Total Systems Gemini II Accelerator for the SE

[Ubik]

I love making the lowly SE faster, since that was my first computer back in the late 80s and early 90's, and accelerator boards were an economical upgrade path compared to the $4,000+ dollars for a loaded SE/30 from the college bookstore.

So I am always on the lookout for cool SE accelerators. I was blessed to become the new steward of a Total Systems Gemini II accelerator (engineered by Quesse) purchased from @nekogahora280 after bricking my previous 16Mhz Gemini 020/030. After my previous experience, I was a bit hesitant to risk bricking another rare board, but old accelerators are fraught with risk, and risk is part of the hobby . Here are my findings and spoiler alert, it did not end with disaster this time.

Clock Speed Upgradeable?​

The crystal oscillator can be replaced on the Gemini series of accelerators. Using a 50Mhz 68030 and 40 Mhz 68882 from reliable sources, I tried FOX oscillators at 50 Mhz and 40 Mhz, but could not get the board to boot no matter what setting or memory configuration was tried. In the end, 33 Mhz clock speed is the only speed that would boot on this Gemini II, which is consistent with Earl Grey's Mac Plus Gemini II findings. In the day, the Gemini II was offered in 33, 40, or 50Mhz configurations, but I could find no certain information that it was possible to consistently upgrade the speed of an existing 33Mhz Gemini II by simply replacing the clock and CPU/FPU. The later Gemini Ultra, had improvements to make the board more upgradeable to higher clock speeds.

Memory​

The faster accelerators for SE have their own onboard fast RAM with a 32 bit data path to the 68030 CPU. But bypassing the SE's onboard RAM required some complex engineering with programmable logic array (PLA) chips. Therefore, not all 30 pin SIMMS work with the on-board memory on this card. For example, modern, fast, high quality Jurassic Computing 2-chip 60ns SIMMS would not work. The board came with 80ns RAM, but I wanted to use 70ns RAM to experiment with lower wait states. The 70ns 1MB SIMMS that came with my old Gemini 020/030 accelerator worked well at 2 wait states (P4 jumpers removed), so I assume this RAM was handpicked by Quesse to work with their board.

In the near future (stay tuned), I am planning to test with 4MB SIMMS and use the compatible Compact Virtual to see if I can address the 16MB of RAM that is reportedly possible.
Theory: I think the helpful note from Jurassic Computing may apply - " systems may only work with 4MB modules with 8 or 9 chips (11 address lines A0-A10 available)"

Performance​

This is a quick but not blazing board that subjectively feels faster than a stock SE/30. I really do like this board and it makes the SE much faster overall, but it is not a rocket like the Gemini Ultra.

Compared to my older Gemini 020/030 board, the Gemini II seems more robust and tolerates experimentation well. It is also even faster than the quick, modern MacEffects Performer (clone of the MicroMac Performer) for the SE. (Side Note - All SE owners should buy a MacEffects Performer!)

Norton System Info 3.2.1 Benchmarks (Tested on System 7.1.1)​

Overall System Rating

CPU

Video/Graphics

FPU

Speedometer 3.2.3 Benchmarks (System 7.1.1)​

Speedometer 3 is not a consistent benchmarking tool, as I've found its results may vary widely from run to run. Nevertheless, here is a comparison with another 33Mhz Mac, the later speed-bumped LCIII.

---
Related thread: https://tinkerdifferent.com/threads...-020-to-030-for-macintosh-se.4437/#post-39011

 

[nekogahora280]

I love making the lowly SE faster, since that was my first computer back in the late 80s and early 90's, and accelerator boards were an economical upgrade path compared to the $4,000+ dollars for a loaded SE/30 from the college bookstore.

