Shop OBEX P1 Docs P2 Docs Learn Events
114Mhz Propeller 1 @ $0.55 — Parallax Forums

114Mhz Propeller 1 @ $0.55

Alexis64Alexis64 Posts: 42
edited 2011-05-20 15:43 in Propeller 1
Hi,

We've been testing the propeller's speed while running our telemetry units as well as a variety of graphic video drivers and we've found that using a 14.31818Mhz Crystal ($0.55 at digikey) will reliably clock the Propeller at 114,54544 Mhz with a pllx8 ... The test was conducted at different temperatures for the past 48 hours.

Now that you have this information, go and get a 14.31818Mhz Xtal (or rescue it from surplus PCBs) and start developing a better, cooler, video driver :) ...

P.S. both the loaded (18pf caps) and the unloaded crystal worked fine - although I personally prefer the 18pf loaded crystal

The beginning of your demos (like graphics.spin etc) will need to be altered as follows:
  _clkmode = xtal1 + pll8x
  _xinfreq = 14_318_180

Video: http://www.youtube.com/watch?v=rd12opYAa3Y
«1

Comments

  • Cluso99Cluso99 Posts: 18,066
    edited 2011-05-16 15:42
    That is the top frequency that I found with the RamBlade. RamBlade uses a 6.5MHz xtal (104MHz) and I also supply a 13.5MHz xtal (108MHz) (xtal is plugable).

    However, overclocking is reliant on the proper decoupling capacitors right at the prop power pins. What prop are you using? (DIP or QFP) I found the DIP prop seems to perform slightly faster. Sapieha has run my TriBlade at 120MHz using 15MHz xtal for many months. There is another thread about proper decoupling and speeds.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2011-05-16 16:16
    I'd be more inclined to use a crystal at half that frequency and use PLLx16. The reason is that 14+ MHz is way, WAY out on the outer fringes of what the PLL will accept as the driving frequency, and some Prop chips might not work at all being driven like that, or will run at the wrong frequency. BTW, the spec says to keep the PLL driving frequency between 4 and 8 MHz. So you'd get the same result from a 7+ MHz crystal and PLLx16 and be within the driving frequency specs.

    -Phil
  • Cluso99Cluso99 Posts: 18,066
    edited 2011-05-16 17:43
    Unfortunately there are no readily available xtals at 1/2 14.31818MHz :(

    IIRC Chip said the osc was actually specd to run at 14.31818MHz tops. I do know my RamBlade fails at 7.3xxMHz. I had looked for a 7MHz xtal but no luck so far (cheap and readily avail). I was interested to see where the prop actually fails in both 8x and 16x.

    Of course, we really do not know whether the whole chip works at these higher frequencies, only that our programs work.

    I now have 2 standard xtals in use... 6MHz (96MHz) for any USB work and 6.5MHz (104MHz) otherwise, with 13.5MHz (108MHz) supplied as an option.
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-16 19:11
    Has anyone tried a 15MHz crystal with PLL8X for 120MHz? If so under what conditions tested, for how long?
    Something like:
    http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=535-9874-1-ND
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-16 19:26
    Also looking at the graph in the Propeller data sheet:
    I am pondering an attempt to run the Propeller 1 at 3.8V, as the max stable frequency seems to increase with operating voltage, and Chip seems to imply that they have successfully run the Prop at up to 5V (look at post #4 in http://forums.parallax.com/showthread.php?85474-How-to-safely-interface-a-5v-signal-to-the-propeller-See-Post-Reply-104 ) (though he says not to do so) for a short time. If that does work, then it should be possible to run the Propeller at up to 130MHz with 3.8V at usable temps, if I am reading the graph in the Propeller Data Sheet correctly.
  • Alexis64Alexis64 Posts: 42
    edited 2011-05-16 19:32
    Cluso99 wrote: »
    That is the top frequency that I found with the RamBlade. RamBlade uses a 6.5MHz xtal (104MHz) and I also supply a 13.5MHz xtal (108MHz) (xtal is plugable).

    However, overclocking is reliant on the proper decoupling capacitors right at the prop power pins. What prop are you using? (DIP or QFP) I found the DIP prop seems to perform slightly faster. Sapieha has run my TriBlade at 120MHz using 15MHz xtal for many months. There is another thread about proper decoupling and speeds.

    qfp44
  • jazzedjazzed Posts: 11,803
    edited 2011-05-16 19:38
    I'll stick with 6MHz at pllx16 (96MHz CLKFREQ). The power consumption is just slightly higher than 80MHz CLKFREQ and it's conservative with the chips over temperature and voltage ranges. Power consumption is a key factor for battery powered devices which I'm selling starting at UPEW. Also 96MHz CLKFREQ is supposed to play nicely with PAL solutions (not that I'll ever ship anything other than a prototype to Europe). No way in heck i'm paying $1.50 for a 6.25MHz crystal though unless I have no choice. I've used 6.5MHz oscillators but would rather not bother because of power consumption. Extreme over-clocking is an accident waiting to happen - nothing is wrong with offering a YMMV disclaimer though unless you know it will fail :)
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2011-05-17 02:21
    I have run a Prop at 4.2 Volts, for a bit,, it was a half way cludge between 3.3 V and the 5 V that the Z80 wanted. It didn't fail but I did fret for the poor thing.

