1. VintHPcom@groups.io
Topics

Upgrading HP9000/380 to 385 or '385+' specification #VintageHPComputers

Martin Trusler
 

Re: Upgrading HP9000 model 380 to 385 or '385+' specification
I should start by saying that I haven't done this yet! However, I think I mostly know what to do. Both 33 MHz and 40 MHz 68040 can be found on eBay (part number MC68040RC33A or MC68040RC40A, 168-pin PGA package). The latter were used in some Amiga computers and I have just bought one on eBay. There are are also two oscillator chips on the processor board that need to be changed. Both are 14-pin DIP format packages (but they actually only have pins 1, 7, 8 and 14 in place). The first is located right next to the CPU chip and should run at twice the CPU clock rate. On my (25 MHz) model 380, the oscillator is a Kyocera series K680S-G running at 50 MHz. I found the data sheet for that series and learn that it is a tri-state clock oscillator, which I think means that it has the capacity to be shutdown, presumably to allow the CPU to rest. The Amiga store (Vesalia Computer  online https://www.vesalia.de/e_oscillators[7297].htm) has both the 50 MHz and 80 MHz versions of this chip. I'm shooting for 40 MHz CPU clock speed, so have ordered the 80 MHz oscillator. Probably there are other sources for a 66.67 MHz oscillator, e.g. Mouser. The pinouts are as follows: 1 = Control, 7 = GND, 8 = O/P, 14 = +5 Vdc

 

The purpose of the second oscillator isn't totally clear to me but my guess is that it's something to do with memory refresh. Maybe someone knows for sure? It is easy to locate on the processor board because it looks the same as the CPU oscillator. On my 380, it is a standard clock oscillator from Saronix running at 40 MHz (series NCT 070 C). The pinouts are: 1 = N/C, 7 = GND, 8 = O/P, 14 = +5 Vdc. So this oscillator runs at 1.6 times the CPU clock frequency, which implies 64 MHz for a 40 MHz CPU or a rather odd 53.33 MHz for a 33.33 MHz CPU. One also needs RAM boards with 60 ns memory chips (70 ns might do). These I don't have yet and I'm still searching! The ones I have are 80 ns chips. Finally, next to the CPU, you will find a jumper with positions for 25 and 33 MHz CPUs. The idea of going up to 40 MHz came from George, who has done this. He also told me about the memory speed issue. A 40 MHz CPU should run fine at 33 MHz, so my plan is to try for 40 MHz (if/when I get some fast RAM) but go down to 33 MHz if it doesn't work. I'll share my success or failure when I get there.

If anyone can add to or correct these details, please do!

Martin
HP75_8687
 

380 ( 68040 25 MHz bus clock, 50 MHz processor clock) and 375 ( 68030 50 MHz ) share the same type of memory, HP 98229E 2 x 16 MB simm 80 ns.
If the memory is fast enough for 50 MHz bus clock when combined with the 68030 processor in a 375, why should there be a problem with an overclocked 68040 ( bus clock 33 or 40 MHz ) ?
Martin Trusler
 

The 68040 runs at half the external clock frequency, whereas the 68030 doesn’t divide by two. Therefore both the 25 MHz 68040 in the 380 and the 50 MHz 68030 (in the 375 have the same external clock frequency. The advice about needing faster memory for a 40 MHz 68040 (80 MHz external clock) is from George at ADC Colorado. He mentioned that later revisions of the 98229x series RAM boards used 60 ns chips. I’m not sure if these are actually necessary for operation with a 33 MHz 68040, possibly not. One could check inside a model 385 or 433 but I don’t have either. 
Martin


On 22 Jul 2019, at 20:00, HP75_8687 <michael.fehlhammer@...> wrote:

380 ( 68040 25 MHz bus clock, 50 MHz processor clock) and 375 ( 68030 50 MHz ) share the same type of memory, HP 98229E 2 x 16 MB simm 80 ns.
If the memory is fast enough for 50 MHz bus clock when combined with the 68030 processor in a 375, why should there be a problem with an overclocked 68040 ( bus clock 33 or 40 MHz ) ?

HP75_8687
 

Thanks for that explanation.

 

Currently there is a 98229E for sale on Ebay; the RAM chips are labelled as 80ns, unfortunately.

The modules seem to have special non-standard connectors and form factors, compared to other simm modules. So it seems to be difficult to find or build a faster module. Are you planning to replace the RAM chips on the module?

Unfortunately I haven’t got a 385 or 433 machine either.

 
Martin Trusler
 

Thanks for the info! Right now I’m after any of those RAM boards, especially the larger capacity - would you send me the link please? I saw some recently on eBay that were labelled 70 ns but, when I tried to buy them, the vendor reported them lost!

