Slow A4000 after overhaul

[Screechstar]

Hi all - would anyone have experience of, or any guidance, with regards to a slow running stock A4000D Rev B with an 25MHz 68EC030 A3630 CPU card, with FPU added?

- 68EC0030 & 68882 FPU
- 3.0 ROMs
- 2MB 80ns FPM chip ram
- 16MB 70ns EDO fast ram (and tried 4 & 8MB 70ns FPM too)
- original PSU recapped and voltages are good within the board


I was lucky enough to buy one locally and the cap leakage was worse than the battery leakage and the board looked not too bad, so I wanted to 'rescue' it.

I decided to go full-on and remove everything that had corroded solder joints so I could clean underneath the IC's and sockets... and then decided it was easy enough to replace with new parts so I used the invaluable A4000 Bill Of Materials on Github and sourced some NOS parts where new parts weren't available. The pics show the extent of the works around the memory sims, up to the mouse/game ports and up into the audio section.

I tested continuity and resistance for all connections for the new components fitted and then after finding & fixing 2 corroded via's, the machine booted!
Sysinfo 4.4, AIBB 6.5 & Amigatestkit and the latest DiagRom seem to show everything works great... but I realised it's running slow!

I searched YouTube and Google and can see the AIBB and Sysinfo show Dhrystone scores around 50% of most stock 030 A4000's and also around half that of their reference A3000's with the same CPU.
Sysinfo shows around 1900 dhrystones and CF HD speed is <1mb/s.

- I've tried stock WB3.0, a modified WB3.0 with Setpatch 44.38 & a HSTWB install of WB3.1 (so lots of new patches and software).
- I tested various FPM and EDO ram modules as chip- and fast- RAM, only as fast as 70ns
- I've tried a new 50MHz crystal


I'll carry on looking for any missed corrosion damage and broken via's but any guidance would be a real help!
... perhaps one of those new chips aren't as compatible within a mixed old / new system as I'd hoped!

Thanks

[Boing-Ball]

Set the clock jumpers correctly for the CPU card?

[Screechstar]

Hi, thanks - good shout to check that!
I’ve just double checked them all and they look ok. The main board is set up for internal clock (the 25MHz clock) and for an 030 and the A3620 had the correct jumpers for the FPU I added.

Ta - definitely worth checking!

Almost forgot… I am running benchmarks with caches and burst on (not data burst as that normally defaults off anyway).

[thebajaguy]

Edit - misread the machine type. Deleted my comment.

[patrik]

What does sysinfo say the 030 clock frequency is?

[Screechstar]

Hi,
It’s an interesting mix…
- Sysinfo CPU 25 MHz
- AIBB - CPU & FPU 25MHz
- WhichAmiga - CPU 25 MHz & FPU 24MHz
- SysSpeed - CPU 25MHz & FPU 230MHz!!!!!!!?

I’ve added a few screen shots… I think Sysinfo and AIBB looks setup properly?

I don’t have an oscilloscope to be able to probe properly alas and I am an enthusiastic novice so when things don’t work I can kind of work things out from schematics as to what broke… but not working 100%… eek! That’s a new challenge for me!
Ta!

[Screechstar]

I was just thinking if it could still a ram speed issue?
AIBB memtest shows 2.74Mb/s or 35% of that from an A3000/030 25MHz.
SysSpeed shows the attached results.

I’m not sure if that’s good or not… I’m struggling to find a benchmark. I’ve googled honest!

I did replace only one of the many F245 IC’s, along with many memory related chips around the battery and one of the fast ram sockets has a bodge wire (corroded via!) having said that I have tested with only 4mb fast ram and that’s not on the bodge-wired socket.

I’ll maybe try bustest as I see people use that for memory speed benchmarks? I’m not sure which is considered the most ‘reliable’ tool but any advice gratefully received!

[patrik]

Good idea with bustest. I have a few ones recorded from my A4000/030 when it had the 030 card:
http://www.megaburken.net/~patrik/am..._200ns-ROM.txt
http://www.megaburken.net/~patrik/am..._160ns-ROM.txt
http://www.megaburken.net/~patrik/am...SkipRamsey.txt

[Screechstar]

Wow! That speed data is fantastic - thank you Patrik, that's exactly the comparison data I was after!

So my results (below) are waaaaay slower and the cycle times are higher. That's true even for the ROM.

So the question is - is this the cause of the slow speed or an effect of something else being slow... my suspicion is that I've changed so much of the memory related IC's that this could be the primary cause.

The difference in our systems is that I'm running KS3.0 with WB3.0 and only have a Rev 9 Super Buster, but could these differences alone really make such a speed difference?

I think I'm going to spend a bit more time investigating the spec of the new chips I installed compared to some old datasheets... but once again, any advice gratefully received!