So I am always on the lookout for cool SE accelerators. I was blessed to become the new steward of a Total Systems Gemini II accelerator (engineered by Quesse) purchased from @nekogahora280 after bricking my previous 16Mhz Gemini 020/030. After my previous experience, I was a bit hesitant to risk bricking another rare board, but old accelerators are fraught with risk, and risk is part of the hobby . Here are my findings and spoiler alert, it did not end with disaster this time.
View attachment 22588

Clock Speed Upgradeable?​

The crystal oscillator can be replaced on the Gemini series of accelerators. Using a 50Mhz 68030 and 40 Mhz 68882 from reliable sources, I tried FOX oscillators at 50 Mhz and 40 Mhz, but could not get the board to boot no matter what setting or memory configuration was tried. In the end, 33 Mhz clock speed is the only speed that would boot on this Gemini II, which is consistent with Earl Grey's Mac Plus Gemini II findings. In the day, the Gemini II was offered in 33, 40, or 50Mhz configurations, but I could find no certain information that it was possible to consistently upgrade the speed of an existing 33Mhz Gemini II by simply replacing the clock and CPU/FPU. The later Gemini Ultra, had improvements to make the board more upgradeable to higher clock speeds.

Memory​

The faster accelerators for SE have their own onboard fast RAM with a 32 bit data path to the 68030 CPU. But bypassing the SE's onboard RAM required some complex engineering with programmable logic array (PLA) chips. Therefore, not all 30 pin SIMMS work with the on-board memory on this card. For example, modern, fast, high quality Jurassic Computing 2-chip 60ns SIMMS would not work. The board came with 80ns RAM, but I wanted to use 70ns RAM to experiment with lower wait states. The 70ns 1MB SIMMS that came with my old Gemini 020/030 accelerator worked well at 2 wait states (P4 jumpers removed), so I assume this RAM was handpicked by Quesse to work with their board.

In the near future (stay tuned), I am planning to test with 4MB SIMMS and use the compatible Compact Virtual to see if I can address the 16MB of RAM that is reportedly possible.
Theory: I think the helpful note from Jurassic Computing may apply - " systems may only work with 4MB modules with 8 or 9 chips (11 address lines A0-A10 available)"

Performance​

This is a quick but not blazing board that subjectively feels faster than a stock SE/30. I really do like this board and it makes the SE much faster overall, but it is not a rocket like the Gemini Ultra.

Compared to my older Gemini 020/030 board, the Gemini II seems more robust and tolerates experimentation well. It is also even faster than the quick, modern MacEffects Performer (clone of the MicroMac Performer) for the SE. (Side Note - All SE owners should buy a MacEffects Performer!)

Norton System Info 3.2.1 Benchmarks (Tested on System 7.1.1)​

Overall System Rating
View attachment 22612

CPU
View attachment 22613

Video/Graphics
View attachment 22615

FPU
View attachment 22616

Speedometer 3.2.3 Benchmarks (System 7.1.1)​

Speedometer 3 is not a consistent benchmarking tool, as I've found its results may vary widely from run to run. Nevertheless, here is a comparison with another 33Mhz Mac, the later speed-bumped LCIII.
View attachment 22617

---
Related thread: https://tinkerdifferent.com/threads...-020-to-030-for-macintosh-se.4437/#post-39011

I'm glad I entrusted this board to you.
I didn't have 60ns RAM so I searched for a 4MbSIMM as junk and installed it, but it only recognized it as a 1MbSIMM. You found out from your article that it was due to the number of RAM chips.
I had a GeminiULTRA 25Mhz before, but at the time I didn't know how to replace a crystal oscillator with a higher frequency one, so I searched and finally got a GeminiULTRA 25Mhz, which I was disappointed and was struggling with money, so I sold both to Byee. I read your article (I had two of them, so I found out I need to search for GeminiULTRA again.
I believe that if you can replace the MacSE with a small power supply to create space and have a SCSI extension cable, it will be possible to speed up SCSI if you can connect it to the high-speed SCSI terminal on the Gemini II. However, since the MacSE is a new model, the current SCSI transfer speed may be sufficient. If you use the individual you have given it to have the crystal oscillator set to about 36Mhz, the GemStart screen will display 40Mhz, which will make you feel very happy. I would like you to experiment as much as possible. I'm writing on a translation site so I'm sorry if it's difficult to read.