I have no idea of how to make replacement boards!
HP75_8687
 

You might not have seen the offer on Ebay.co.uk because the seller (USA) doesn't offer international shipping, but you might be able to persuade him to change that, or you can make use of a forwarding service.

https://www.ebay.com/itm/98229E-HP-32MB-Memory-Module-2x16MB-98229-66524/173965778173?hash=item28812b1cfd:g:jYkAAOSwiR5dLM2i
Martin Trusler
 

I have now experimented with upgrading the model 380. I used a 40 MHz MC68040 sourced from eBay. Upgrading to operation at 33 MHz is straight forward - just replace the CPU, change the clock oscillator to 66 (or 66.6667) MHz and move the jumper located near the CPU to the 33 position. It does not seem to be necessary to make any other changes. Basic/WS then operates 33% faster as expected and HP-UX 9.1 runs perfectly (at least in my limited tests).

I tried operating at 40 MHz also. Basic/WS appear to function absolutely normally and certainly runs 60% faster than with the 25 MHz CPU. I have not checked for subtle errors. Unfortunately, HP-UX exhibits strange behavior and does not run reliably. Initially, I had 32 Mb of 80 ns RAM on board and had severe errors with HP-UX.  I then fitted 32 Mb of 70 ns RAM. In this case, HP-UX booted without error but the system hung in HP-VUE. So it looks like 33 MHz is the fastest for reliable operation. 

I have not attempted to increase the speed of the second oscillator (Y2, 40 MHz) and I also notice a third crystal (Y1) with a frequency of 20 MHz, function unknown.

I've uploaded a file with step-by-step instructions for the 33 MHz upgrade - I hope this helps.
HP75_8687
 

interesting, thank you Martin for sharing your experiences.
You mentioned that you replaced 80 ns memory with 70 ns RAM.
Was this necessary for reliably running BASIC/WS at 33 MHz ?
And from where did you receive 9000/380 RAM SIMMs in the faster version? Seems to be almost impossible to obtain these.
Martin Trusler
 

Both Basic/WS and HP-UX (9.10) seem to run completely normally at 33 MHz using the standard 80 ns RAM. This makes sense as the board and memory modules were clearly designed to run up to 33 MHz. It's only when I tried to push it up to 40 MHz that problems appeared with HP-UX. Basic/WS continued to function normally at 40 MHz in as far as I could tell but more extensive checking is needed to be fully confident. HP-UX definitely didn't work at 40 MHz. However, it did at least boot error free when I had the 70 ns RAM on board but not with the 80 ns RAM, so there is some evidence there that RAM speed might be important. I didn't like to test HP-UX further at 40 MHz for fear of corrupting my hard drive.

The 70ns RAM I found on eBay was from Dataram: 2 x 16 Mb modules, module part number 69311 B1 Rev.190500024. The following post has information on memory modules for 375/380/385 from HP, Kingston and Dataram:
Not listed there are the later 4 Mb modules from HP, part number 29229-66525 (A3110-56) based on 4Mbit chips.

George at ADC Colorado told me that 60 ns modules were made but I'm not sure which brand and I've never seen one.

By the way, in my last post I incorrectly described the second clock oscillator (40 MHz) as Y2 but it's actually Y3 (Y2 is the CPU clock oscillator). See here for a nice photo of the processor board:

Martin 


On Sun, 11 Aug 2019 at 15:40, HP75_8687 <michael.fehlhammer@...> wrote:
interesting, thank you Martin for sharing your experiences.
You mentioned that you replaced 80 ns memory with 70 ns RAM.
Was this necessary for reliably running BASIC/WS at 33 MHz ?
And from where did you receive 9000/380 RAM SIMMs in the faster version? Seems to be almost impossible to obtain these.



--
J P Martin Trusler
Tel: +44 (0)7734 590826
HP75_8687
 

Following the instructions given by Martin I upgraded my 380 ( 80ns RAM ) to 33 MHz and ran a test program for about one hour ( under BASIC/WS 6.4).
Everything worked flawlessly, speed up factor as expected was close to the 33/25 ratio.

Interesting observation: I ran the test program which mainly calculates sine function values in a loop at first under BASIC 6.2, then under BASIC 6.4.
The results are amazingly different. Under 6.4 the program is about 4 times faster!
The FPU of the 68040 processor doesn't implement trigonometric functions in hardware (the 68882 does), instead the operating system has to provide a software library. ( Earlier versions of BASIC don't know about the 68040 and therefore are not compatible with this CPU.)
The floating point library obviously has been heavily optimized from version 6.2 to 6.4.
Martin Trusler
 

That's great. Are the MCMATH binaries different at 6.2 and 6.4, or is the difference in the kernal?
Martin
J-F Garnier
 

On Sun, Aug 18, 2019 at 07:55 AM, HP75_8687 wrote:
Interesting observation: I ran the test program which mainly calculates sine function values in a loop at first under BASIC 6.2, then under BASIC 6.4.
The results are amazingly different. Under 6.4 the program is about 4 times faster!
The FPU of the 68040 processor doesn't implement trigonometric functions in hardware (the 68882 does), instead the operating system has to provide a software library. ( Earlier versions of BASIC don't know about the 68040 and therefore are not compatible with this CPU.)
The floating point library obviously has been heavily optimized from version 6.2 to 6.4.
A improvement factor of 4 is very surprising and intriguing !
This would mean that the previous versions were very poorly optimized!

