Failing 8641 drivers

[Joerg Hoppe]

Hi,

I now had a failing 8641 in my Unibone setup, for the 2nd time.

For development I run my Unibone in a bare DD11 backplane on my working bench, without case and without cooling. I only have Russian 8641s in use here, which seem to get a bit warmer then the Western originals. And the Unibone is powered-on for months non-stop.

Perhaps the russian clones are bad, or the 8641 generally need cooling.

8641s also tend to break on vintage DEC boards ... maybe UniBone/QBone is just too compatible here.

You suffer from this issue too?

kind regards,

Joerg

[Joan Touzet]

Hi Joerg,

On 2022-09-04 02:52, Joerg Hoppe wrote:

> For development I run my Unibone in a bare DD11 backplane on my working
> bench, without case and without cooling. I only have Russian 8641s in
> use here, which seem to get a bit warmer then the Western originals. And
> the Unibone is powered-on for months non-stop.

Cooling would be the fix in my opinion. None of the original machines would have been without a case and without cooling for more than a brief time. If you have an IR camera you could check the temps of the chips in your setup; the DS8641 datasheet specs a maximum temp of 70C.

Have you noticed if it's always the same chip on the board every time? If you're able to measure power draw for that one chip, that would be informative.

> Perhaps the russian clones are bad, or the 8641 generally need cooling.
>
> 8641s also tend to break on vintage DEC boards ... maybe UniBone/QBone
> is just too compatible here.
>
> You suffer from this issue too?

When we build QBones we do a burn-in test on the drivers for an hour or two for regular units, and we did 24+ hours for the Gold units. So far this hasn't failed any chip -- chips fail within the first 10 seconds if they're going to fail. Overall failure rate of the КР559ИП3 is higher than the DS8641 chip, but both are in the single-digit percentages.

I'm conducting tests with modern chip replacements now for future (non-QBone) work and will let you know, but so far, AM26S10 or SN75138 do look compatible - at least for "small" setups (single QBus backplane, no larger than a 12-slot BA123).

-Joan

[Joerg Hoppe]

Hi Joan,
> Hi Joerg,

>
>
> Cooling would be the fix in my opinion. None of the original machines would
> have been without a case and without cooling for more than a brief time. If
> you have an IR camera you could check the temps of the chips in your setup;
> the DS8641 datasheet specs a maximum temp of 70C.

"Warm" feels more like enhanced body temperature, far far below the 70°.

> Have you noticed if it's always the same chip on the board every time? If
> you're able to measure power draw for that one chip, that would be
> informative.

I'll have a look at it, when the 3rd fails.

>> Perhaps the russian clones are bad, or the 8641 generally need cooling.
>>
>> 8641s also tend to break on vintage DEC boards ... maybe UniBone/QBone
>> is just too compatible here.
>>
>> You suffer from this issue too?
> When we build QBones we do a burn-in test on the drivers for an hour or two
> for regular units, and we did 24+ hours for the Gold units. So far this
> hasn't failed any chip -- chips fail within the first 10 seconds if they're
> going to fail. Overall failure rate of the КР559ИП3 is higher than the
> DS8641 chip, but both are in the single-digit percentages.

I also do an >1h selftest burn-in for every board.
Can't remember any ebay-chip failed there, not even the КР559ИП3.


kind regards,

Joerg

[Ethan Dicks]

On Sun, Sep 4, 2022 at 2:52 AM Joerg Hoppe <ioerg...@gmail.com> wrote:
> I now had a failing 8641 in my Unibone setup, for the 2nd time.
>
> For development I run my Unibone in a bare DD11 backplane on my working bench, without case and without cooling. I only have Russian 8641s in use here, which seem to get a bit warmer then the Western originals. And the Unibone is powered-on for months non-stop.
>
> Perhaps the russian clones are bad, or the 8641 generally need cooling.

I used to make Unibus boards in the 80s and 90s. We sold close to
1500 boards with 8641s. That was never the part that failed in dead
boards (RAM was probably the most common failure). We didn't have any
special cooling on the boards but in those days, our products were
usually run in data centers/computer rooms where the rooms kept around
18C. If the Russian clones run warm, that could definitely be an
issue though. I don't remember the National Semiconductor parts ever
getting noticeably warm.

-ethan