Every console launch I can remember has demonstrated a significant disparity between the quality of launch titles, and more importantly, between the best looking launch titles and 2nd and 3rd generation software. This thread is being created for your feedback, pro and con, on this disparity in graphic quality. Since the Xbox 360 is almost here and X05 has given us a good idea of what to expect from the launch software I felt it was a good time to take a jab at this subject. The 360 will be my personal main focus, but feel free to broaden your own comments *within the scope of this topic*.
Feel free to add your reasons or post explainations why certain reasons are not valid because correcting them wont result in better looking titles.
1. Developers are being Rushed.
a. Like most launches developers are contrained to a specific a fairly short timeframe and must have their games out within a couple months buffer period to make a launch. When you add in...
b. Beta Kit issues, well, this is a problem. As IGN noted in their special last week, the Beta Kit situation did not go as MS planned. Final dev kits were not received until Augest, and even then there were rumblings about underclocked packages.
c. It does not help that the Xbox 360 has developers tieing their head around a new CPU. Xbox1 and PC devs have been spoiled with large OOO chips with significant amounts of cache at high frequencies. Now they are being asked to multi-thread on an in-order CPU that has 1MB of cache to share. There is also a transition from the x86 ISA to the PPC one. Oh, and you don't get to touch the CPU until at most 6 months before launch. So all those ideas of experimenting, testing, and designing new code ideas to deal with the transition? Forget it, you get a couple months to clean up your code and make the game playable.
d. Devs also get to bang their heads against a new GPU. Yes, they get a thin DX-like API. But with eDRAM and a significant shift toward Shaders and away from fixed function hardware and heavy TMU use it requires a new line of thought to exploit the potential of the chip. Yet this requires having access to the hardware to test shader code and see what works. Not only have developers not been working on GPUs without Unified Shaders and eDRAM, they have been working on GPUs without SM3.0 support, no HDR effects, no hardware tesselation, no HOS, no MEMEXPORT, no vertex texturing, no dynamic branching, etc. Devs were given SM2.0 cards (9800/X800) and while many titles look great this limitation shows pretty clearly.
=> 2nd and 3rd gen titles wont be pushed by a launch date and will have the luxury of having working development kits from day 1.
2. Ports.
a. Current Gen Console ports. A number of titles like Gun, King Kong, BF2:MC, FFXI, NFSMW, etc are Xbox1/PS2 ports with some additional graphical bells and whistles in HD.
b. Xbox1 upgrades. Kameo, PDZ, etc. These titles started life on the Xbox1. Their art design was concieved within the limitations of the Xbox1 and many of their art assets and rendering engine betray this.
c. PC ports. Quake 4, Prey, CoD2, Oblivion, etc. All good looking games, but also games that are designed to be compatible with DX8 (and older!) GPU architectures.
=> Games designed for next gen consoles or the Xbox 360 (like Gears of War with UE3 and PGR3) really demonstrate the gap between even a 1st gen next gen game and porting over an existing franchise/engine/assets. As developers draft design concepts with the hardware in mind and build their art assets and engines accordingly we should see a significant change.
3. Transition away from DX7/DX8 fixed function design.
a. As being discussed in the X1600XT thread we are seeing the limitations of tossing current rendering engines/techniques at Shader-centric hardware. To roughly summerize, the R530 has 12 PS units (pixel shader) but only 4 TMU units (texture mapping). The result? In games with DX7/DX8 class hardware in mind for redundancy the X1600XT acts, for the most part, like a 4x1 GPU at 590MHz. In shader heavy code the X1600 excels amazingly.
Xenos is in the same boat. It has 16 TMUs @500MHz. This is basically the same as the X800XT (16 @ 500MHz). The difference being that the X800XT has a 1:1 ration between PS and TMUs whereas Xenos has a (total) Shader ratio of 3:1.
b. Similarly, Xenos has 8 ROPs. While being on the eDRAM and designed for 4xMSAA @ 720p indicates they are fairly powerful for that task, it must be noted that the X800XT has 16 ROPs. In certain circumstance Xenos, where certain ROP tasks are heavily leaned upon amd were the bottleneck, Xenos would be no faster than an 8 pipe GPU.
Since Shader performance is Xenos' forté game designs that favor heavy ROP or TMU usage wont benefit in the same way a design focused on Shaders will.
c. The concept of shaders can be frustrating because we use the term "programable shaders" but what is the end result in games? I know frequently I think of shaders as the hardware that allows normal maps, specular highlights, bump mapping, displacement mapping, subsurface scattering, and so forth. Some of those are run of the mill effects that build on features already seen in DX7 class cards, others have been unusable due to the lack of shader power and/or important features (e.g. USA and vertex texturing should make displacement maps a reality). So what is the big deal? The recent Toy Store slides give a good idea of how shaders can be used in a real game design to give a graphic punch so far not seen in games. Without seeing the slides I would have thought such a design was very CPU intensive, but almost all the effects are Shaders. Of course this type of design could never be done if one was looking for DX7/DX8 fallback.
