Daystar Universal PowerCache P33 in SE/30

JDW said:

Please post a nice photo of the FRONT and BACK of that PDS slot adapter.  I've not seen one before.

Click to expand...

Oh its nothing special, its the IIsi Daystar Adaptor.  I can still take pictures if you'd like!

JDW said:

You also are forgetting the need to benchmark ENHANCED PowerMath, which I mention in the following thread (which I will benchmark for you later):

Click to expand...

I have tried that too, it does not affect the Processor benchmark in Speedometer.  Thank you for benchmarking in advance!

BTW, I tried the Command-Option clicking like you mentioned in the other thread, and it has no effect on my benchmarks!  So, while you say it degrades your performance, it makes no changes to mine, which leads me to believe that the cache portion of the PowerCache is not active, instead of the cache being on and the toggle being broken. :-(  

Speedometer 3 test variations are these:

• PowerCache On, PowerMath On
  
• PowerCache Off, PowerMath On
  
• PowerCache On, PowerMath Off
  
• PowerCache Off, PowerMath Off (same as if PowerCache control panel not loaded at boot time)
  
• PowerCache On, Enhanced PowerMath On
  
• PowerCache Off, Enhanced PowerMath On
  

Hold down Option & CMD keys while clicking on the PowerMatch switch in the control panel to enable Enhanced PowerMath, which will then be shown by "PowerMath" being underlined.

With PowerCache OFF I get about 9.5 in Speedometer, and with PowerCache ON I get 13.7.  And yes, I do see a boost in FPU performance for some operations with Enhanced PowerMath ON versus regular PowerMath ON.

I am still building Speedometer 3 Machine Records for you, which I will post in SIT format for you in a few days (sorry, I'm quite busy the next few days).  My disk tests will be slow because I booted off my FloppyEMU.  I even got a "FATAL ERROR" display on the FloppyEMU just now during disk testing for some unknown reason.  Anyway, I used my Micron Xceed Grayscale adapter to complete all the video tests.  In monochrome mode though, the Color30HR is only as fast as stock graphics.

I look forwarding to seeing your photos.

JDW said:

 

I am still building Speedometer 3 Machine Records for you, which I will post in SIT format for you in a few days (sorry, I'm quite busy the next few days). 

Click to expand...

Appreciate it!  Hoping to hear from others as well.  

Below are some photos:

Thank you for the photos.  Do you have the original documentation for it?  I'm curious if Daystar officially says it will work properly in the SE/30, or if it is intended only for the IIci.

OK.  Below are my Speedometer 3 benchmark files (one file is v1.5.1 .sit and the other is the same .sit file but subsequently Binary encoded), with each benchmark saved in its own file.  When you view the filenames, please note:

PC = PowerCache

PM = PowerMath

EPM = Enhanced PowerMath

View attachment Speedometer3_JDW.sit

View attachment Speedometer3_JDW.sit.bin

Let me know if the SIT file is usable as-is.  Not sure if Binary encoded is needed when uploaded to this forum.

JDW said:

 Do you have the original documentation for it?  I'm curious if Daystar officially says it will work properly in the SE/30, or if it is intended only for the IIci.

Click to expand...

Nope I do not I'm afraid.

JDW said:

Let me know if the SIT file is usable as-is.  Not sure if Binary encoded is needed when uploaded to this forum.

Click to expand...

Great I will try this out once I get home, thanks!

Thanks @JDW!  I ran some new benches, and this is what I see.  It really does seem my setup is really under performing yours!

Now I am wondering is it just me or other Universal PowerCache + TwinSpark users?

Mine-Universal-PowerCache-On-vs-JDW-Socketed-PowerCache-Off



Mine-Universal-PowerCache-On-vs-JDW-Socketed-PowerCache-On



View attachment Mine-PC-On-vs-JDW-PC-Off

8.83 isn't too much under 9.4, but it's clear that because you get the same exact scores regardless of the Control Panel setting that something is wrong in the hardware, with that card and adapter combo not being compatible with the SE/30.  Being "not compatible" doesn't mean it won't work, but it's clear the performance gain isn't there, although you would get a gain over the stock SE/30 configuration.