    The only DIP40 that loved 13-14MHz Xtals was the one that I tested at 9 Volts. R.I.P.
  • AleAle Posts: 2,363
    edited 2011-05-17 02:30
    Toby... why not using a 3.3V-powered prop with resistors in series with the Z80 ? It is a NMOS (or CMOS if you are that lucky!) device accepting good old 2.4 V as high... ? just a thought.
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2011-05-17 03:24
    Yes it was a CMOS 10MHz Z80. I wonder just how much slower it would need to run at given 3.3 Volts (if it would run at all) and just bolt it on dirrectly.

    Pullmoll got the Nascom emu running and the interest wained.
  • HumanoidoHumanoido Posts: 5,770
    edited 2011-05-17 05:25
    I have run a Prop at 4.2 Volts, for a bit,, it was a half way cludge between 3.3 V and the 5 V that the Z80 wanted. It didn't fail but I did fret for the poor thing.

    The only DIP40 that loved 13-14MHz Xtals was the one that I tested at 9 Volts. R.I.P.

    As Mike Green has pointed out, running the Propeller beyond normal operating voltages will stress the chip even when performance appears normal, and can result in shortening the props life span.
  • HumanoidoHumanoido Posts: 5,770
    edited 2011-05-17 05:27
    Alexis64 wrote: »
    Hi, We've been testing the propeller's speed while running our telemetry units as well as a variety of graphic video drivers and we've found that using a 14.31818Mhz Crystal ($0.55 at digikey) will reliably clock the Propeller at 114,54544 Mhz with a pllx8 ... The test was conducted at different temperatures for the past 48 hours. Now that you have this information, go and get a 14.31818Mhz Xtal (or rescue it from surplus PCBs) and start developing a better, cooler, video driver :) ...P.S. both the loaded (18pf caps) and the unloaded crystal worked fine - although I personally prefer the 18pf loaded crystal. The beginning of your demos (like graphics.spin etc) will need to be altered as follows:
      _clkmode = xtal1 + pll8x
      _xinfreq = 14_318_180
    
    Have fun.

    Does the prop have any rise in temperature?
  • prof_brainoprof_braino Posts: 4,313
    edited 2011-05-17 06:00
    Humanoido wrote: »
    beyond normal operating voltages will stress the chip even when performance appears normal, and can result in shortening the props life span.

    Do folks use extra cooling when overclocking micros? The PC overclockers say that a 20% overclock doesn't significantly affect life as long as the temps are kept at low (closer to room) temperature. Of course an overclocked PC is obsolete after 6 months so maybe "life span" is defined differently in these contexts.
  • Alexis64Alexis64 Posts: 42
    edited 2011-05-17 06:06
    Humanoido wrote: »
    Does the prop have any rise in temperature?

    It's been running for 3 days now (running full graphics screens) and honestly, the QFP44 does not seem to be any different in temperature than a similar board testing next to it running @ 5Mhz ... I am sure (logic dictates) that there will be a difference in power consumption and temperature, however how much of a difference still need to be determined.
  • Alexis64Alexis64 Posts: 42
    edited 2011-05-17 07:15
    Here is a video of the speed difference between 80 Mhz and 114 Mhz

    http://www.youtube.com/watch?v=rd12opYAa3Y
  • Toby SeckshundToby Seckshund Posts: 2,027
    edited 2011-05-17 08:55
    As I said the accidental test at 9 Volts did cause a shortening of the Prop's life span. Interestingly the SD card, which got ********* hot, worked again once it cooled down.

    The Prop, 512KB memory and the EEPROM weren't so charitable.

    As for cooling I found that the Prop could sometimes read a SD card when frozen but not at room temp. That turned out to be unwanted capacitance or an open circuit lead (stub)
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2011-05-17 08:59
    [quote=Toby Seckshund[/quote]The Prop, 512KB memory and the EEPROM weren't so charitable.[/quote]
    But imagine the high they must have felt up to the very end!!

    -Phil

    "Friends don't let friends do Vin."
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 09:15
    "Friends don't let friends do Vin."
    :)
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 09:58
    Ok I am making a test. I am going to put together a 3.8V power supply, and use a 8MHz Crystal to see how the Prop does at 128MHz with a 3.8V Vdd, I will also test at 4.0V and 4.2V just to see. Once I have the Crystals, I will run multiple boards at 128MHz with the mentioned power supplies for 1 week at each voltage, running high res video on each cog to push its limits.
  • jazzedjazzed Posts: 11,803
    edited 2011-05-17 10:21
    Ok I am making a test. I am going to put together a 3.8V power supply, and use a 8MHz Crystal to see how the Prop does at 128MHz with a 3.8V Vdd, I will also test at 4.0V and 4.2V just to see. Once I have the Crystals, I will run multiple boards at 128MHz with the mentioned power supplies for 1 week at each voltage, running high res video on each cog to push its limits.
    Your test will be worthwhile for just that chip and that application.

