BLAZING. FAST. MAC. The Performer PDS 68030 Accelerator by MacEffects!

[Ron's Computer Videos]

I'm pretty impressed with MacEffect new accelerator for the Macintosh SE. Clocks in at roughly the same speed as a Macintosh II... and shows actual speed improvements in programs (not just in artificial benchmarks). Plus, it looks amazing in my new SE-Mini!

I made a short video showing off the card and doing some benchmarking.

Check it out:

MacEffects Performer PDS 68030 Accelerator for Apple Macintosh SE

MacEffects Performer accelerator is the perfect hardware upgrade for your Macintosh SE. The Performer accelerator features a 25mhz 68030 processor (clocked to 16mhz to match SE/30 performance) and 25Mhz 68882 FPU (math coprocessor) to speed up CPU performance and math calculations. With the...

maceffects.com

 

[Garrett]

On the original Performer card, the FPU will run at 16MHz unless a separate 25MHz oscillator is installed. Not sure if MacEffects is claiming the FPU is running faster than the 68030, but based on the photos on their site, it's going to be 16MHz. Either way, a great improvement to a stock SE!

 

[JDW]

I also have a Performer-SE from MacEffects and have been testing it on my SE, both with and without an optional Crystal Oscillator. Results are as follows, inline with the results from @Ron's Computer Videos :

JDW's TESTS​

JDW's Speedometer 3.23 Test (No FPU XTAL, stock 16MHz operation):

LEFT: 33MHz FPU XTAL | RIGHT: no XTAL (16MHz FPU)

(I saved the Machine Record on my Mac SE but later opened it on my modern Mac within Basilisk II, which is why the bars above are color.)


Norton System Info: With & Without 33MHz XTAL:

Daystar Power Demo app shows 16MHz for both CPU and FPU, even with 33MHz XTAL installed!

The XTAL I installed:

---

Ron's TESTS​

(extracted from his nice video and straightened)


Ron's Speedometer 3.23 Test (No FPU XTAL):

---

CONCLUSIONS:​

1. My stock 16MHz Performer-SE benchmark results are quite similar to Rons.
   
   
2. Oddly, my FPU benchmark results didn't increase much at all when I installed a 33MHz crystal. Remember, 33MHz is twice the base clock speed! Hmmm....
   And yes, I did indeed remove the solder blob at SJ1 before soldering on the crystal. And I confirmed that connection to be broken before I soldered in my XTAL. And I checked it again after my crystal was installed, as shown here:
   
   
   
   
3. We need more Performer-SE owners to post their benchmark results, especially those folks who have added a crystal to boost the FPU clock speed (it does not boost the CPU clock speed).

I ordered my 33MHz TCXO (Temperature Compensated Crystal Oscillator) from this AliExpress seller. I told them I wanted 33MHz instead of 25MHz, and they made the change for a few bucks extra.

 

[YMK]

This is a very neat product! Quite cost effective as well.

@JDW, your FPU benchmarks are in line with what I've seen, doubling FPU speed on an SE/30:

https://tinkerdifferent.com/threads/se-30-fpu-upgrade.3022/

 

[JDW]

@JDW, your FPU benchmarks are in line with what I've seen, doubling FPU speed on an SE/30…

Very interesting. Thank you for confirming that I am not losing my mind. I would have thought a doubling of clock speed would result in a far bigger boost to FPU benchmark scores, but your own tests show that to not be the case.

 

[JDW]

NEW BENCHMARKS​

I ran Speedometer 3.23 on my stock SE/30 just now, since my Machine Record file didn't have SE/30 scores.