The problem I am having is with lockups.  While running those benchmarks for you, I actually had more than one lockup.  It seems to occur as the system heats up, and this socketed accelerator really heats up, let me tell you.  (It can't be the analog board or PSU or motherboard capacitors, because all 3 are recapped.)

Have you contacted Manabu Sakai of ARTMIX to ask about compatibility with his TS Adapter and your card?  If not, please ask him.  His English isn't perfect, but he can surely understand that question.  Just be sure to state which TS adapter version you have so he isn't confused about that.  Use simple English sentences and you should get a reasonably good response from him.

JDW said:

8.83 isn't too much under 9.4

Click to expand...

Right, so it’s as if the PowerCache setting has no effect and is off in my setup no matter what.  It is way under your 13.5 which is what I would expect!

JDW said:

Have you contacted Manabu Sakai of ARTMIX to ask about compatibility with his TS Adapter and your card?

Click to expand...

I plan to do so now with these benchmarks in hand, I wanted to rule out that I had wrong or misleading benchmarks.  I am sure I have loaded the Control Panel right through my PowerMath flipping, and also tested OS 7.5 through a boot floppy, so I am reasonably confident there is a gap.  

I still do do wonder what other folks with a Universal PowerCache + Adaptor are getting.  I also wonder if @trag benched his cacheless 50mhz Accelerator he won a few weeks back  

I have a theory... The Artmix adapter is a clone of the adapter that was made for the DiiMo Cache 030.

It looks like that one handles the /STERM signal different than the PowerCache. I will modify one of my adapters and see if the performance rating changes if I run /STERM through the GAL just like on the Artmix adapter or wire it up straight through between the PDS and cache slot on the adapter like Daystar did on the IIsi adapter.

Not sure how they did on their official SE/30 PowerCache adapter as I do not have one of those.

If the performance rating changes you can run a simple wire (or change the code in the GAL)

On the SE/30 DiiMo they are doing this in one of the GALs and I am pretty certain the same is going on in the Artmix/DiiMo SE/30 adapter as well. (at least I did this in my SE/30 adapter and I am getting the same results so I again I am sure this is what Artmix and I am doing)



If I get this right this will align /STERM to the 16MHz clock on the PDS interface and then delay it for a few nanoseconds (depending on the propagation delay of the GAL that's used) and finally output that signal to the accelerator.

I can imagine this might throw off the bus/cache logic on the PowerCache if it wasn't designed to wait for that delay.

In the end the SE/30 could be something totally special and require even more magic on the adapter to use the Universal PowerCache to its full potential.

EDIT: changing /STERM to straight through wiring like on the IIsi adapter doesn't do anything to the performance rating. :/

That's really weird... I will see how the Speedometer rating behaves when using the P33 PowerCache in a IIsi, II and LC (and I figured it's going to be ok in the IIci/IIvx as those have the correct slots right away)

Both of my PowerCaches are behaving the same in the SE/30, so it's not an issue with them. Someone with the original Daystar adapter would be helpful here now.

Ok, so on all other Macs the cache switch in the control panel does something and the changes get reflected in the Speedometer benchmark as well.

Cache works fine in the Mac II using the original Daystar adapter.

Cache works fine in the Mac LC and CC using the original Daystar adapter.

Cache works fine in the IIsi using one of my Topper adapters in Daystar configuration (which matches the original Daystar IIsi adapter)

Cache does not work however in the SE/30 using neither the DiiMo/Artmix nor the Daystar IIsi configuration of my adapter.

The only difference between the IIsi adapter and the original SE/30 adapter is a 74ACT86 that does some kind of clock buffering... at least the C16M and CPUCLK line from the PDS get jiggled around that IC.

I am not sure if our issue here is is related to the 16MHz clock signal going to the PowerCache though.