A4000 REV B
A3630 030@25MHz
2MB chip 80ns FPM
16MB fastmem 70ns EDO
ROM timing at 200ns setting
Super Buster 390539-09
Kickstart 3.0
Workbench 3.0

BusSpeedTest 0.19 (mlelstv) Buffer: 262144 Bytes, Alignment: 32768
===============================================
memtype addr op cycle calib bandwidth
fast $07078000 readw 342.8 ns normal 5.8 * 10^6 byte/s (-28% slower)
fast $07078000 readl 688.8 ns biased 5.8 * 10^6 byte/s (-52% slower)
fast $07078000 readm 667.4 ns normal 6.0 * 10^6 byte/s (-55% slower)
fast $07078000 writew 658.8 ns normal 3.0 * 10^6 byte/s (-63% slower)
fast $07078000 writel 658.1 ns normal 6.1 * 10^6 byte/s (-62% slower)
fast $07078000 writem 657.1 ns normal 6.1 * 10^6 byte/s (-66% slower)
chip $00028000 readw 1178.8 ns normal 1.7 * 10^6 byte/s (-23% slower)
chip $00028000 readl 1172.3 ns normal 3.4 * 10^6 byte/s (-23% slower)
chip $00028000 readm 1119.9 ns normal 3.6 * 10^6 byte/s (-42% slower)
chip $00028000 writew 1149.6 ns normal 1.7 * 10^6 byte/s (-51% slower)
chip $00028000 writel 1145.8 ns normal 3.5 * 10^6 byte/s (-50% slower)
chip $00028000 writem 1190.9 ns normal 3.4 * 10^6 byte/s (-51% slower)
rom $00F80000 readw 615.3 ns normal 3.3 * 10^6 byte/s (-57% slower)
rom $00F80000 readl 613.6 ns normal 6.5 * 10^6 byte/s (-41% slower)
rom $00F80000 readm 670.0 ns normal 6.0 * 10^6 byte/s (-49% slower)

[patrik]

The ROM version should not affect the speed, however I did not put this in writing but I had run SetPatch before these tests which enables instruction cache burst and data cache without burst. If setpatch is not run before, it will only have instruction cache without burst enabled. When testing on my A3000 with 030@25MHz, enabling the caches via setpatch gives for example ~8MB/sec instead of ~6MB/sec for fast readw.

Please check how much difference before/after setpatch makes for you and that septatch enables the caches (check with cpu command)

One thing that is very odd with your results are that all .l and .m (32-bit read and writes) are more than halved compared to mine. Like it is doing 16-bit bus accesses when it should be doing 32-bit ones and it also affects the chip access. This is something the 030 is capable of and is signaled to it via /DSACK0, /DSACK1 (async 8-, 16- and 32-bit) and /STERM (sync 32-bit).

Buster version should not affect the speed of motherboard resources.

[hooverphonique]

Speed of the chips (logic/ram) should not affect bandwidth - if they are too slow (i.e. latency is high), you will get corrupted memory data, not slow access.

[Kin Hell]

Originally Posted by Screechstar

Hi,
It’s an interesting mix…
- Sysinfo CPU 25 MHz
- AIBB - CPU & FPU 25MHz
- WhichAmiga - CPU 25 MHz & FPU 24MHz
- SysSpeed - CPU 25MHz & FPU 230MHz!!!!!!!?

I’ve added a few screen shots… I think Sysinfo and AIBB looks setup properly?

I don’t have an oscilloscope to be able to probe properly alas and I am an enthusiastic novice so when things don’t work I can kind of work things out from schematics as to what broke… but not working 100%… eek! That’s a new challenge for me!
Ta!

SysInfo should be showing around 4.xx MIPS. No way is your A4K 25Mhz 030 & 68882 0.4 against an A3000. Iirc, it should be around 0.9xx the speed of an A3000.

Don't worry about the bench programs showing different parameters. - That's normal.

[Screechstar]

Thanks for all the comments there - this all helps a lot.

So yes, I checked and had the same cache & burst settings for the BusSpeed results I showed (instruction cache burst and data cache without burst).

I'm running Setpatch 44.38 with 68030.library version 46.6 and mmu.library in libs. But I do get the same low Sysinfo speed results on a fresh 'stock' WB3.0 also.

I did just test BusSpeed without any cache or burst... wow, that was REALLY slow! haha! The Rom and Fast ram were most affected in bandwidth and cycle time. The chip ram though was less affected overall - still slower but not as bad.

I will check continuity across the board for the /DSACK0, /DSACK1 and /STERM lines just to be sure nothing bad has happened to these... thanks for the tip with those

[Kin Hell]

Re-worked boards can always be a pain.

