Sonnet 7200 PCI programming

[grelbfarlk]

Originally Posted by Wrangler

Thanks, that's good info but I'm still mystified why the actual voltage on 5v is so critical to the PMC card operating

Well like we've seen with other issues of instability, running a 12V fan on 5V causes noise. What the issue is here I can't rightly say. To throw theories out there, maybe the BFG9060 is causing noise. I take it the Ragnarok runs fine when the BFG9060 is clocked at 50MHz?

Maybe I have a situation I can reproduce here. I put the Ragnarok in my test machine with a newer ATX power supply and I'm measuring 4.98V on a drive connector and 4.86 or thereabouts on the cap next to the RAM on the motherboard. And I can't seem to get the Ragnarok working correctly. Previously it had the 550MHz Rapture which was pretty much working alright as I recall, I haven't played with this machine in a while.

*EDIT*
Strange behavior, I try the PMC239/F with a Rapture and it acts mostly the same, becoming crash city in less than a minute. Trying some of the other PMC adapters I have around I put the 600Mhz Rapture in which eh, I didn't have cooling set up and it immediately is working at full load until it overheats, the good news is I lost some more fingerprints. Though Ragnarok is just as bad on the other busboards.

I take it everyone is using an MK-IV busboard with a 3.3V regulator on the busboard?

In supporting your point I threw all kinds of configurations at this system and eventually plugged in some 12V fans running at 5V and it started to work fine.

[Stedy]

Originally Posted by grelbfarlk

Well like we've seen with other issues of instability, running a 12V fan on 5V causes noise. What the issue is here I can't rightly say. To throw theories out there, maybe the BFG9060 is causing noise. I take it the Ragnarok runs fine when the BFG9060 is clocked at 50MHz?

Maybe I have a situation I can reproduce here. I put the Ragnarok in my test machine with a newer ATX power supply and I'm measuring 4.98V on a drive connector and 4.86 or thereabouts on the cap next to the RAM on the motherboard. And I can't seem to get the Ragnarok working correctly. Previously it had the 550MHz Rapture which was pretty much working alright as I recall, I haven't played with this machine in a while.

*EDIT*
Strange behavior, I try the PMC239/F with a Rapture and it acts mostly the same, becoming crash city in less than a minute. Trying some of the other PMC adapters I have around I put the 600Mhz Rapture in which eh, I didn't have cooling set up and it immediately is working at full load until it overheats, the good news is I lost some more fingerprints. Though Ragnarok is just as bad on the other busboards.

I take it everyone is using an MK-IV busboard with a 3.3V regulator on the busboard?

In supporting your point I threw all kinds of configurations at this system and eventually plugged in some 12V fans running at 5V and it started to work fine.

Hey, try something for me. Can you get a 3.3 ohm, 10W resistor and connect it across the 3.3V ATX supply output?

I have seen oddities when there is no load on the +3.3V supply, which have in turn affected the +5V output.

You're not using the +3.3V from the ATX supply, instead relying on the busboard +3.3V converter.

Your PMC cards appear to be 5V tolerant and you're using a universal PCI to PMC adaptor, that should not cause issues.

[grelbfarlk]

It's weird I was actually thinking about putting a cap on the 3.3V VRM on the Mediator today.
But mostly we aren't using the 3.3V on the busboard, the PMC239/F does its own conversion. From what I can tell the 3.3V from the slot isn't even connected to the voltage regulation on the adapter.
Then if we are using Voodoo, that's 5V, or if people are using Radeons, they probably got 5V tolerated Radeons that do their own conversion.
So that's why I was asking if people are mostly using MK-IV that included the 3.3V busboard.

[grelbfarlk]

I couldn't find a suitable resistor, so I just removed the VRM. On quick testing no real difference to be seen.

[grelbfarlk]

So here's what I gathered, which is pretty much useless.
With Radeon+Ragnarok, Radeon must always come first in the slots otherwise Ragnarok will suck up all of the PCI address space and throw a RAMLIB error as soon as the pci.library is started.

With it in this configuration:
4) 10mb NIC
3) Ragnarok
2) Radeon
1) SB128 (CT5808)
System will boot but takes a long warmup time before it is stable.

