > Overclocking DSP Clock is not a very good idea, and is not > > adviced by the manufacturer. The Core do work for significant amount > > of time at room temperature however at Elevated Temp we had seen > > spurious behaviour. > > While I do agree on your reply, just as a notice I've been using a > 500Mhz DM642 oc to 600Mhz without any problem, working full time > on MPEG4 compression/decompression. This has been tested on a batch of > 20 boards. I've also tested a 720Mhz DM642 oc to 792Mhz without any significant > problem. Notice that using an hairdrier and a termomether I've tested > them at 109 Celsius degrees. :) > This of course does not mean oc is a safe procedure. Hairdrier is not valid -- that blows air also, i.e. forced air "cooling". Use a lamp or oven instead! (Be safe). -Jeff PS. NO I DON'T THINK IT'S A GOOD IDEA TO OVERCLOCK DSPs. My comments are relevant to extreme "where is the limit" type of testing only. |
Reply by ●March 3, 20052005-03-03
Marco- > > Hairdrier is not valid -- that blows air also, i.e. forced air "cooling". Use a lamp > > or oven instead! (Be safe). > > Yes but I measured the temperature with a sensor put directly on the > chips, so the air can cool, but the temperature is what I read. For heat-testing DSP boards, I would suggest a) to use a method that simulates what would really happen, b) raise the temperature of all components, including SDRAM, logic (if any), etc. If there is forced-air in your box, then let it stay, if not then don't add it. Let's put it this way: if my box failed under extended temp conditions, and my customer asked me "how did you test it?", I would not want to say "I used a hair drier". On the other hand, if your testing is only for debug or academic purposes, then my comments don't apply. -Jeff |
Reply by ●March 3, 20052005-03-03
Marco- > > Hairdrier is not valid -- that blows air also, i.e. forced air "cooling". Use a lamp > > or oven instead! (Be safe). > > Yes but I measured the temperature with a sensor put directly on the > chips, so the air can cool, but the temperature is what I read. For heat-testing DSP boards, I would suggest a) to use a method that simulates what would really happen, b) raise the temperature of all components, including SDRAM, logic (if any), etc. If there is forced-air in your box, then let it stay, if not then don't add it. Let's put it this way: if my box failed under extended temp conditions, and my customer asked me "how did you test it?", I would not want to say "I used a hair drier". On the other hand, if your testing is only for debug or academic purposes, then my comments don't apply. -Jeff |
Reply by ●March 3, 20052005-03-03
Hi Jeff, > Let's put it this way: if my box failed under extended temp conditions, and my > customer asked me "how did you test it?", I would not want to say "I used a hair > drier". Eheheheh!! Of course. I was just talking about my personal tests, nothing serius or scientific. Simply I was amazed seeing my complex app still run at 792Mhz with really hot air blowing on the board. Nothing more. :) -- Best regards, Reply to: Marco Braga Insignis Technologies |
Reply by ●March 3, 20052005-03-03
EEks, thats a yuckkky way to test.... Temperature Chamber is the way to test. in event of failure, use Heat Gun(s) (described in IEEE Specifications). <note: guns is plural> It is important that components whose characterstic changes (like Analog PLL, Oscialators, VCXO) with the change in Temperature must be tested and in turn test the compnents which depends on the characterstic/performance of those components. Also remeber that the Board Characterstic Changes (i.e. Z) with change in Frequency (which depends on Temperature) Temprature Tests provides the following functionality of the processor: 1. OverClocking / Underclocking 2. Static and Dynamic Timing Analysis 3. Signal Integrity, Reflections, SSO/Cross Talk (we are talking of the oscillator characterstics to change with increse in Tempertaure) Analysis of a Waveform attached. Anurag Darbari ----- Original Message ----- From: "Marco Braga" <> To: "Jeff Brower" <> Cc: <> Sent: Thursday, March 03, 2005 8:55 AM Subject: Re: [c6x] Re: 6713 overclocking > Hi Jeff, > >> Let's put it this way: if my box failed under extended temp conditions, >> and my >> customer asked me "how did you test it?", I would not want to say "I used >> a hair >> drier". > > Eheheheh!! Of course. I was just talking about my personal tests, > nothing serius or scientific. Simply I was amazed seeing my complex > app still run at 792Mhz with really hot air blowing on the board. > Nothing more. :) > > -- > Best regards, Reply to: > Marco Braga > Insignis Technologies |
|