Stock SE/30:​

Stock SE/30 (left) versus Performer-SE (no XTAL added, CPU=FPU=16MHz):​

Stock SE/30 (left) versus Performer-SE (with 33MHz FPU XTAL):​

CONCLUSIONS​

1. Performer-SE's CPU Score isn't as fast as a stock SE/30, but still 3.3x faster than a Mac Classic (or SE). So nothing too unusual about that. It's a nice boost.
2. Graphics score is a major difference, with the Performer being only 50% of a stock SE. But why? PDS slot a bottle neck? This should be investigated.
3. Disk Score is slower than a stock SE/30, but not substantially so.
4. Even with a 33MHz FPU XTAL installed, the Performer's Math Score is still slower overall than a stock SE/30. Also note how many of the Math Scores such as Dhry, Towers, etc. are exactly the same for the 16MHz FPU Performer as they are with the 33MHz FPU Performer, which makes no sense at all! Again, the reason for this should be investigated.
5. KWhet Score blows away the stock SE/30, with or without the 33MHz XTAL added.
6. Other math tests like Dhry and Towers are much lower than a stock SE/30.
7. All 4 FPU tests at the bottom show the Performer's FPU to be slower than the SE/30.

​

So how does the Performer-SE "feel" versus a stock SE/30?​

Well, that's where the Graphics performance comes into play, I think. I base the "feel" on how I work in the Finder, doing things like opening Windows and clicking on pull-down menus. The time it takes for a menu to drop down when clicked is faster on the SE/30. No need to time it, I just feel the difference. Ditto for opening windows by double-click. It's faster on a stock SE/30. CPU performance plays a role in this too.

Does the Performer-SE feel faster than a stock 8MHz Mac SE?​

Yes, but graphics performance probably comes into play there too because if you check my earlier post, we see the Speedometer 3.32 score for Graphics is 1.88 on the Performer and 1.00 on the SE. Plus the CPU is 3.32x faster than a stock SE too. So overall, yes, you will feel the difference if you use a Performer versus a stock SE. It just doesn't feel as fast as a stock SE/30, and the benchmarks also explain why that is. So you should buy the Performer-SE with that in mind.

DISCLAIMER: This is my objective look at the Performer. It doesn't seek to excessively praise it or tear it down.

Now it's your turn!​

If anyone else out there has a Performer-SE, please share you Speedometer 3.23 scores and experiences, with or without an FPU Crystal Oscillator added.

 

[phipli]

Even with a 33MHz FPU XTAL installed, the Performer's Math Score is still slower overall than a stock SE/30. Also note how many of the Math Scores such as Dhry, Towers, etc. are exactly the same for the 16MHz FPU Performer as they are with the 33MHz FPU Performer, which makes no sense at all! Again, the reason for this should be investigated.

There is something strange going on here certainly. Normally, in part because of the asynchronous operation, FPU performance scales fairly linearly with clock speed.

So, these cards are electrically identical to the Total Systems Mercury - I have one, but not set up. I could compare your performance.

@JDW What driver software are you using? Can I recommend installing GEMStart and seeing what difference that makes to your FPU scores? Find it...

here : https://macintoshgarden.org/apps/accelerator-card-drivers

Install version 3 for System 7

 

[phipli]

Further to my previous post, I had a look in my phone for photos of old benchmarks I've run. In the following "Upgraded Macintosh SE" is an SE with the Total Systems Mercury fitted - as mentioned, this is electrically identical to the card your card is a clone of. Different brands, same card.

I can't remember for absolute certain if this (Upgraded Macintosh SE) Norton FPU benchmark score is from when I was running the FPU at 16MHz, or 20MHz, but I suspect it is 16MHz, it is very close to the performance of an SE/30. The reason I suspect it is 16MHz is because from neighboring photos, I know for a fact that "Current System" 14.7, was another, different accelerator with a 20MHz FPU. For interest, note that (14.7/20MHz)*16MHz = 11.76.

I don't have this on hand at the moment - this is a 2 year old photo.

Like I said, try GemStart and see where that gets you!

Also, yes, my desktop is covered in all the different driver versions I was testing at the time

Edit - For interest, here is my card (after I fitted a 20MHz clock) :

 

[YMK]

Also note how many of the Math Scores such as Dhry, Towers, etc. are exactly the same for the 16MHz FPU Performer as they are with the 33MHz FPU Performer, which makes no sense at all! Again, the reason for this should be investigated.