It starts on first try and runs just fine in this orientation:
4) Ragnarok
3) (haven't added the NIC back yet but in another system 100MB NIC goes here)
2) Radeon
1) SB128

[trixster]

Interesting. My configuration is:

4) SB128
3) Radeon
2) Ragnarok
1) NIC
0) Mediator logic board

Although Deneb means swap Config is jumpered, perhaps that makes a difference? (I thought that just meant pci cards get configured first?)

[Stedy]

Originally Posted by grelbfarlk

So here's what I gathered, which is pretty much useless.
With Radeon+Ragnarok, Radeon must always come first in the slots otherwise Ragnarok will suck up all of the PCI address space and throw a RAMLIB error as soon as the pci.library is started.

With it in this configuration:
4) 10mb NIC
3) Ragnarok
2) Radeon
1) SB128 (CT5808)
System will boot but takes a long warmup time before it is stable.

It starts on first try and runs just fine in this orientation:
4) Ragnarok
3) (haven't added the NIC back yet but in another system 100MB NIC goes here)
2) Radeon
1) SB128

What does PCIinfo and MUscan give you for both combinations?

Something cacheable, that shouldn't or overlapping Base Address Register (BAR) can cause weird crap to happen. Would be interesting to see how the interrupts get mapped to (pciinfo) output.

[starship]

Does ragnarok works on last build with 100mb nic?

[grelbfarlk]

Originally Posted by trixster

Interesting. My configuration is:

4) SB128
3) Radeon
2) Ragnarok
1) NIC
0) Mediator logic board

Although Deneb means swap Config is jumpered, perhaps that makes a difference? (I thought that just meant pci cards get configured first?)

I tried this exact arrangement, swap is jumpered also because of the A4091 and RAMLIB still blowing up as soon as pci.lib is called, via pciinfo.

[trixster]

Very odd!

[DarrenHD]

Has anyone got a BFG9060? I just installed one in my A4000 but I am struggling to get the PPC to work reliably, either at 50Mhz or 100Mhz (on the BFG).

I can get it to work more consistently at 50Mhz on the BFG, but when I run getinfo I get a PowerPC exception window pop up, although getinfo does run and then I can run PPC programs. I've run the mmu rexx script to generate a MMU-configuration file and I merged that with the recommended mmu config file for the BFG.

At 100Mhz on the BFG its less likely to work at all - ramlib hangs when trying to start the PPC.

[grelbfarlk]

Originally Posted by DarrenHD

Has anyone got a BFG9060? I just installed one in my A4000 but I am struggling to get the PPC to work reliably, either at 50Mhz or 100Mhz (on the BFG).

I can get it to work more consistently at 50Mhz on the BFG, but when I run getinfo I get a PowerPC exception window pop up, although getinfo does run and then I can run PPC programs. I've run the mmu rexx script to generate a MMU-configuration file and I merged that with the recommended mmu config file for the BFG.

At 100Mhz on the BFG its less likely to work at all - ramlib hangs when trying to start the PPC.

Yeah that's kind of what we're talking about here-sort of?

Originally Posted by Wrangler

Trixster suggested I share my findings on voltages because they're quite unexpected but may help other users. I'm also looking for some suggestions on fixes!

This post relates to the Apocalypse and Rapture cards (maybe Ragnarok too - I haven't tested) ie PPC cards that use the Ramix carrier card.

TL;DR version: the voltage on the 5v power line to the Ramix being too high makes the cards unreliable/non-functional. This is the opposite to what I thought - I had thought that they needed a nice clean 5v on the 5v line but that is not the case. If your PPC is unreliable you might want to try removing any power injector to the Zorro slots if you have one. Or even stick some extra Zorro/PCI cards in to soak up some current!

Background

(Note: all voltage measurements are from a CPLDICY card that is either measuring 5v at the sensor on the card - ie the Zorro bus voltage - or direct from the 060 CPU card itself so are about as accurate as you can get)

I was trying out the BFG9060 in combination with the PPC. Obvs I wanted the '060 at 100MHz. To do that, you need at least about 4.8v to the CPU card whereas the card was getting about 4.72v, so I upped the voltage. That gets the '060 nice and stable. But the PPC wouldn't work at all.