Can you post your test program, so we can check on other machines?

On Sun, Aug 18, 2019 at 12:50 PM, Martin Trusler wrote:
That's great. Are the MCMATH binaries different at 6.2 and 6.4, or is the difference in the kernal?
The MCMATH bin is the Motorola library for the 68040 and is unchanged from 6.2 to 6.4 (if we trust the bin version 6.0):
MCMATH 6.0   Copyright HP 1991   MCMATH Motorola Math Libraries

@Martin
In an other message, you said you have version 6.3. Would it be possible to make disc images (hpi ot td0) of the installation discs (if you have them)?
6.3 is the last version from HP, 6.4 was made by TAMS, Having the 6.3 would be useful to find when the changes were made.

J-F
Martin Trusler
 

Annoyingly, I seem to have lost my version of BASIC/WS 6.3 - sorry about that! Is the one on Bitsaver (http://www.bitsavers.org/bits/HP/HP_9000/BASIC/) a WS or HP-UX version? It's an image of a tape archive, so it might be for HP-UX.
Martin
J-F Garnier
 

Yes, it's a WS version in a LIF disc image format, as we discussed it recently. It's better than nothing but it's a system file (no individual bin files), apparently built for a 68040 (there is the MCMATH bin in it) and I'm not sure it works correctly on other machines, actually  it doesn't work very well in Olivier's Series 200 emulator, I got strange errors for a simple math program.

Any other copy of version 6.3 somewhere?

J-F
green
 


hi Martin, group,

Does anyone know:

Could this CPU speed up be done on the HP R/332 Controller ?

thank you,
r

**
Martin Trusler   08/10/19   #1991
08/10/19   

I have now experimented with upgrading the model 380. I used a 40 MHz MC68040 sourced from eBay. Upgrading to operation at 33 MHz is straight forward - just replace the CPU, change the clock oscillator to 66 (or 66.6667) MHz and move the jumper located near the CPU to the 33 position. It does not seem to be necessary to make any other changes. Basic/WS then operates 33% faster as expected and HP-UX 9.1 runs perfectly (at least in my limited tests).

I tried operating at 40 MHz also. Basic/WS appear to function absolutely normally and certainly runs 60% faster than with the 25 MHz CPU. I have not checked for subtle errors. Unfortunately, HP-UX exhibits strange behavior and does not run reliably. Initially, I had 32 Mb of 80 ns RAM on board and had severe errors with HP-UX.  I then fitted 32 Mb of 70 ns RAM. In this case, HP-UX booted without error but the system hung in HP-VUE. So it looks like 33 MHz is the fastest for reliable operation. 

I have not attempted to increase the speed of the second oscillator (Y2, 40 MHz) and I also notice a third crystal (Y1) with a frequency of 20 MHz, function unknown.

I've uploaded a file with step-by-step instructions for the 33 MHz upgrade - I hope this helps.
Martin Trusler
 

Ummm ... there are some important differences. The processor board for the 375/380/385 was designed for the 68040 CPU running at up to 33 MHz. As far as I know, the 332 and R/332 processor boards weren’t intended to run the 68030 any faster than 16.7 MHz. That said, you might be able to get it to run a bit faster. I’d try putting a 25 MHz 68030 in the CPU socket and experiment with increasing clock speed. So you’d need several trial oscillators and also the FPU will need a speed upgrade if you have one fitted. Memory speed could be a key limitations but it would be fun to try!
Martin

On Sun, 10 Jan 2021 at 17:33, green <hrgerson@...> wrote:

hi Martin, group,

Does anyone know:

Could this CPU speed up be done on the HP R/332 Controller ?

thank you,
r

**
Martin Trusler   08/10/19   #1991
08/10/19   #1991  

I have now experimented with upgrading the model 380. I used a 40 MHz MC68040 sourced from eBay. Upgrading to operation at 33 MHz is straight forward - just replace the CPU, change the clock oscillator to 66 (or 66.6667) MHz and move the jumper located near the CPU to the 33 position. It does not seem to be necessary to make any other changes. Basic/WS then operates 33% faster as expected and HP-UX 9.1 runs perfectly (at least in my limited tests).

I tried operating at 40 MHz also. Basic/WS appear to function absolutely normally and certainly runs 60% faster than with the 25 MHz CPU. I have not checked for subtle errors. Unfortunately, HP-UX exhibits strange behavior and does not run reliably. Initially, I had 32 Mb of 80 ns RAM on board and had severe errors with HP-UX.  I then fitted 32 Mb of 70 ns RAM. In this case, HP-UX booted without error but the system hung in HP-VUE. So it looks like 33 MHz is the fastest for reliable operation. 

I have not attempted to increase the speed of the second oscillator (Y2, 40 MHz) and I also notice a third crystal (Y1) with a frequency of 20 MHz, function unknown.

I've uploaded a file with step-by-step instructions for the 33 MHz upgrade - I hope this helps.

--
J P Martin Trusler
Sent from my iThing
1 - 16 of 16