Overclocking the TF330. From success to failure.
 

During the week I successfully overclocked my TF330.

You can watch my video on it here: https://youtu.be/BG2vLDR-1EY

Long story short I swapped the 100mhz crystal for a 108mhz to bring the clock speed up to 54mhz. All was working fine. I used the machine that evening of the overclock for several hours without issue.

Going out to it this morning though and it's dead. Power on and black screen only, occassionaly you'll get the jingle with no display or sometimes the display signal will drop off.

First thing I did was to go over my soldering at the crystal which was fine.

Then I tested the console without the riser then with the riser but no tf330. Fine.

Then I tried the tf328 in the riser and that's fine too so nothing wrong with the console or the riser. (I didn't think there would be but you need to check these things).

So I then swapped back to the 100mhz crystal and same result. Black screen or occasional jingle or occasional display drop out.

Last thing I tried was a different 108mhz crystal but same result.

If left on the black screen for any length of time the cpu gets very hot. When running I was getting a maximum chip temperature of about 40°C. After maybe a minute on black screen the pu got to just over 47°C

So what's happened?

Has the cpu died?
Has another chip on the tf330 died?
Would it be worth getting the diagrom chips to test what's going on?

Unfortunately I don't have another 030 to test on the card.

What PSU are you using? Maybe the extra power draw from the overclock pushed something to break?

Just using the standard power supply. I suppose that could be the issue but to be honest I'm leaning heavily on the cpu has died. To be sure though I'll measure the 5v rail to make sure its not dropping.

Having over-clocked one or two things in my time with Amiga's & indeed PeeCee's, I'd be betting on the CPU being fried. Those types of CPU package are not as resilient as PGA Ceramic. (Like the ones used on the Blizzard Cards)

I believe the CPU on those cards is a 40Mhz part, (or I've only seen pre-release boards) already over-clocked to 50Mhz. Hardly surprising the extra 4Mhz was just too much to handle Cap'n! - Apologies if they now ship with 50Mhz fabrications.

When over-clocking, it isn't only down to how far the Fabrication will go, & not all fabrications do the same mileage. Surrounding components & their respective circuits also take a strain & should also be considered.

Perhaps one of the most unceremonious Over-clocking events I had the pleasure of experiencing was with a CS MKIII. I over-clocked the CPU Synchronously with the SCSI & saw some absolutely fantastic SCSI speeds @ 38Mb/sec. Unfortunately & in a very short time, the card died for frying the SCSI's Logic chip. - Cest La Vie!

Might be worth contacting Stephen Leary & asking. - Would be great to hear success again for you. :great

You are correct in that the CPU was a 40mhz model and I agree that it was probably pushed to far and that is most likely the cause of the failure. I'm getting a 50mhz gold top chip so I'm hoping that will solve the issue and be safe at the 54mhz speed.

I've decided if I get it up and running again at 108/54mhz I'll just leave it at that with some cooling and be grateful its working. I don't want to push out to 120/60mhz and risk damaging anything further.

Glen,

It may still be worthwhile chatting to Stephen and asking if the IDE is in Sync or out of Sync. Guessing it is in Sync as there is only one timing X-Tal on the board but I'm not the guy designing the board.

I have heard of the 030 being pushed to 70Mhz & has been happy for years @ that speed, but would guess active cooling is employed.

Presuming the IDE takes it's timings from the X-Tal for the CPU (there's a divide by two counter on the card somewhere) it would be handy to know the speed limitations of the IDE Controller chip. It could be a cheap part & have a smaller tolerance than other chip brands.

Don't forget the Ram Speed too.

More Power, More Speed! :D

Charlie

PS - Be wary of MC68030's from China, or any I.T. component for that matter. - A lot of them are "Mickey Mouse" event's & by that I mean, "Fictitious".

I'll try to get hold of Stephen and see what he thinks the maximum potential clock speed of the board would be if its limited to purely how far you could push the CPU or another factor.

I have an full 030 50mhz coming from a trusted source so no problems there.

The thing I find most odd is that the system was stable for a good 2 hours the night I done the mod and recorded the video. That's sitting idle, running demos, playing games etc so it was well used and rock solid. It was the next day I went to power it on and nothing.

Actually at this point I'm hoping it is a dead CPU and not something else!