These are integer tests, so they don't involve the FPU.

Graphics score is a major difference, with the Performer being only 50% of a stock SE. But why? PDS slot a bottle neck?

This probably has to do with the BBU's interleaved CPU/video sharing of DRAM that was designed around 68000 timing.

Even with a 33MHz FPU XTAL installed, the Performer's Math Score is still slower overall than a stock SE/30.

The reason here is probably also memory bandwidth.

 

[phipli]

This probably has to do with the BBU's interleaved CPU/video sharing of DRAM that was designed around 68000 timing.

Graphics score is a major difference, with the Performer being only 50% of a stock SE. But why? PDS slot a bottle neck? This should be investigated.

I didn't see this in my first read through. This doesn't match my experience with my version of the card. For Norton's quickdraw benchmarks, the accelerated SE scored... 62% faster than a stock SE.

Again, I'd suggest installing GemStart and re-running benchmarks.

Edit - Here is a summary of the scores for my card.

	Stock SE	16MHz Accelerated SE	SE/30
Overall	4.37	13.4	28.7
CPU	5.11	11.9	26.2
Video	15.3	24.8	44.5
Disk	42.4	49.3	85.3
FPU	0.0813	10.9	11.1

 

[phipli]

The reason here is probably also memory bandwidth.

The FPU runs asynchronously though? Plus the many clock cycles it takes to do floating point, even with an FPU, means that memory shouldn't be a limit within reason. Also, this behavior doesn't match my experience with basically the same card, so it is either a clone issue, or a setup issue I suspect.

 

[JDW]

@phipli
I installed the official Control Panel and Extension shown on the MacEffects Performer page.

I will give your recommended drivers a try later today and report back with Speedometer 3.23 benchmark results at that time.

@YMK
There is indeed a huge memory bandwidth difference, and I forgot about the impact the BBU might have. Thanks also for the tip about Integer tests which don’t use the FPU.

Mark Jozaitis told me that clocking the “CPU” (not FPU) higher requires a reprogramming of the GAL chips, and that is beyond his technical capability. But of course, that would result in a bigger performance boost to the card than overlocking the FPU alone. (FPU overclocking can be done without touching the GALs, apparently.

 

[phipli]

Mark Jozaitis told me that clocking the “CPU” (not FPU) higher requires a reprogramming of the GAL chips, and that is beyond his technical capability. But of course, that would result in a bigger performance boost to the card than overlocking the FPU alone.

More than that, it would require a redesign of the card - this is a fairly simple card and its timings aren't disconnected from the SE's primary bus speed. I think it uses the motherboard 15.67MHz clock signal.

FPU overclocking can be done without touching the GALs, apparently.

The 68882 runs completely asynchronously and happily works with whatever clock you feed it (within overclocking limits - but they overclock quite well). For example, if you lift the clock pin on the 68882 on a IIci and pipe 33MHz into it, you'll see an improvement in floating point performance on a IIci, no other mods required.

 

[phipli]

@JDW...

Just a thought... but I've worked out part of what is going on.

You removed the blob of solder?

So if you break the connection between pin 8 (Output) on the clock, and the FPU itself... how is the FPU clock meant to see the signal from your 33MHz clock?

I think there has been a misunderstanding here and people have been given instructions that disconnect the socket, instead of disconnect the other clock. Is there another jumper somewhere for disconnecting the onboard 16MHz clock?

Where did you get those renders from? Can you share them? I might be able to say where you need to cut with a knife (or if there is another place to disconnect the signal path) to isolate the FPU from the 16MHz built in clock, because it looks like... perhaps... the jumper was put on the wrong signal.

 

[JDW]

You removed the blob of solder?

Yes, as per the instructions.

Similar photo showing solder blob appears on @Bolle 's Github page.
On that same Github page, the following text appears (I added clarity in brackets):

Close [with solder blob] SJ1 to run the optional 68882 off the 16MHz accelerator clock. Install an oscillator and open SJ1 [remove solder blob] to run the 68882 at any speed. The original design used 25MHz which has been verified to work on the clone as well.

