Qu'est ce que je disais
Allez, encore une couche:
I remember when the PS2 was first announced and the technical specifications that were bandied about at that time: 75 million polygons/second, unlimited streaming texture potential, 48GB/s of memory bandwidth, and so on. It wasn't long after this that technology analysts began to question Sony's numbers.
Polygon Performance
The 75 million number was reduced to 66 million. Afterwards, it was admitted that these PS2 numbers were a peak performance figure for flat -shaded, identically shaped polygons. Unfortunately, the image of the PS2 as some sort of polygon monster had already become firmly entrenched in the minds of the mainstream media.
Sega chose a more conservative approach, which is in keeping with their new business philosophy - to regain the trust and confidence of gamers. Since its introduction two years ago, Sega has never mislead gamers about the Dreamcast's power. 3+ million polygons is all that Sega ever claims, even though new games like Test Drive: LeMans push closer to five million in 3D scenes loaded with effects.
The truth is that the PS2 has never displayed more than 2-3 million polys in a game. The main problem is a memory one. With only a 4MB VRAM cache on its GS graphics processor, the PS2 is severely limited in what it can achieve on screen. While it's true that 32MB of main memory and the fairly powerful Emotion Engine processor are capable of producing in the neighborhood of 10-12 million textured and lit polygons/second, the poor design of the GS and its small pipeline to main memory restrict the final number to roughly half of that.
What? You mean, regardless of the power of the EE processor and the large amount of available memory, the PS2 is still only capable of displaying 5-6 million on-screen polygons? The answer, unfortunately, is yes. By contrast, the Dreamcast has only 16 MB of main memory and a processor that is only capable of one-half the number of polygons/second - ie. 5- 6 million - but the whole point of the exercise is to get these onto your television. An intelligent memory saving technique known as differed rendering, coupled with the PowerVR2DC graphics chip's hardware texture compression abilities, allow the Dreamcast to display all of its generated polygons.
To better understand the PS2's limitations and the Dreamcast's strengths, you need only look at the available video memory for your answer. While the DC has 8MB of VRAM, the PS2 has only 4MB of VRAM. The main problem arises because a polygon takes up roughly 40 bytes of RAM. When you have 5 million of them in a given second, this amounts to 5 million/60fps = 83,333 polygons in a give frame of animation. If each of these polygons uses 40 bytes of VRAM, you will use 3.33 MB displaying these 5 million PPS. This doesn't leave the PS2 much room for it's framebuffer which uses around 1.2MB just to display the end data, not to mention that you still need to leave room for textures to put on those polygons.
Now, there are a few tricks which will allow the PS2 to display 5-6 million PPS, even though it only has a 4MB VRAM cache. One of them is to update the cache more frequently than once a second. But, there are other bandwidth limitations that prevent this from happening more than two or three times per second and the net result is that the PS2 is still limited to 5- 6 million PPS.
Here is a table which summarizes the polygon performance of both next- generation machines:
System Processor Stage Graphics Stage Best Example**
PS2 EE + 32MB
12 million PPS* GS + 4MB
6 million PPS* Madden NFL 2001
2 million PPS*
DC SH4 + 16MB
6 million PPS* PVR2DC + 8MB
5 million PPS* Ferrari F355 Challenge
3 million PPS*
* All polygons are textured and lit and represent peak performance
** Only games available right now were considered
Unfortunately, this isn't the PS2's only shortcoming. The reason I emphasize polygon performance at all is because these number have become the defacto standard for judging a console's power, when in fact they tell less than half the story. The main disadvantage of this expensive architecture is it's poor texturing ability.
Texturing Performance
The way texturing works is simple. Polygons and texture data arrive into video memory, textures are applied to the polygons and the result is displayed on screen. Most PC users are used to games with 16MB or more of texture data. A diehard Quake III player might have a setup capable of delivering 32MB of textures during the game. 32MB? But the PS2 and DC only have 4MB and 8MB of VRAM respectively. How can they hope to compete? The answer is that consoles do not hold all of a scene's texture data in memory at once. Usually, the data is streamed over the bus from main memory in a continuous manner.
The Dreamcast is a wonderful texturing beast, due in large part to the efficiency of the PVR2DC's graphics methodology. Two things help the PVR2DC - hardware texture decompression and infinite planes deferred rendering. Unlike the PS2's GS graphics processor, the PVR2DC is capable of decompressing textures on the fly. Thus, DC programmers usually take 20-25MB of texture data and compress it at a 5:1 (sometimes 8:1) ratio to reduce the amount of texture data to only 4 or 5MB. Then, the texture data is sent over the bus to the PVR2DC which simply decompresses the data at the moment of rendering into it's original huge size.
