A TF534 build in 2021

TF534 - 68030 + More RAM Board (More compatible with amiga hardware)

Moderators: terriblefire, Terriblefire Moderator

Post Reply
Eduardo L M
Posts: 15
Joined: Thu Apr 08, 2021 7:27 pm

A TF534 build in 2021

Post by Eduardo L M »

I'm a little late in the party, but I would like to share my experiences building a TF534 this year. I live in Brazil, and it was very difficult to get some parts (specially SRAMs). It was also hard to reunite all the information needed to do the build, but reading every post in this forum helped a lot. I also had a great amount of help from Keith Mcgerr, thanks a lot, Keith!

The most difficult (and costly) part was the SRAMs. I first tried buying them from ebay / aliexpress, despite all warnings. And guess what? Got fakes. Then I gave up and bought them from Mouser (very expensive!).

I got all the other parts from China. Surprisingly, CPLDs worked very well (I guess, keep reading). I was surprised also with plastic 68030 processors: I got 7 of them from a chinese eBay seller, and all of them (but one) worked well at 50MHz. I didn't expect that.

I build 3 boards. During builds, I had the usual problems with tiny solder bridges on CPLD legs, and bad contacts on 68000 socket. I solved the last one by replacing the socket with turned pin header strips, which makes excelent and stable contact.

I also tested with chinese FPUs, and surprisingly, almost all of them worked flawlessly at 50MHz.

The only issues that remain are these:

- I have some video artifacts (horizontal flashing lines) during Frontier opening sequence. Here's the video:



I suspect it may be due to the 74AC16245 I bought from China. It looks like it's remarked, it may be a slower version (74HC16245), causing glitches in the communication between accelerator and Amiga buses. But it might be even the CPLDs, I'm not sure. An important detail is that the artifacts are on video only, there are no crashes or freezes, so stability is great so far, and this fact leads me to think the CPLDs are not the issue.

- I need to find a heat management solution: some of the 68030s I tested get very hot. Not to the point of burning my fingers, I can stand the heat, but it's way beyond a warm and comfortable temperature. I'm looking for low profile heatsinks, so I can close the A500 housing.

Now I'm looking for a good stress test to ensure the boards are really stable and reliable. Is there any standard procedure or suggestion for that?

Eduardo
dalek
Posts: 224
Joined: Thu Nov 08, 2018 11:03 am
Location: NSW Australia

Re: A TF534 build in 2021

Post by dalek »

I would try swapping the oscillator for a 40MHz one and see if that gets rid of the flashing. I seem to remember I had the same issue and it was either fixed by dropping back to 40MHz or adding heatsinks to the CPU and larger CPLD.

Not had much luck with the CPLDs from China - some work, some "mostly" work to the point I'd tear my hair out trying to work out what was wrong. If the above doesn't fix the flashing, try getting genuine CPLDs.

If you really want 50MHz go for a TF536.

For stress test, see here: https://exxosforum.co.uk/forum/viewtopic ... 717#p46717
User avatar
8 Bit Dreams
Moderator Team
Moderator Team
Posts: 785
Joined: Fri Nov 09, 2018 7:12 am
Location: Germany

Re: A TF534 build in 2021

Post by 8 Bit Dreams »

Yes, there are,
The fact that Frontier boots doesn't mean anything, You may run Povray test to be sure Your CPU/FPU are fine under load.
Choose proper test with FPU for TF536, it will run about 30 - 40 minutes, show pictures at the end, and check hashes of them, if these mach - you'll see "OK", if mismatch (FPU makes error)- You'll see "Failed", if CPU craps out - script will crash..

SAVE_20210830_191556.jpg
SAVE_20210830_191556.jpg (79.67 KiB) Viewed 3469 times
Retro computer hardware & repair in Germany
alenppc
Moderator Team
Moderator Team
Posts: 906
Joined: Thu Nov 08, 2018 12:59 pm

Re: A TF534 build in 2021

Post by alenppc »

It's a miracle your board boots at 50 MHz at all. This SEVERELY overclocks the 55ns rams (the timings are from the 12ns TF530 rams) and the vast majority won't even boot at that speed. I have written this a million times in vain... the upper limit for the 534 is 40 MHz and you should stick to that.
Eduardo L M
Posts: 15
Joined: Thu Apr 08, 2021 7:27 pm

Re: A TF534 build in 2021

Post by Eduardo L M »

Thanks a lot for the replies and povray tip! I'll do the stress test as soon as I can. The flashing is not bothering me too much, but I'll get genuine CPLDs and bus transceiver to see if it makes any difference.

@allenppc, indeed, I saw some posts from you stating 50MHz is too much for a TF534, but I really think there's something else going on here. I built 3 boards, all of them work very well at 50MHz, except for the flashing during Frontier opening (by the way, I get the same flashing at 40MHz too). I ran many cycles of ATK memory test at 50MHz, and had not even a single error. To be honest, I had problems with a single RAM chip I forgot to solder properly. But as soon as I soldered it, errors were gone.

I'm using the latest CPLD binaries TerribleFire published at GitHub (tf53x_2019_02_18_alpha.zip). Are you sure this version doesn't include any kind of wait state, to allow the TF534 to achieve 50MHz without overcloking the 55ns RAM chips too much? I think it's strange all my 3 boards ran so well so far.

Is there any particular RAM pin I could scope to check the real speed they're running?

Anyway, I will be sure only after povray tests, I'll post the results here!

Eduardo
Eduardo L M
Posts: 15
Joined: Thu Apr 08, 2021 7:27 pm

Re: A TF534 build in 2021

Post by Eduardo L M »

Regarding the 50MHz question, I was wondering..

The period of a 50MHz clock signal is 20ns. And I don't think the CPU ever do memory operations in consecutive clock pulses, they should be at least 2 clocks apart, giving a 40ns time for the memory to work. This is surely less than 55ns, but not much less. And this is considering a memory access every 2 processor clock cycles, what I think is too much, the operations should take several clock cycles in a CISC processor like the 680x0 family. I'll do some research on this!

And the 12ns memory the TF530 uses is much more than needed, that means a 83MHz clock! So this memory can go a lot faster than it goes on the TF530.

Eduardo
Eduardo L M
Posts: 15
Joined: Thu Apr 08, 2021 7:27 pm

Re: A TF534 build in 2021

Post by Eduardo L M »

I still have to do the PovRay test, I'm having a hard time figuring out how it works.

Anyway, I did a very long memory test on ATK. It ran for 75 testing cycles at 50MHz, and I had not a single glitch, with a gold CPU.

With a plastic one, I had memory fails, but those don't happen with the gold one.

Eduardo
Post Reply

Return to “TF534 (OBSOLETE)”