RISC OS news and information from Drobe


Beta! \| About us \| Contact \| RSS	Webspace \| Tech docs \| Downloads \| BBC Micro \| Gallery \| Wallpaper

StrongARM to FastARM

Published: 21st Jan 2002, 12:45:06.

When the going gets tough, the tough get faster and silly

Cards on the table time. RISC OS needs new hardware, no secret there. We've had 300MHz StrongARM cards from Castle but they've run out of those so it was back to the 200-odd MHz StrongARMs. Last week, drobe.co.uk learnt that a RISC OS online resource developer had put down his server tools long enough to have a play with some StrongARMs to see just how fast they can really go.

Webmaster of Acornsearch.com, Ian Hawkins, has indeed managed to run a normal StrongARM at 304.8MHz and has made his experimentations public for anyone brave enough to follow. If you are a hardware junkie, this could be an ideal spring project.

You might be aware that processors can be 'over clocked' which basically involves driving them at a much higher clock speed than usual to increases execution speed with the hope of speeding up the computer system as a whole. It's like a slave master banging her drum to a much faster rhythm to get her slaves to work faster. The downside to this is that the processor can heat up and become stressed which reduces the life time of the chip although if you try and cool the processor, you can avoid damaging it in this way.

While Castle used a particular batch of StrongARM chips that could withstand clock speeds of 300MHz (albeit with a large cooling fan), Ian is using a normal StrongARM that is used to clock speeds in the area of 200MHz so it therefore needs an almost ludicrous cooling solution. When drobe spoke to Ian, all he could bring himself to say was, "If you do this yourself, then you're as mad as I".

Links
How to give your StrongARM go faster stripes

Chris, red carpet cleaning division

Discussion

Viewing threaded comments | View comments unthreaded, listed by date

That's nothing new. I did this about a year after the first StrongARM cards became availeble (as soon as the warranty expired). I remember showing it off on the annual meeting of the German Archrimedes Group, so it might even be mentioned in the magazine. Anyway, back then I clocked my StrongARM at up to 528 MHz (where it became a little unstable) and usually used it at 496 MHz. I've just updated a file wich I wrote back then and put it up on my website at www.zimmerle.de for anyone who's trying to do something similar. I suggest someone point Ian to my site, I haven't got his email address.

- Julian G. F. Zimmerle

on 22/1/02 12:54AM [ Reply | Report ]

Thats all quite nice, except the clock multiplier on the board DOESNT LOCK stably above 320MHz. fool.

on 22/1/02 9:23AM [ Reply | Report ]

I've heard that earlier boards were quite able to do this. A friend of mine was once involved in the beta of the StrongARM and RO3.7. It could be clocked (without modification other than flicking dip switches) up to 500MHz, or around that, and remain stable at around 450MHz, if very slightly flaky (crash once every coupld of days) and dramatically reducing the lifespan of the silicon.

on 22/1/02 12:54PM [ Reply | Report ]

The anonymous poster who called me a fool up there should either keep his mouth shut or identify himself. I've been trained in electro-technical engineering for two years and six months before I had to stop because of health-problems and consider myself two have quite a bit of knowledge on the subject. I'm happy to share all the infos I can remember or find on my HDD on the subject to enable people to do the same thing. Specifically I'd be interested in performance results with an up-clocked Kinetic. Just anonymously calling me a fool won't help anyone, except the people who think it's best to keep as much information to themselves as they possibly can. I got some very negative reactions from some people a few years back, when I did it, wich was the reason for me to stop giving away any information on the subject. Yesterday I read about Ian's experiments, looked at the photos of his card, noticed he hadn't put any cooling-equipment on some vital components and thought it might be a good idea to point that out to him, as otherwise he'll probably damage these components pretty soon if he hasn't done so already. It isn't nice to be called a fool when all you were trying to do was to offer some help to someone who obviously needed it.

- Julian G. F. Zimmerle

on 22/1/02 3:18PM [ Reply | Report ]

I use my RiscPC at the bottom of the ocean and attain 15Ghz speeds due to the huge pressures keeping the chip cold. Oh yes - I'm a fish.

on 22/1/02 5:21PM [ Reply | Report ]

With due respect to your above experience, and my own experience with tuned circuits I/F stages, RF amps, and so forth (being a Radio Amateur) for over 6 years and building/tinkering with various valve/silicon based transcievers, could I politley ask what you think happens to a clock doubler/multiplier that's been designed to multiply from 3.5Mhz to 200-300MHz, and you start feeding it twice the clockrate (8MHZ in your case) into it?

In my experience with it both in testing with the SA board it started misfiring after 5.5MHz and above, resulting in instability of the chip, and a lower multiplied frequency. This weekend I will endeavour to experiment some more by testing the output of the clock multiplier with 6 - 12MHz xtals to see exactly what is happening.

I would be interested to see any information regarding how you measured the frequency of the SA whilst running at over 400MHz (the output of a frequency counter would have been my first choice, then a logic probe on various addr/interrupt lines whilst running some basic excercises on the chip, and last some basic internal timing loops (like the port of bogoMips) which run inside the SA cache).

Please accept my apologies for my above anonymouse message whereby I called you a fool.

Regards,

- Ian Hawkins (g0tai).

on 22/1/02 5:25PM [ Reply | Report ]

Drobe seems to have a problem with long comments, so I'll just split it in two:

on 22/1/02 11:45PM [ Reply | Report ]

Ah, finally I've found a way to contact you. I coudn't find your email-address anywhere on that site. Anyway, maybe it would be a good idea to continue this conversation by email, so please feel free to send your reply to julian@zimmerle.de.

