Sony knew that 3D hardware can get very messy to develop for, for this reason their new console will keep its design simple and practical… Although this may come with a cost!
The main processor is a modification of LSI’s CoreWare CW33300 which, at the same time, is binary-compatible with SGI’s MIPS R3051. This chip runs at 33.87 MHz and it features:
Like other MIPS R3000-based CPUs, it supported configurations with up to four coprocessors, Sony customised it with two:
Additionally, there’s an extra processor found in the main CPU:
Sometimes any subsystem (graphics, audio or CD) may require large chunks of data at a fast rate, however the CPU may not be able to always keep up with the demand.
For this reason, the CD-ROM Controller, MDEC, GPU, SPU and the Parallel port have access to an exclusive DMA controller that takes control of the main bus whenever they require it.
Graphics data is manipulated by the CPU with the cooperation of the GTE and then sent to Sony’s Proprietary GPU for rendering.
If you’ve been reading the Sega Saturn article, let me tell you that the design of this GPU a lot simpler!
Insomniac’s Spyro: Year of the Dragon will be used as example to show how a scene is drawn.
To start with, the GPU uses triangles as primitives to form 3D models. Being the only available primitive means that backgrounds and foregrounds make no difference in terms of composition (both are made of triangles). 2D games inherit the same nature: They are just flat polygons (two triangles joined to form a quadrangle). The GPU also includes routines for handling quadrangles as sprites.
That being said, the CPU sends geometry data to the GPU by filling its internal 64 byte FIFO buffer with commands (up to three). A command basically states how and where to draw one primitive. Now, if there are lots of polygons to render, for performance reasons it’s better to have them sorted by the distance from the camera, with the nearest ones first. For this, the GPU implements an ordering table: A dedicated table where each entry is indexed using a depth value (also called ‘Z-value’) and contains the address where the command resides. The CPU needs to work out which depth does the polygon have, and then reference it in the correct entry of the table, which will be handled by the GPU. Multiple DMA functions are provided to assist the CPU/GPU with the creation and traversing of this list.
Once the geometry is received, clipping is applied to skip operations over unseen polygons (residing outside the camera’s viewport or behind other polygons).
Compared to the more complex Sega Saturn, the GPU only requires a single frame-buffer.
In order to apply lighting effects over these polygons, the system provides two algorithms:
The reason of having this choice comes down to the fact that Flat shading provides ~2.5 times more polygons per second than Gouraud, so it’s important to optimise which polygons need a more realistic shading than others.
Finally, shaded surfaces are blended with texture maps to produce the final result.
The GPU performs a very inexpensive routine called Affine Texture Mapping to stamp textures on our polygons. This technique only operates using 2D coordinates (X/Y values) while discarding the third coordinate used for perspective (Z-value).
The unit also includes the following effects available to use:
It’s worth mentioning that the PS1 happened to excel at those effects!
Here are some examples of game characters designed from the ground up for the 3D era, they are interactive so I encourage you to check them out!
The system features 1 MB of VRAM that will be used to store the frame-buffer, textures and other resources the GPU will need to access almost instantly. In theory, with the available amount of VRAM one could allocate a massive frame-buffer of 1024×512 pixels with 16-bit colours or a realistic one of 960×512 pixels with 24-bit colours allowing to draw the best frames any game has ever shown… This sounds pretty impressive right? Well, it does raise a couple of issues, for instance:
Alright, so let’s have a 16 bpp 640x480 buffer instead, which leaves 424 KB of VRAM for materials. So far so good? Again, such resolution may be fine on CRT monitors, but not really noticeable on those 90s TVs everyone had at their homes. Then, is there any way to optimise the frame-buffer? Introducing adjustable frame-buffers.
In essence, instead of wasting valuable VRAM by using ‘unappreciated’ resolutions, this console’s GPU allows to decrease the dimensions of the frame-buffer to effectively increment the space available for other resources.
A common setup consisted in dividing the 640x480 frame-buffer into two 320x480 ones, then using a technique called page-flipping to render multiple scenes at the same time. Page-flipping consists in switching the location of the frame for display between the two available whenever the game wants it, allowing the game to render one scene while displaying another, thus hiding any flickering effect and improving loading times (something that the player will certainly appreciate!).
