David Cook
Well-known member
This Does Not Compute recently posted a video showing the excellent performance of the Applied Engineering TransWarp 4300 accelerator card in a Color Classic.
I purchased two TransWarp cards from eBay to see the performance for myself. Below, notice a couple of bodges, and also a purposeful hole machined in the add-on cache. (More on that later.)

My cards did not come with a FPU. So, I added the 68882 math coprocessor into the card's socket.

There has been some question as to what the finger-edge connector is for. It appears to designed for a video card.

Oddly, the accelerator PCB is not labeled with a product name. It turns out there is a little sticker hiding underneath the cache board. I could not wiggle the cache out of the connector and didn't want to break it. So, below is the best picture that I could take of the label.

The card barely fits in a standard LC pizza-box case. Look how tall it is!

At first, I didn't believe it fit. Maybe the cache is removable because it doesn't fit? Then I realized that the slot cut in the cache is designed to allow the plastic peg in the LC cover to fit through and hold down the main board. I guess it is designed to fit.

But, what about the four wires atop the ADB inductor? Surely they short-circuit the bottom of the accelerator?

I thought those wires would make contact with the crystal leads. Instead the PLCC socket rests on the ADB inductor. Ok. It fits, but with no room to spare.

Cache Size
The cache is 128KB. This is secret to this board's performance. The LC, LC II and Color Classic all have 16-bit data buses, as opposed to 32-bit data buses on better computers. This means memory delivers half as much data per read.
In the graph below, notice how much faster the cache is than main memory. Three times the speed!

This doesn't directly translate into 3x computer performance, because it depends on how often a computer is in a tight loop within the CPU, versus hitting the Applied Engineering cache, versus needing something from main memory, versus how often it is waiting for the disk drive.
040 Competitor
Before I show you the overall results, an alternative to the Applied Engineering 68030 @ 40MHz is a Sonnet Presto 68040 @ 25MHz. The 68040 cpu has a larger internal cache than the 68030 cpu, and it can execute parts of multiple instructions in a row.
I have two of these Sonnet accelerators. I noticed a couple of differences between the Sonnet boards. Although they indicated the same PCB revision, one has bodge wires on U5 but the other doesn't. Secondly, one uses a Motorola 88915 clock driver, the other uses a Qualcomm QS5917. Lastly, and most importantly, one is slightly overclocked using a 55 MHz crystal vs a 50 MHz crystal, resulting in a CPU @ 27.5MHz vs 25 MHz.

The bodge wires on U5 seem to force two unused (?) gates to GND and Vcc respectively. Likely to eliminate electrical noise.

Results
The Applied Engineering TransWarp 4300 with 128KB cache + FPU is slightly better than the Sonnet Presto 68040 @25 MHz, but slightly worse than @27.5. The fact that the 40MHz 68030 is comparable to a 25MHz 68040 shows the importance of the 128KB cache for a 16-bit memory bus machine.

Either accelerator is a game changer for the LC, LC II, and Color Classic. The Applied Engineering card did not work in an LC III, Color Classic II, or LC 550. Based on the Color Classic II performance, it wouldn't have made much of a difference.
- David
I purchased two TransWarp cards from eBay to see the performance for myself. Below, notice a couple of bodges, and also a purposeful hole machined in the add-on cache. (More on that later.)

My cards did not come with a FPU. So, I added the 68882 math coprocessor into the card's socket.

There has been some question as to what the finger-edge connector is for. It appears to designed for a video card.

Oddly, the accelerator PCB is not labeled with a product name. It turns out there is a little sticker hiding underneath the cache board. I could not wiggle the cache out of the connector and didn't want to break it. So, below is the best picture that I could take of the label.

The card barely fits in a standard LC pizza-box case. Look how tall it is!

At first, I didn't believe it fit. Maybe the cache is removable because it doesn't fit? Then I realized that the slot cut in the cache is designed to allow the plastic peg in the LC cover to fit through and hold down the main board. I guess it is designed to fit.

But, what about the four wires atop the ADB inductor? Surely they short-circuit the bottom of the accelerator?

I thought those wires would make contact with the crystal leads. Instead the PLCC socket rests on the ADB inductor. Ok. It fits, but with no room to spare.

Cache Size
The cache is 128KB. This is secret to this board's performance. The LC, LC II and Color Classic all have 16-bit data buses, as opposed to 32-bit data buses on better computers. This means memory delivers half as much data per read.
In the graph below, notice how much faster the cache is than main memory. Three times the speed!

This doesn't directly translate into 3x computer performance, because it depends on how often a computer is in a tight loop within the CPU, versus hitting the Applied Engineering cache, versus needing something from main memory, versus how often it is waiting for the disk drive.
040 Competitor
Before I show you the overall results, an alternative to the Applied Engineering 68030 @ 40MHz is a Sonnet Presto 68040 @ 25MHz. The 68040 cpu has a larger internal cache than the 68030 cpu, and it can execute parts of multiple instructions in a row.
I have two of these Sonnet accelerators. I noticed a couple of differences between the Sonnet boards. Although they indicated the same PCB revision, one has bodge wires on U5 but the other doesn't. Secondly, one uses a Motorola 88915 clock driver, the other uses a Qualcomm QS5917. Lastly, and most importantly, one is slightly overclocked using a 55 MHz crystal vs a 50 MHz crystal, resulting in a CPU @ 27.5MHz vs 25 MHz.

The bodge wires on U5 seem to force two unused (?) gates to GND and Vcc respectively. Likely to eliminate electrical noise.

Results
The Applied Engineering TransWarp 4300 with 128KB cache + FPU is slightly better than the Sonnet Presto 68040 @25 MHz, but slightly worse than @27.5. The fact that the 40MHz 68030 is comparable to a 25MHz 68040 shows the importance of the 128KB cache for a 16-bit memory bus machine.

Either accelerator is a game changer for the LC, LC II, and Color Classic. The Applied Engineering card did not work in an LC III, Color Classic II, or LC 550. Based on the Color Classic II performance, it wouldn't have made much of a difference.
- David