By contrast, the PS2's GS processor has no ability to decompress textures on the fly. This means that all texture data must flow over the relatively small pipeline between main memory and the GS 4MB VRAM cache, at it's original large size. Currently, this fact has limited PS2 games to only around 10 MB of texture data/frame, and this is why the buildings look so similar in Ridge Racer 5. Lack of variety in texturing has made most PS2 games look extremely plain when compared to Dreamcast games like Sonic Adventure, Shenmue, and even Draconus: Cult of the Wyrm.
Moreover, the PVR2DC belongs to the only processor family on the market that uses deferred rendering to texture only those polygons which are facing the gamer in any given frame. Other graphics chips must texture the backs of polygons as well as the front facing polygons. The net effect is to reduce the amount of texturing that the DC has to perform in a given scene by a factor of two or three depending on the complexity of the scene. The greater the scene complexity, the more you see the benefits of deferred rendering. This is why you never see any really large free-roaming 3D games on the PS2. Crazy Taxi, Ecco the Dolphin, and Shenmue are simply not possible on the PS2, because it doesn't have deferred rendering.
Test Drive LeMans on the Dreamcast.
GT3 on the PS2.
Here is another table which summarizes texturing performance for bother machines:
Texture Data Streaming Capacity System Capacity Decompressed Texturing Ability Best Example*
PS2 10MB/frame (Main Memory -> GS Memory) 10MB/frame on screen Dead or Alive 2: Hardcore*
DC 5MB/frame (Main Memory -> VRAM 25MB/frame on screen Shenmue, Ecco the Dolphin*
* Only games available right now were considered
These two performance measure give you a pretty good idea of why the PS2 is, technically-speaking, a poor hardware design. The biggest problem of all with this architecture, however, is the difficulty that development houses are having extracting reasonable performance out of the machine. All the power in the world under the hood, doesn't do anyone much good if the games don't look good.
Development Environment
The PS2 shipped to developers with incomplete kits last year. By contrast, Sega has been giving excellent support to developers both large and small. Most DC developers are using 5th generation development kits, known as Set 5 Dev Kits. Sony mistakenly made the assumption that third-party PS2 developers would want bare bones development kits so they could program the hardware directly like they have during the last days of the PSX. Unfortunately, key features that are very hard to implement, like anti-aliasing to remove jagged edges from on-screen polygons have not yet surfaced.
Developers have responded to these PS2 programming challenges in a number of ways. Some developers like THQ (Summoner) have used a form of CRT (Cathode Ray Tube) blending to fake the effects that true anti- aliasing would offer. This is something which the DC has had for over two years, but unlike the DC CRT method, the PS2 method results in washed out, blurry textures. Tekken Tag Tournament is the perfect US launch title example. While they have eliminated the jagged edges which plague the Japanese version, the end result is that all of the textures in the game seem blurry or washed out. Hardly what I would call revolutionary for a next-generation console.
Another developmental problem, which is the reason for the jaggies in the first place, is serious lack of kit functions that will intelligently enable developers to overcome some of the limitations of the small size of the GS VRAM cache. While all Dreamcast games run at 640x480 resolution, many PS2 games only utilitize a 640x240 field- rendered display which fakes a 640x480 display. Bad jaggies are the result, and these need to be hidden through some form of anti-aliasing (AA, not yet available), or by using the CRT method described above, with all its unintended consequences.
Moreover, the EE processor is actually three separate CPUs in one core. Most developers, for lack of proper tools, are using only one third of the EE's processing ability, because both vector units (VP1 & VP2) are too hard to program. Certainly future games will take advantage of these units, thereby freeing the main CPU to implement some fairly nice AI routines, but the cost of developing these techniques has become enormous - something which I will outline in the next article.
The sad fact is that only a few development houses like EA have been able to extract reasonable next-generation performance out of the PS2 architecture. Even Namco and Konami, the kings of PSX development during the 32 bit era, are having a hard time getting more than 2-3 million PPS out of what is supposed to be the end-all of gaming machines. The fact of the matter is that Namco's 18 month old Soul Calibur on Dreamcast looks worlds better than the newly released Tekken Tag Tournament on PS2. Not very impressive compared to the promises that have been made by Sony and it's cabal of industry sycophants.
Overall
The Dreamcast is the best machine on the market. Tomorrow nothing will have changed. Technically speaking, nothing on the PS2 comes close to the beauty of Shenmue or Ecco, the speed and power of F355 Challenge or Test Drive: LeMans 24, and the sheer elegance and gaming grace of games like Metropolis Street Racer and Jet Grind Radio. If one full motion video demo of Metal Gear Solid 2 has convinced you that the PS2 is the better machine, then you haven't opened your eyes to the reality before you. The best next-generation machine from a technical standpoint is the Sega Dreamcast. Let other less informed individuals buy a machine capable of less, on the promise of one game thirteen months from now. In the meantime, you and I will be enjoying the technically best games for months to come.