=> DX9 hardware has been on the market since fall of 2002 (Radeon 9700/R300 launch). Low end PC GPUs are finally getting decent DX9 performance with the X1300. On the PC end DX9 will soon be the "bare minimum" and with millions of Xbox 360s and PS3 sold developers will have even more reasons to base their minimum featureset on DX9 and focus on shader heavy code. As the R530 and Xenos show there is a shift away from ROPs/TMUs toward shader units and over time this should bear fruit.
4. Under utilized / un-used features.
a. We already covered the launch "crunch". The impact is that a lot of features are currently under utlized and even un-used. HOS, hardware tesselation, SM3.0 (e.g. geometry instancing, dynamic branching), FP16/10, vertex texturing, alpha-to-mask, and so forth. Since the Alpha Kit hardware either did not support many of these features or were hacks it was difficult to test the final hardware and design with these features in mind. And therefore many of these features are un-used or slapped on ala "lens flares".
b. XeCPU<>Xenos cache lock. One of the neat "streamlined" features of the Xbox 360 is the ability to lock the cache and stream data directly to Xenos without touching the system memory or memory bandwidth at over 10.8GB/s (not including the 2:1 compression). Since many of the titles are ports from other platforms and none of them had access to this ability until to late it is understanble that it has gone un-used. But XeCPU is designed with some graphical work in mind (e.g. the beefed up VMX units with 128bit registers and 1 Dot Product per cycle). With almost 40GFLOPs of peak performance per core, it seems that using a CPU core as an aid to rending was always in the plan (maybe as the Xbox 360 equivalent of a geometry shader?). Offloading certain rendering techniques to the CPU should further push the graphical abilities of the Xbox 360.
=> 2nd and 3rd gen titles will have the advantage of building engines with these features in mind. The un-used features should allow a significant jump in graphical fidelity, and certain features being used as last minute slap patch jobs will look much better when part of the initial concept art focus.
5. Up coming 3D Engines--UE3, id Software, CryTek, Offset, etc--are aimed at the new consoles.
=> With a base featureset close to the 360's spec these new engines should allow development teams using these development tools to quickly get a handle on the advanced features in the console and allow the creation of art assets, from day 1, with the features and focus of the hardware in view. These are, for the most part, not upgraded DX7/DX8 engines like Source, CryTek, and Doom 3. While they wont be as good as a native engine because they are cross platform and have some redundancy, the ability to work on art assets from day 1 is a huge advantage compared to working with out dated technology.
6. As multi-core goes, performance shall follow.
=> Only a couple development teams are using more than 1 CPU core in the Xbox 360 due to the obvious timeframe constraints. Interestingly, the titles that use more than 1 core appear to be running very smoothly. e.g. CoD2 is running at a solid 60fps on the Xbox 360 and is using 1 core for particles and effects. The X1800XT (R520) @ 1280x1024 with 4xAA and 8xAF runs at 55fps; taking into account the need to sync with the TV (either 30Hz or 60Hz) it is unlikely that the X1800XT could be stable at 60fps @ 720p. This demonstrates the advantage of multithreading, but it also gives a glimpse of Xenos on newer software. While CoD2 was not designed specifically with Xenos in mind (it is a PC title with DX8 support and it does not use many of Xenos' features like HDR) it gives us a small hint that Xenos is not the reason many Xbox 360 games do not overwhelm us graphically or many titles struggle to hit 30fps. CoD2, one of the best looking Xbox 360 launch titles, can run at 60fps then that is a good indication that future Xbox 360 specific titles should not only run smoothly but also look really good.
I could give more reasons but I think I will stop there and allow other people to add their own comments
For me, the conclusion seems obvious: The Xbox 360 is being under utilized by launch titles. And not by a small margin. I think we are seeing the paradigm shift and the old paradigm of 3D engines is not jiving well with the future. We are seeing a lot of "solar flares" and "bloom" type patch jobs and really are not experiencing what the hardware can really do.
I fully expect 2nd and 3rd generation titles to look degrees of magnitude better than the best launch titles. I believe GOW and PGR3 are just the tip of the iceberg. 18-24 months from now, when the first software reaches store shelves that has been designed from day 1 with a beta kit in hand will really contrast with the titles we are seeing at launch. Especially for the developers who are doing launch titles, who have already knocked on the hardware some, I think they will have a pretty good vision of what the hardware can do and will really take the time to sit down with their design staff and craft a game concept that will graphically accent the closed box's strengths.