nekogahora280

 

[Ubik]

I'm glad I entrusted this board to you.
I didn't have 60ns RAM so I searched for a 4MbSIMM as junk and installed it, but it only recognized it as a 1MbSIMM. You found out from your article that it was due to the number of RAM chips.
I had a GeminiULTRA 25Mhz before, but at the time I didn't know how to replace a crystal oscillator with a higher frequency one, so I searched and finally got a GeminiULTRA 25Mhz, which I was disappointed and was struggling with money, so I sold both to Byee. I read your article (I had two of them, so I found out I need to search for GeminiULTRA again.
I believe that if you can replace the MacSE with a small power supply to create space and have a SCSI extension cable, it will be possible to speed up SCSI if you can connect it to the high-speed SCSI terminal on the Gemini II. However, since the MacSE is a new model, the current SCSI transfer speed may be sufficient. If you use the individual you have given it to have the crystal oscillator set to about 36Mhz, the GemStart screen will display 40Mhz, which will make you feel very happy. I would like you to experiment as much as possible. I'm writing on a translation site so I'm sorry if it's difficult to read.

nekogahora280

Thanks @nekogahora280. I will try a 36Mhz clock. BlueSCSI v2 does not do well on SE benchmarking. The disk seems fast when using the Mac, but I'm going to flash the latest BlueSCSI V2 firmware to see if this may help benchmarks. On the memory I ordered 9 chip 4MB 30 pin 70ns SIMMS, which were often used for 486 gaming PCs of the period.

 

[JDW]

Your CPU score of 8.69 in Speedometer is really good.

I made these benchmarks of the MacEffects Performer, 33MHz FPU XTAL at left, and 16MHz at right below. I am always shocked at how little difference that 33MHz FPU makes.



I made these last year in the WarpSE thread:



We were testing various versions of Zane's firmware so the disk performance differs. But while a CPU score of 4.12 may not seem that much bigger than 3.31, it sure feels like a huge difference when using WarpSE. I can feel the Performer is faster than stock, but when I install WarpSE, it is a very noticeable speed boost. And what's most interesting is WarpSE uses a 68000, while the Performer has an 030.

Also note the differences between the "Graf" (graphics) score among all 3 accelerators. WarpSE scores a 3.4! But because your Gemini II's CPU score is more than double WarpSE, it surely must be faster overall.

 

[Ubik]

Your CPU score of 8.69 in Speedometer is really good.

View attachment 22625

I made these benchmarks of the MacEffects Performer, 33MHz FPU XTAL at left, and 16MHz at right below. I am always shocked at how little difference that 33MHz FPU makes.

View attachment 22626

I made these last year in the WarpSE thread:

View attachment 22628

We were testing various versions of Zane's firmware so the disk performance differs. But while a CPU score of 4.12 may not seem that much bigger than 3.31, it sure feels like a huge difference when using WarpSE. I can feel the Performer is faster than stock, but when I install WarpSE, it is a very noticeable speed boost. And what's most interesting is WarpSE uses a 68000, while the Performer has an 030.

Also note the differences between the "Graf" (graphics) score among all 3 accelerators. WarpSE scores a 3.4! But because your Gemini II's CPU score is more than double WarpSE, it surely must be faster overall.

Hi @JDW! Thanks for the exciting news on the WarpSE. The WarpSE looks like a fantastic and very balanced accelerator! Your are right, there will be a big difference between 3 and 4 CPU results in Speedometer 3. I can't wait until the WarpSE is ready for sale!

Regarding the Speedometer 3.23 tests, I did some experimenting this evening and had some amazing results when I changed the System 7 Memory Cache settings and tested the Gemini II. Increasing the cache to the max 768k resulted in a massive increase in the graphics score, which verified through both Norton System Info and the subjective feel of the SE, which was much, much faster in games like Spectre. Conversely, reducing the Memory Cache to 128k reduced both the graphics and CPU scores in Speedometer. The disk continued to show as slow, but copying large files did not seem subjectively slow, so I'm not sure about the disk scores.