On the PowerCache itself the CPU clock signal from the host PDS is only used in one single location @GAL U14 so I don't see exactly why buffering should be necessary.  :huh:

Bolle said:

Cache does not work however in the SE/30 using neither the DiiMo/Artmix nor the Daystar IIsi configuration of my adapter.

Click to expand...

Wow @Bolle thanks for looking into it and reproducing!  Kinda glad I am not the only one! ;-)  

Hardware is over my head, but if its switchable in software then something must sense an address/instruction/data to "flip a bit".  It also seems that PowerDemo is able to "read/sense that bit" - does that stay correct in the Macs where the cache is enabled?

I mean, it could also be the Control Panel setting different addresses expecting a different adaptor in an SE/30 - maybe because of Slot $E being used for Video unlike the other Machines?

A quote from a certain "jt" gentlemen from a thread on thinkclassic.org.  I don't claim to understand much of it.

I think I may have figured out the source of confusion. In GttMFH2e, pp.389-392 list the signals available on the IIci cache slot. The bolded entries are signals Apple placed on the connector for diagnostics purposes only and would be the jumping off point for those "determined to design an expansion card other than a cache memory card," such as the PowerCache accelerators.

My theory is that something about Daystar's implementation using machine level interrupts /IPL0-IPL2 or some other "extraneous" signals available for operations "other than cache" puts a hurt on the function of cards at location $E that are not designed to operate within the framework of the Slot Manager operations. The video subsystem of the SE/30 may appear to occupy $E to Sot Manager, but there was no reason for Apple to operate within its own Slot Manager design guidelines for a "card" in a slot location in the SE/30 that's unimplemented in hardware. 

My theory is that operations of the PowerCache accelerator interfere only with those operations of slot $E that fall outside standard Slot Manager Design guidelines. Slot $E would therefore be fully functional in six slot Macs and the LCIII PDS that fall within those standard Slot Manager operations . . .

...but well and truly bork the SE/30's video subsystem, hence, the CPLD adaptation?

Click to expand...

Also, there is an adaptor on eBay right now with some decent pictures, it gives a good view of the 74ACT86 and you can _almost_ make out traces in one of the layers.  I'll try to clean it up and put it up here and hopefully that can help.

Ok... the one place the CPU CLK from the SE/30 PDS is going on the PowerCache is to GAL U14 as I said before.

That GAL seems to do some kind of clock distribution for that clock signal. It also connects to the "unused" pins on the cache connector.

It looks like they use those pins to determine the kind of adapter/machine the PC is plugged into and adjust the clocks accordingly (i.e. shift them in phase a little bit using the internal feedbacks of the GAL in some "unused pin" configurations)

There is an external feedback network hooked up as well using two caps but those two caps are not soldered into place on the PowerCache.

This is what U14 looks like in the schematics:







Not sure if this leads us somewhere but it's the only thing I could imagine that keeps the cache portion from working correctly in the SE/30 because why would they have used the 74ACT86 on the official adapter to do weird stuff to the 16MHz clock if the clock wasn't important.

The "unused" pins (lets call them machine sense pins) could be connected to ground/vcc in a different manner on the SE/30 adapter as well.

Ok, way easier than all that:

Connect /CENABLE (pin C13 on the PowerCache slot on the TwinSpark) to ground and the cache will work in the SE/30.

Spotted that little difference from the IIsi adapter while looking at pics of the SE/30 adapter.

Not sure if there is a jumper or something like that on the TwinSpark that will let you do this conveniently.

I just soldered in a patch wire on one of my adapters and I am now getting a score around 11.5 with my 40MHz P33.

I also noticed that the score gets changed a little bit when changing configuration of the "machine sense pins"

The external cache setting in the PowerDemo now also reflects the setting in the control panel and turning it on and off has an effect on how fast the shiny stars are redrawn :tongue:

Wow! That sounds super! I want to try that with my TwinSpark! Do you have a picture?

There is a jumper! Two jumpers, and three slots as a matter of fact.

Can you buzz out what those jumpers are connected to?

I’ll give it my best shot. Hang on.

I might as well send some more pictures, and I’m not used to this.