And I am excited to see what fall 2007 brings! (On a side note I expect PS3 devs to go through a similar process, although I believe their alpha-kit situation is far superior to that of the Xbox 360 developers so I have big hopes for the PS3 come fall 2006).
Feel free to add your reasons or post explainations why certain reasons are not valid because correcting them wont result in better looking titles.
1. Developers are being Rushed.
a. Like most launches developers are contrained to a specific a fairly short timeframe and must have their games out within a couple months buffer period to make a launch. When you add in...
b. Beta Kit issues, well, this is a problem. As IGN noted in their special last week, the Beta Kit situation did not go as MS planned. Final dev kits were not received until Augest, and even then there were rumblings about underclocked packages.
c. It does not help that the Xbox 360 has developers tieing their head around a new CPU. Xbox1 and PC devs have been spoiled with large OOO chips with significant amounts of cache at high frequencies. Now they are being asked to multi-thread on an in-order CPU that has 1MB of cache to share. There is also a transition from the x86 ISA to the PPC one. Oh, and you don't get to touch the CPU until at most 6 months before launch. So all those ideas of experimenting, testing, and designing new code ideas to deal with the transition? Forget it, you get a couple months to clean up your code and make the game playable.
d. Devs also get to bang their heads against a new GPU. Yes, they get a thin DX-like API. But with eDRAM and a significant shift toward Shaders and away from fixed function hardware and heavy TMU use it requires a new line of thought to exploit the potential of the chip. Yet this requires having access to the hardware to test shader code and see what works. Not only have developers not been working on GPUs without Unified Shaders and eDRAM, they have been working on GPUs without SM3.0 support, no HDR effects, no hardware tesselation, no HOS, no MEMEXPORT, no vertex texturing, no dynamic branching, etc. Devs were given SM2.0 cards (9800/X800) and while many titles look great this limitation shows pretty clearly.
=> 2nd and 3rd gen titles wont be pushed by a launch date and will have the luxury of having working development kits from day 1.
2. Ports.
a. Current Gen Console ports. A number of titles like Gun, King Kong, BF2:MC, FFXI, NFSMW, etc are Xbox1/PS2 ports with some additional graphical bells and whistles in HD.
b. Xbox1 upgrades. Kameo, PDZ, etc. These titles started life on the Xbox1. Their art design was concieved within the limitations of the Xbox1 and many of their art assets and rendering engine betray this.
c. PC ports. Quake 4, Prey, CoD2, Oblivion, etc. All good looking games, but also games that are designed to be compatible with DX8 (and older!) GPU architectures.
=> Games designed for next gen consoles or the Xbox 360 (like Gears of War with UE3 and PGR3) really demonstrate the gap between even a 1st gen next gen game and porting over an existing franchise/engine/assets. As developers draft design concepts with the hardware in mind and build their art assets and engines accordingly we should see a significant change.
3. Transition away from DX7/DX8 fixed function design.
a. As being discussed in the X1600XT thread we are seeing the limitations of tossing current rendering engines/techniques at Shader-centric hardware. To roughly summerize, the R530 has 12 PS units (pixel shader) but only 4 TMU units (texture mapping). The result? In games with DX7/DX8 class hardware in mind for redundancy the X1600XT acts, for the most part, like a 4x1 GPU at 590MHz. In shader heavy code the X1600 excels amazingly.
Xenos is in the same boat. It has 16 TMUs @500MHz. This is basically the same as the X800XT (16 @ 500MHz). The difference being that the X800XT has a 1:1 ration between PS and TMUs whereas Xenos has a (total) Shader ratio of 3:1.
b. Similarly, Xenos has 8 ROPs. While being on the eDRAM and designed for 4xMSAA @ 720p indicates they are fairly powerful for that task, it must be noted that the X800XT has 16 ROPs. In certain circumstance Xenos, where certain ROP tasks are heavily leaned upon amd were the bottleneck, Xenos would be no faster than an 8 pipe GPU.
Since Shader performance is Xenos' forté game designs that favor heavy ROP or TMU usage wont benefit in the same way a design focused on Shaders will.
c. The concept of shaders can be frustrating because we use the term "programable shaders" but what is the end result in games? I know frequently I think of shaders as the hardware that allows normal maps, specular highlights, bump mapping, displacement mapping, subsurface scattering, and so forth. Some of those are run of the mill effects that build on features already seen in DX7 class cards, others have been unusable due to the lack of shader power and/or important features (e.g. USA and vertex texturing should make displacement maps a reality). So what is the big deal? The recent Toy Store slides give a good idea of how shaders can be used in a real game design to give a graphic punch so far not seen in games. Without seeing the slides I would have thought such a design was very CPU intensive, but almost all the effects are Shaders. Of course this type of design could never be done if one was looking for DX7/DX8 fallback.