Actually I fed it about 217 percent of the clockrate wich is recommended in the documentation. And to be honest, I don't know what exactly happens when you feed that clockrate to a chip that's been cooled down by about 60 degrees celcius. I guess we'd have to ask the chip-developers at Intel what happens, as it's embedded into the chip and they were the ones who designed it.

I didn't actually measure the CCLK as AFAIK it never gets out of the chip unless you pull up TCK_BYP (pin 130) to get the signal out of the chip on TESTCLK (pin 131) wich according to the documentation wouldn't have been a good idea. It says "There are important restrictions on the use of TESTCLK that must be observed to avoid permanent damage to the chip."

So I just had a look at the documentation und figured out at what clockrate the core was supposed to run at and fed the higher clockrate into the chip. Then I ran some simple BASIC programs (because they fit completely into the cache) to see wether they ran much faster. And as they did and the machine didn't crash I assumed that the chip must have been running at the frequency I calculated.

on 22/1/02 11:46PM [ Reply | Report ]

Anyway, if you're willing to risk your chip, then I can recommend the Voltcraft MXC-1600, but be careful when measuring around on the card because AFAIK Acorn used a higher core-voltage VDD (I think it was 0.2V too high, but am not sure) than specified. That's probably also the reason why the chip runs so well at higher clockrates. :-)

Given the enormous production-tolerances I find it quite likely that only very few chips will take an external clockrate of 8MHz and not get anywhere near the 500MHz mark. All I can say is that my chip (almost) did without making the machine unstable.

Hmm, that I'm thinking about it: I feel tempted to get back into the overclocking-scene, now that there're almost noiseless watercooling-systems around. Maybe after the HobbyTronic computer show, when I get some more time...

- Julian G. F. Zimmerle

PS. Of course I accept your apologies, I didn't know it was you. I just react a little allergic to anonymous posters calling me names. ;-)

on 22/1/02 11:47PM [ Reply | Report ]

Someone tell me how to overclock a Kinetic, and I'll give it a try. -- The Doctor

on 23/1/02 12:49AM [ Reply | Report ]

Now that *would* be worthwhile overclocking.

- Ian Hawkins

on 23/1/02 8:50AM [ Reply | Report ]

Don't get Ian started again!

on 23/1/02 10:34AM [ Reply | Report ]

Well, to overclock a Kinetic I would start by trying diffrent jumper-settings (the equivalent of the dip-switches on the StrongARM cards). Castle indicated that they would be willing to provide people with the necessary information. Other than that, you can find the clock-generator (crystal) for the core-frequency at the bottom right of the Kinetic card. It's got a square metal case and the frequency (3.579545) is printed on it. You can try to replace it with one that gives a higher frequency. Start out with crystals that only give a slightly higher frequency, like 4MHz, then work your way upward until the machine gets unstable and revert to the last stable setting. You get the core-frequency by multiplying the crystal-frequency with the multiplier-value (wich you set with the jumpers/DIP-switches - see my website). With all these things you might have to change the environment the chip runs in to get it to operate stable. Specifically you should cool it down as much as you can. Use peltier heat-pumps to cool the chip and cool them with heatsinks and fans or water cooling systems (have a look at www.aquacomputer.de). Modern peltier heat pumps should generate a temperature difference of about 60 degrees celsius. You can increase that value by sticking several of them on top of each other. So if you have a room temperature of 20 degrees celsius, one side of the peltier heat pump cools down to -10 degrees and the other side heats up to 50 degrees as soon as you switch the thing on. But be careful, the CPU might not be the only thing on the card wich needs cooling. I would talk to the hardware engineers at castle about the project first, to get as much information as possible. You might also need to rise the core-voltage a bit to get it to operate reliably.

on 24/1/02 9:31PM [ Reply | Report ]

Sorry, I forgot to append my signature to the previous post. And BTW, I don't take any responsibility if you break your hardware by following my advice.

Best wishes,

Julian G. F. Zimmerle

on 24/1/02 9:37PM [ Reply | Report ]

Just one question. what do you intend to do with all that extra speed anyway. ? It's not as if you need it for super fast gaming. Playing DVD's (er no) So 500Mhz is wasted on a any Risc Pc with a 16Mhz data bus running the same old cheesey Risc OS programs. So with that in mind I decided to have a go at overclocking something useful. My washer is a brand new Hover AAA1400. By overclocking the CPU I can get a faster wash cycle. I have tested this with a program called "cotton wash" which goes through a cycle of soakwashpre rinsrinsspin. The whole thing takes only 15 seconds, however my clothes come out still dirty. Can anyone help or advise me where I am going wrong.

on 12/2/02 6:56PM [ Reply | Report ]

With reference to the dirty clothes... you /did/ add the washing powder, didn't you?

on 25/2/02 3:25AM [ Reply | Report ]

With reference to the comment before the dirty clothes one

The extra speed helps things like ArtWorks, PhotoDesk, as well as some games chug along nicely. You seem to forget that quite a bit of code can fit into the instruction cache of a StrongARM.

on 25/2/02 1:18PM [ Reply | Report ]

Please login before posting a comment. Use the form on the right to do so or create a free account.

Create a new account
Forgot your password?

Search this website