After a lot of faffing about with hardware and software , I discovered that the PPC won't operate properly if its voltage is above about 4.7v. Below that (within reason), it works absolutely fine and reliably. Above that, it will either hang quickly or not even be found by InitPPC.

The result is for my combo I need high 060 volts and low PPC volts. Basically, there is no overlap in those two on my set up - it's one or the other.

Questions
1. I thought the Apocalypse and Rapture cards only ran off 3.3v as supplied and regulated by the Ramix. So they shouldn't care whether the 5v input is 4.7v or 4.9v or whatever. Right? Clearly not. Can anyone identify where 5v is actually used by the Ramix/PPC card pair?

2. Any creative ideas on how to try to create that sweetspot of high 060 volts, low PPC volts? I imagine it would mean hacking the Ramix somehow to drop 5v to 4.7v but PCI has many 5v lines so that's a lot of hacking. Any other ideas?

I wonder if anyone is using an A4000T or an A4000 in a tower kit?

Also ask I asked before report your system config, are you all using MK-III? I'm using an MK-IV in this system.
A4091 is in the top Zorro slot, maybe this is also a contributing factor.
What slot is the Deneb in?

[DarrenHD]

Originally Posted by grelbfarlk

Yeah that's kind of what we're talking about here-sort of?




I wonder if anyone is using an A4000T or an A4000 in a tower kit?

Well I was having stability issues with my system (which was using the AmigaKit ATX adapter). I was only getting about 4.3V measured near the motherboard simm sockets. It was causing graphical glitches and lockups. I changed back to the stock A4000D Power supply now I get 4.9V and my machine is stable now. However with the BFG now installed, the PPC somewhat works at 50Mhz on the BFG, but I still get PowerPC exception crashes, and sometimes the PPC works sometimes it doesn't. If I put the BFG at 100MHz, InitPPC will just spike up ramlib to 100% for a few seconds then it will go back down and hang.

I may end up putting my CS MK2 (@100Mhz) back in, as it worked with that. And put the BFG into another machine.

Update: Put the CS MK2 @100MHz back in and the PPC works fine.

Darren

[BULI]

I done quick upgrade on this list:
https://github.com/Sakura-IT/SonnetA...brary-version)

[Apollon]

is it possible to make the pci express killer nic cards compatible somehow? or technically not possible? i have a MK4 Mediator and it says something about PCI to PCIe adapters.

[grelbfarlk]

Maybe some troubleshooting is needed, as I understand it the PCIe Killer NIC card's RAM doesn't show up as PCI space. Possibly just a configuration issue with the card.

Something I was looking to recently, I was digging up the Marvel PCI bridge docs and think I found some which might be relevant. I'm not sure if I'm incepting the idea we did this already and determined that even with the docs it wouldn't work anyway? Or maybe this is just the first iteration of the temporal loop?

Or maybe we should just crowdfund a new card. Something with an E6500 core and RAM. In PCI format.

[Apollon]

would be interesting to find out if the pci e cards run somehow, they are cheaper than the 64 mb pci cards and easier to find. there would be nothing wrong with a PowerPC e6500 card, but that much power would have to get os 4.1 support in addition to classic support, i think.

Edit
where do you get the PowerPC e6500 CPUs and what would such a card cost?

[grelbfarlk]

Originally Posted by Apollon

would be interesting to find out if the pci e cards run somehow, they are cheaper than the 64 mb pci cards and easier to find. there would be nothing wrong with a PowerPC e6500 card, but that much power would have to get os 4.1 support in addition to classic support, i think.

Edit
where do you get the PowerPC e6500 CPUs and what would such a card cost?

Digikey has plenty in stock. Depends on how much gofast people want. You can get e6500s for $200, maybe less, or something in the $90 range for an E500mc at above 1GHz. Let's say RAM costs $30 or so for 512MB. Board and associated logic for another $40.

Then get someone to design, build, debug and test it, $50,000. (Unless you find someone who is bored).



I could see it being possible to sell them for $500.

[Stedy]

Hi,

Where is the best place to provide comments on the code, here or https://github.com/Sakura-IT/SonnetAmiga?

I've had a look at the TSI107 code and have a few comments to make, looks like 2 errata are not patched (6 & 23).

[grelbfarlk]

I think Hedeon is not around much any more so I'd say do it on the git, fork I guess.