All PCB files and renders at shown at Bolle's Github. I simply dropped the files into the Gerber viewer at JLCPCB to see the board layout, and that was what I used to create the screenshots in my earlier post.

I cannot explain anything technical about the card beyond what I've already stated because Micromac designed the card and Bolle recreated it. Joe at Joe's Computer Museum played some role as well based on the silkscreen text:

As to the proper orientation of my TCXO XTAL, it is correct. Pin-1 corresponds to the sharper edge part of my XTAL:

 

[phipli]

Yes, as per the instructions.

View attachment 18847

Similar photo showing solder blob appears on @Bolle 's Github page.
On that same Github page, the following text appears (I added clarity in brackets):



All PCB files and renders at shown at Bolle's Github. I simply dropped the files into the Gerber viewer at JLCPCB to see the board layout, and that was what I used to create the screenshots in my earlier post.

I cannot explain anything technical about the card beyond what I've already stated because Micromac designed the card and Bolle recreated it. Joe at Joe's Computer Museum played some role as well based on the silkscreen text:

View attachment 18848


As to the proper orientation of my TCXO XTAL, it is correct. Pin-1 corresponds to the sharper edge part of my XTAL:

View attachment 18849

@JDW You misunderstand me - I'm saying those instructions are incorrect, or there is an error in the PCB design.

By following those instructions you DISCONNECT the socket from the FPU.

But... I don't just mean put it back, because then you would have two clocks at once. I was asking about where you got the diagrams so I could see both sides of the board and follow traces.

 

[JDW]

I recommend installing GEMStart and seeing what difference that makes to your FPU scores? Find it...

here : https://macintoshgarden.org/apps/accelerator-card-drivers

Install version 3 for System 7

More specifically, you seem to be referring to DL#14...

And it seems I need to expand the file named "gemstart3.sit":

Very interesting...

It uses the same icon as the v2.2.1 Control Panel I am using now...

I must assume that means we must use the file named "gemstart.sit" with System 6?

 

[phipli]

More specifically, you seem to be referring to DL#14...
View attachment 18850

And it seems I need to expand the file named "gemstart3.sit":
View attachment 18851

I must assume that means we must use the file named "gemstart.sit" with System 6?

I think they both work in System 6

You get to watch the silly startup animation.

But the main issue here is that you never actually clock the FPU faster. Stick a frequency meter on the FPU clock pin if you can - you'll find it is always just under 16MHz.

Not sure why your graphics were slow though.

 

[JDW]

But the main issue here is that you never actually clock the FPU faster. Stick a frequency meter on the FPU clock pin if you can - you'll find it is always just under 16MHz.

I have very limited time today, time that will be used only to test your recommended control panel. Tomorrow will be very busy for me as well, but if I can check the frequency tomorrow, I will try. Otherwise, that must wait for Monday.

Even so, what you are suggesting is that the documentation on Github is wrong. All I did was follow those instructions very strictly, while consulting with @MacEffects .

 

[phipli]

I have very limited time today, time that will be used only to test your recommended control panel. Tomorrow will be very busy for me as well, but if I can check the frequency tomorrow, I will try. Otherwise, that must wait for Monday.

Even so, what you are suggesting is that the documentation on Github is wrong. All I did was follow those instructions very strictly, while consulting with @MacEffects .

We all make mistakes, even on github - look at the trace you're breaking - it leaves the output pin of the clock and heads towards the FPU.

Combine that with the fact that the weird behavior you're seeing would be exactly explained by this - the speed doesn't change when you add a fast clock... almost like it isn't connected?

Here is a good test - remove your clock, just unplug it from the socket. If the FPU still runs, it is connected to the 16MHz, not the socket, which it shouldn't be if you're trying to use the socket. If things are wired correctly, the FPU will not work with the clock removed.

That's a nice, definitive and quick and easy test