=> DX9 hardware has been on the market since fall of 2002 (Radeon 9700/R300 launch). Low end PC GPUs are finally getting decent DX9 performance with the X1300. On the PC end DX9 will soon be the "bare minimum" and with millions of Xbox 360s and PS3 sold developers will have even more reasons to base their minimum featureset on DX9 and focus on shader heavy code. As the R530 and Xenos show there is a shift away from ROPs/TMUs toward shader units and over time this should bear fruit.
4. Under utilized / un-used features.
a. We already covered the launch "crunch". The impact is that a lot of features are currently under utlized and even un-used. HOS, hardware tesselation, SM3.0 (e.g. geometry instancing, dynamic branching), FP16/10, vertex texturing, alpha-to-mask, and so forth. Since the Alpha Kit hardware either did not support many of these features or were hacks it was difficult to test the final hardware and design with these features in mind. And therefore many of these features are un-used or slapped on ala "lens flares".
b. XeCPU<>Xenos cache lock. One of the neat "streamlined" features of the Xbox 360 is the ability to lock the cache and stream data directly to Xenos without touching the system memory or memory bandwidth at over 10.8GB/s (not including the 2:1 compression). Since many of the titles are ports from other platforms and none of them had access to this ability until to late it is understanble that it has gone un-used. But XeCPU is designed with some graphical work in mind (e.g. the beefed up VMX units with 128bit registers and 1 Dot Product per cycle). With almost 40GFLOPs of peak performance per core, it seems that using a CPU core as an aid to rending was always in the plan (maybe as the Xbox 360 equivalent of a geometry shader?). Offloading certain rendering techniques to the CPU should further push the graphical abilities of the Xbox 360.
=> 2nd and 3rd gen titles will have the advantage of building engines with these features in mind. The un-used features should allow a significant jump in graphical fidelity, and certain features being used as last minute slap patch jobs will look much better when part of the initial concept art focus.
5. Up coming 3D Engines--UE3, id Software, CryTek, Offset, etc--are aimed at the new consoles.
=> With a base featureset close to the 360's spec these new engines should allow development teams using these development tools to quickly get a handle on the advanced features in the console and allow the creation of art assets, from day 1, with the features and focus of the hardware in view. These are, for the most part, not upgraded DX7/DX8 engines like Source, CryTek, and Doom 3. While they wont be as good as a native engine because they are cross platform and have some redundancy, the ability to work on art assets from day 1 is a huge advantage compared to working with out dated technology.
6. As multi-core goes, performance shall follow.
=> Only a couple development teams are using more than 1 CPU core in the Xbox 360 due to the obvious timeframe constraints. Interestingly, the titles that use more than 1 core appear to be running very smoothly. e.g. CoD2 is running at a solid 60fps on the Xbox 360 and is using 1 core for particles and effects. The X1800XT (R520) @ 1280x1024 with 4xAA and 8xAF runs at 55fps; taking into account the need to sync with the TV (either 30Hz or 60Hz) it is unlikely that the X1800XT could be stable at 60fps @ 720p. This demonstrates the advantage of multithreading, but it also gives a glimpse of Xenos on newer software. While CoD2 was not designed specifically with Xenos in mind (it is a PC title with DX8 support and it does not use many of Xenos' features like HDR) it gives us a small hint that Xenos is not the reason many Xbox 360 games do not overwhelm us graphically or many titles struggle to hit 30fps. CoD2, one of the best looking Xbox 360 launch titles, can run at 60fps then that is a good indication that future Xbox 360 specific titles should not only run smoothly but also look really good.
I could give more reasons but I think I will stop there and allow other people to add their own comments
For me, the conclusion seems obvious: The Xbox 360 is being under utilized by launch titles. And not by a small margin. I think we are seeing the paradigm shift and the old paradigm of 3D engines is not jiving well with the future. We are seeing a lot of "solar flares" and "bloom" type patch jobs and really are not experiencing what the hardware can really do.
I fully expect 2nd and 3rd generation titles to look degrees of magnitude better than the best launch titles. I believe GOW and PGR3 are just the tip of the iceberg. 18-24 months from now, when the first software reaches store shelves that has been designed from day 1 with a beta kit in hand will really contrast with the titles we are seeing at launch. Especially for the developers who are doing launch titles, who have already knocked on the hardware some, I think they will have a pretty good vision of what the hardware can do and will really take the time to sit down with their design staff and craft a game concept that will graphically accent the closed box's strengths.
And I am excited to see what fall 2007 brings! (On a side note I expect PS3 devs to go through a similar process, although I believe their alpha-kit situation is far superior to that of the Xbox 360 developers so I have big hopes for the PS3 come fall 2006).