    However, your test is not substantial enough to prove anything of value for a chip "population" sample. YMMV and especially OUMMV. That is "other users' milage may vary" which is important.

    You should also add temperature variations and low voltage in addition to high voltage for "four-corners" test. Limits specified by the processor should be used. Of course all 8 COGS have to be running flat out with CTRA/CTRB operational; running video only tests PLL mode on one counter. Once you have a fully qualified test, you can run "four-corners" test with each crystal frequency. Each corner should ramp to target temperature within an hour and soak for 8 hours at least twice. After that you can do MTBF life testing.
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 10:26
    Thank you Heater.

    Heater:
    What type of code would you recommend to provide the most even
    over all test for all cogs. As to thermal testing My equipment available will limit me to more crude methods.
  • jazzedjazzed Posts: 11,803
    edited 2011-05-17 10:39
    As to thermal testing My equipment available will limit me to more crude methods.
    This is why Parallax produced the Propeller specification.
    Read and enjoy the benefit of their engineering.
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 10:50
    Jazzed:
    The results of my testing will definitely be YMMV/OUMMV, though I would still wish to test as thoroughly as I can. So what type of code would be needed to fully test out all cogs to there limits in every way that matters? I will be using as good of a cross section of Props as I can with my limited resources.

    Also:
    Is it possible to produce a 1 pin signal with a counter that can be used by the PLL on another Prop? If so I could use a Propeller to generate the clock for testing, and not be limited by the frequency of available crystals.
  • jazzedjazzed Posts: 11,803
    edited 2011-05-17 11:07
    The results of my testing will definitely be YMMV/OUMMV, though I would still wish to test as thoroughly as I can. So what type of code would be needed to fully test out all cogs to there limits in every way that matters?
    I'm not THE expert, but I believe that running video on every COG is very close to a full stress test. I would add producing an NCO signal from each COGs second counter that is within FRQA spec limits. Don't use waitcnt in any COG under test for stress testing.

    It would be good to have a functional sanity check test which could check every function that could be run periodically. That test is a little harder to specify and I don't have time for researching that just now.

    Just using one counter out of spec will heat up a propeller chip. A temperature probe from a DMM would be good for monitoring that.
    Is it possible to produce a 1 pin signal with a counter that can be used by the PLL on another Prop? If so I could use a Propeller to generate the clock for testing, and not be limited by the frequency of available crystals.
    Yes, this can be done easily. You could also use the other propeller to verify the second counter in NCO mode on the test target's COGs are running. Testing with crystals would give better qualified data since it approximates typical use, but you could at least verify some basic assumptions by generating different clocks.
  • Alexis64Alexis64 Posts: 42
    edited 2011-05-17 11:15
    Jazzed:

    Also:
    Is it possible to produce a 1 pin signal with a counter that can be used by the PLL on another Prop? If so I could use a Propeller to generate the clock for testing, and not be limited by the frequency of available crystals.

    That is exactly what I used to test the prop frequencies - Chip wrote that object a few years back - it's in the OBEX ...
  • Cluso99Cluso99 Posts: 18,066
    edited 2011-05-17 19:44
    Sapieha has run a DIP40 Prop at 15MHz (120MHz) for more than 6 months without problems. We have not elevated the supply voltage. My RamBlades are tested and shipped with 6.5MHz (104MHz) and have 13.5MHz (108MHz) xtals supplied as well. However, RamBlades 2 pins reserved for FDX comms or 1pin TV (uses VGA mode) and 1pin Keyboard. The remainder are used for SRAM and microSD interfacing. In the 1pin code all 8 cogs are running, and otherwise 7 cogs are running.

    The prop runs very cool and this is one of the reasons overclocking works so well. When we started discussing 100MHz operation, we got a very unofficial nod that it was probably ok from Chip. overclocking does require special decoupling at the prop pins and there is a thread devoted to this.
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 19:48
    Clusso:
    Sounds like I will not have a problem then. I am currently running 4 test Props at 130MHz with a supply of 3.8V, lets see what happens.
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2011-05-17 20:13
    Here's what I don't understand: Why use a crystal higher than 8 MHz (except to save board real estate) when, say, a 7.5 MHz crystal running at PLLx16 gives you the same performance as a 15 MHz crystal running at PLLx8? The fact is that with a 15MHz crystal, the PLL's VCO has to run at 240 MHz, which is pushing the envelope way into the unreliable zone, while at 7.5 MHz, it coasts at 120 MHz. The VCO's frequency doesn't get multiplied; it gets divided. PLLx8 is really VCO / 2; PLLx16, VCO / 1. Since the VCO itself has to run at the "x16" rate, regardless of the PLLx? factor, why not let it relax and enjoy the benefits of enhanced reliability?

    -Phil
  • davidsaundersdavidsaunders Posts: 1,559
    edited 2011-05-17 20:19
    Phil:
    I agree with you, on using the PLL16 setting. I do not agree with the 8MHz limit, as I am using 8.125MHz (with PLL16X, for 130MHz) for the 4 test props running on my bench at this time, and will be increasing that if all goes well.
Sign In or Register to comment.