Why not get over to AmiBay & buy a Re-Amiga 4KD board? - Hese does ATX & even AGA A3000+ Re-Amiga. (My choice if I had loose cash would be A3K AGA)

[Boing-Ball]

80ns Chip and 70ns RAM

For a start the speed rating for the Chip = Not good! Ideally you should try 60NS.

Fast RAM 70NS EDO. Again try 60NS FPM. I’m guessing the system could be compensating for the lowest denomination of 80NS.. therefore possibly the slow speeds.

EDO will also mean 36-bit Data checking which can mean slow down compared with Fast Page Memory.

Won’t know until you try.

[pandy71]

Originally Posted by Boing-Ball

80ns Chip and 70ns RAM

For a start the speed rating for the Chip = Not good! Ideally you should try 60NS.

Fast RAM 70NS EDO. Again try 60NS FPM. I’m guessing the system could be compensating for the lowest denomination of 80NS.. therefore possibly the slow speeds.

EDO will also mean 36-bit Data checking which can mean slow down compared with Fast Page Memory.

Won’t know until you try.

Doubt on this, if there is no RAM errors then speed difference should not affect significantly performance especially that there is no autoconfig for RAM speed.

And EDO is extension to FPM - data are hold longer so timing conditions can be slightly relaxed.

I would rather check logic for cycle ending - perhaps something floating so R/W cycle is maxed to highest supported value.

Those kind of issues are quite annoying and most difficult to chase - system works but not as expected - or there is something trivial but nasty or there is some quirk...

Probably best would be to borrow and compare two same configurations... but nowadays this can be serious issue...

[Screechstar]

Hey all - thanks for the thoughts

I don't have 60ns but tried 8mb FPM 70ns fast and 2mb FPM 70ns chip.
Bustest results are a *bit quicker here and there but it didn't make a big difference. Good to check it though, just in case.

I did get a faster Sysinfo speed... twice. Seems a cold-boot can somtimes give me around 2500 dhrystones in Sysinfo. Still way slower than expected.

But if I dare press the SysInfo speed button again, it just gets slower and slower, until it settles around 2000 dhrystones. Not sure if that's a clue. It's not consistent through except that the more I use the Amiga the slower it gets until it settles around 1900-2000 Dhrystones.

I did check the /DSACK0, /DSACK1 and /STERM traces across the whole board and up into the CPU card... all are making good connection. Really good to check though just in case.

And yeah... a re-Amiga 4K would be amazing, but I'd love to get this one back up to speed if I can. It may be a long project as yes... these things are really, really anoying!

[Screechstar]

Soooo… was just playing around with SysSpeed again and noticed I don’t get any results for any ‘16’ tests… chip2chip16, fast2chip16, fast2fast16.

Should I? And if so, does anyone know what that could mean?
I found a thread which seems to include a an A4000 with an 060 and it does have one of these ‘16’ results shown in the image and all the other tested systems do too…
http://eab.abime.net/showthread.php?t=98465

Thanks.
I may be clutching at straws now

[pandy71]

Try to remove not necessary components - i would start from RAM modules... perhaps there is something active in background eating CPU and memory cycles - some nasty driver ?

[patrik]

Originally Posted by Screechstar

Hey all - thanks for the thoughts
I did get a faster Sysinfo speed... twice. Seems a cold-boot can somtimes give me around 2500 dhrystones in Sysinfo. Still way slower than expected.

But if I dare press the SysInfo speed button again, it just gets slower and slower, until it settles around 2000 dhrystones. Not sure if that's a clue. It's not consistent through except that the more I use the Amiga the slower it gets until it settles around 1900-2000 Dhrystones.

SysInfo is quite inconsistent on the 030. My experience does however not match what you are describing in the same way. In my experience, you get the same result for repeated speed tests during the same start of SysInfo, but if you exit SysInfo, perhaps start some other applications and then start SysInfo again, it will usually give a different result. Annoying, but I think it has to do with the alignment of some data or code used in the speed test which ends up less optimal for the 030 sometimes.

Perhaps this difference is a clue? Does your Amiga get slower after repeated tests in other speed tests like SysSpeed, AIBB or bustest?

Originally Posted by Screechstar

I did check the /DSACK0, /DSACK1 and /STERM traces across the whole board and up into the CPU card... all are making good connection. Really good to check though just in case.

Though of one more thing regarding this - are those signals connected to any pullup resistors or similar and if so, are those resistors giving the correct reading?

Originally Posted by Screechstar

Soooo… was just playing around with SysSpeed again and noticed I don’t get any results for any ‘16’ tests… chip2chip16, fast2chip16, fast2fast16.

Should I? And if so, does anyone know what that could mean?
I found a thread which seems to include a an A4000 with an 060 and it does have one of these ‘16’ results shown in the image and all the other tested systems do too…

That is most likely the MOVE16 instruction of the 040 and 060.