Overall, our suggested layout only consumed 600 KB of VRAM. The rest (424 KB) was often used to store colour lookup-tables and textures that, combined with 2 KB of texture cache available, resulted in a very convenient and efficient setup.
Finally, it is worth mentioning that VRAM could be mapped using multiple colour depths at the same time, meaning that we could allocate a 16 bpp frame-buffer with 24 bpp textures alongside it. This is another feature allowing further optimisation of space.
Whereas the PS1 had a very simple and suitable architecture, problems ended up arising anyway. Surprisingly, certain issues were tackled with very clever workarounds!
The routines used for handling geometry and applying textures were known to carry some inaccuracies:
These explain why users may notice instabilities while playing some games. The effect is also referred as texture warping, some games often resorted to tessellation (dividing a big polygon into smaller ones) in order to reduce these unwanted effects, others just reduced the amount of textures in favour of solid colours.
If a game strived for a more realistic scenery than the GPU could provide, one available option was to stack two triangles and use the Motion Decoder to feed pre-rendered cinematic on them. FMV video could weight a lot of space, luckily the CD-ROM was prepared for this.
Some games relied on it specifically for composing backgrounds and, honestly, it was quite convincing to see them in a CRT TV, obviously modern emulators with upscaling capabilities will tell on very quickly.
The Sound Processing Unit or ‘SPU’ takes care of this. It supports the enormous amount of 24 channels of ADPCM samples (a more efficient version of the well-known PCM sampling) with a sampling rate of 44.1 KHz (CD quality).
This chip also provides the following capabilities:
512 KB of buffer is used to transfer audio, this memory is accessible from the CPU and CD controller, the latter will reserve it while playing an Audio CD.
Similarly to the Saturn’s boot process, after turning this console on, it will:
Games have all the facilities that the CD medium provides: Large storage (640 MB), good audio quality and a ‘not-so-slow’ read speed thanks to the 2x drive. Additionally, there are two I/O ports (Serial and Parallel) available for add-ons, however these were removed in later revisions of the console due to lack of use and the fact that they could also be applied to crack the copy protection system.
The SDK provided C libraries which used BIOS routines to access the hardware. This is the main factor that helped to emulate the PS1 on wide range of platforms.
The port of the controller and the Memory Card are electrically identical so the address of each one is hardcoded, Sony altered the physical shape of the ports to avoid accidents.
Like any other optical media, in order to fetch data from a CD, a laser beam is used to read the pits (zeroes) and lands (ones) from the track of the disc. Now, conventional discs are not 100% flat and they often have tiny fluctuations in their tracks. These defects are completely unnoticeable while reading the data since lasers can automatically calibrate themselves as they read.
This is what Sony based their copy protection on: The CD reader of the PS1 will only accept discs with a defined frequency known as Wobble Groove which is only applied during mastering. These patterns cannot be replicated through conventional burners.
However, this check is only executed once at the start, so manually swapping the disc just after passing the check can defeat this protection… with the risk of damaging the drive. Alternatively, tiny boards programmed to emulate the wobble signal could be soldered in the console. These boards are known as Modchips and, while legally questionable, were incredibly popular.
The use of emulators were seen as a threat for publishers as well, as a result some games included their own checks (mostly checksums) to combat any type of unauthorised use.
This article is part of the Architecture of Consoles series. If you found it interesting please consider donating, your contribution will be used to get more tools and resources that will help to improve the quality of current articles and upcoming ones.
A list of desirable tools and latest acquisitions for this article are tracked in here:
## Interesting hardware to get (ordered by priority) - First PS1 revision with a controller (£5 - £20 ?) - Any game (£5 ?)
Always nice to keep a record of changes.
## 2020-02-28 - Expanded VRAM section ## 2020-01-27 - Expanded 'Models' section - Added more reasons textures wobble ## 2019-10-29 - Added some 3d models to fiddle with ## 2019-10-09 - Improved Scratchpad term ## 2019-09-17 - Added a quick introduction ## 2019-08-29 - Better explanations ## 2019-08-09 - Corrected vague CPU info ## 2019-08-08 - Ready for publication