Spoke with Stephen via PM over on the exxos forum. The answer was a quick "don't overclock it" so perhaps I should put it back to 100mhz.

I'm a complete &*#$@£+ idiot. An absolute arse. I shouldn't be trusted to open a tin of beans let alone working on hardware. Actually just because this is such an epic screw up of idiotic proportions I was honestly of 2 minds if to post this but its better to own up when you make a mistake. Even if such admission just confirms I'm a complete and total bonehead.

In short there's nothing wrong with the bloody thing. Rather there's a complete lack of anything between my ears.

Was going back over the board today carefully measuring everything then found a jumper lying on the floor that had somehow fell off the board. You know the enable jumper on the top bloody right of damned accelerator. Put that back on and away she went. No issue, booted and has been happily running the Frontier intro for the last hour not missing a beat.

I'm lost for words and am away to chastise myself appropriately.

But that’s a good result! :)

Now you can get back to overclocking it to death

Glen, those jumpers are tiny. Easily done.
Really glad it’s fixed, keen to see it at Amiga Ireland!

I'm curious...
Do you have tried 120Mhz with the older 68030@40Mhz too ?
Perhaps the MMU on the full 68030@50 doesn't like the overclock.

BTW it was a nice try, and you've reached the 10M barrier, well done !

That thought crossed my mind after I had put it back 108mhz but from conversation with Stephen Leary I'm fairly certain it won't work at 120mhz. I don't think the issue was the processor but rather the code written by Stephen in that TF330 that doesn't like the clock speed and that is by no means a criticism of his excellent work.

My reason for this conclusion is that the machine posted to the CD32 screen at 120mhz so all rom tests must have passed fine. Its the IDE that doesn't like the speed. All clocks on the card are derived from the 1 crystal so you're not just overclocking the CPU but the RAM and the IDE. The CPU can take it fine, the RAM also is fine (its modern memory after all) but IDE I imagine is a lot more fussy about the speed it will work at.

I am a little reluctant to try it again to be honest. The crystal has been off and on maybe 5/6 times now and I don't want to risk lifting pads by overheating. Just with the work I've done so far there is already a little damage to the solder mask and you can appreciate I don't want to risk bricking the board. I'm not an expert with SMD work but getting better.

If you can find a 110mhz crystal I think that may work but they seem like hens teeth for whatever reason. 108 and 120 are the only speeds above 100 that I could easily find and even at that they had to come from China.

To be honest I'm more than happy with 10.5Mips in sysinfo. I believe I've proved a small overclock is indeed possible and everything seems stable at 108mhz.

I ran memory tests for a good hour the other night and it didn't miss a beat, well once I got the test working that is. It seems there is something weird going on in my startup sequence causing the test to fail from within workbench. Its perfect if I boot with no start-up sequence though. I think the crash is because of the MMU library that I've installed. Using the MMU tools I've mapped the ROM to fast ram which is then write protected causing the test to crash. I could be wrong but thats my best guess.

could you tell me how you have mapped ROM to FastRAM? I want to do this with a custom kickstart ROM but havnt had much joy figuring it out

A custom Kickstart will need a MMU. Often there is a special tool provided with the accelerator but you can try CPU FASTROM.

no special tool with the TF330 and no MMU either... which is what the poster is using - hence my question.

Yes surely the IDE is the bottleneck and that's why the CF was corrupted i think.
If it's working good with 108Mhz, keep it like this.

What's the benefit of a 68030@50Mhz with MMU vs 68040@40Mhz@50Mhz in terms of performance ?

I've since fitted a full 030 with mmu to my TF330 and I'm using the fast rom tool that comes with the mmu library to map it. Added about 300kbps to my cf card access speed in sysinfo.

I never ran the full 030 at 50mhz but at 54mhz the difference between the 2 chips is very little. Sysinfo jumped from 10.45mips to 10.5mips and as above the cf card is a bit faster. Some of that though might be down to the oddities of sysinfo itself, I find the scores can change each time to run it.

I'm not sure if other software makes use of the mmu for a speed boost.

The only other thing I can say for sure is that the full 030 runs significantly cooler than the EC variant and since I'm pushing it a bit with the overclock a cooler chip can only be a good thing.

I think clock for clock 040 is maybe twice the speed of 030