Any theories why the cache would have such an impact? I always thought the cache settings was just for disk access, but I guess it impacts the accelerator's access to its on board fast RAM??

Gemini II System 7.1.1 Memory Cache Setting Performance Impact​

Speedometer 3.23​

Norton System Info 3.21​

 

[JDW]

Very surprising. Not sure about the right answer.

QuickDraw is disk intensive, and benchmarks do load from disk. The only way to know if that has something to do with it would be to boot from a RAM Disk with System Info and Speedometer on it. You'd need a fair amount of RAM to do that, however. But if you booted from a RAM disk and re-ran the tests and then showed little to no difference despite the Disk Cache difference, then such would be proof it is possibly QuickDraw or other disk access from the Benchmark utilities that are benefitting from a larger Disk Cache setting in the Memory Control panel. But if you boot from a RAM disk and still find the same huge different between 128K and 768K disk cache settings, I can only speculate that your Gemini II card is using the Disk Cache setting to impact it's performance. But if that's true, it's something totally unique that I've never seen before when using any other accelerator.

 

[nekogahora280]

Very surprising. Not sure about the right answer.

QuickDraw is disk intensive, and benchmarks do load from disk. The only way to know if that has something to do with it would be to boot from a RAM Disk with System Info and Speedometer on it. You'd need a fair amount of RAM to do that, however. But if you booted from a RAM disk and re-ran the tests and then showed little to no difference despite the Disk Cache difference, then such would be proof it is possibly QuickDraw or other disk access from the Benchmark utilities that are benefitting from a larger Disk Cache setting in the Memory Control panel. But if you boot from a RAM disk and still find the same huge different between 128K and 768K disk cache settings, I can only speculate that your Gemini II card is using the Disk Cache setting to impact it's performance. But if that's true, it's something totally unique that I've never seen before when using any other accelerator.

I never thought that disk cache would affect speed. Until now, the benchmark has only been run with the default settings. I wanted to try and see if there is reproducibility on my machine.

 

[Ubik]

This is a theory, but I think this may provide a clue of why the cache impacts increases performance of the Gemini II, and it's the higher 768k that really drives up the video performance. Like the IIsi, the board uses its onboard RAM for video. In the case of the Gemini it's the ROM to RAM function in the Gemstart 3.0 extension that pushes the Toolbox into the board's ROM, including Quickdraw. Clearly, there something about the Gemini II that's unique, since I have not noticed this in my other SE accelerators with on board RAM. I discovered this ancient article from GOPHER (the grandfather of the web) which provided a clue on how the max cache pushes System 7 in certain types of hardware to read Quickdraw from the fast RAM. It's actually Gemini AI that found this deeplink.

From https://web.archive.org/web/20161031165244/http://ccat.sas.upenn.edu/gopher/other/special/help/Macintosh/Macintosh System FAQ

The IIsi and the IIci use system RAM to store the video image on your screen. (Other Macs with internal video have video RAM separate from the main system RAM so this trick doesn't apply to them.) The internal video competes with the System for use of this RAM, and that competition slows down your Mac like two children fighting in the back seat of your car adds an hour to the time it takes to get to the beach. To stop the fighting a smart parent will put one child in the front seat and one in the back. A smart Mac owner will put the internal video in the front seat and the system in the back seat. To push the system out of the front seat set a IIsi's cache to between 384K and 768K which will take up all the space in the front seat not occupied by the internal video and force the system to sit in the back. The exact value depends on the type of monitor you have installed. Experiment to see what works for you. Unfortunately this trick doesn't work when virtual memory is turned on, but if you're using virtual memory you're probably more concerned about saving memory than gaining speed anyway.

From this discussion it sounds like using virtual memory may negate this benefit. It's possible if I extend the memory of the card to 16MB and use Compact Virtual the video-cache performance link may not work.