seismologist
Regular
Wow that looks stunning. Hope we'll see a US release before PS3 comes out amd spoils us.
Ikusagami Screenshots
- Thread starter Slay
- Start date
pixelbox said:
Which numbers? There's a few there. Mostly i got them from here though. Peak in game figures seem to be around 15M polys per sec. That translates to 500k per frame at 30fps.
Average numbers will be much lower than that.
Hey now . THe tiled based defered renderer would have had a field day with that scene . It just would have drawn the parts of the characters you could see . :smile:The Dreamcast...
It would also draw it at full 60 updates per second on VGA, with better internal color precision and better texture filtering.
Transform comes before the visible surface determination, so bounding box tests or smart game based visibility culling would be needed to cut down on a lot of the T&L.
Overdraw definitely plays to a DC strength for pixel fillrate (and bandwidth, shading, and texturing), and the savings on external memory from not needing a bigger z-buffer, with the great many depth positions that DC's high floating point precision sorts unconditionally, affords DC designers a lot of freedom for games with a massive numbers of objects.
Overdraw definitely plays to a DC strength for pixel fillrate (and bandwidth, shading, and texturing), and the savings on external memory from not needing a bigger z-buffer, with the great many depth positions that DC's high floating point precision sorts unconditionally, affords DC designers a lot of freedom for games with a massive numbers of objects.
Last edited by a moderator:
jvd said:
Generally no. You need to move your polys first before you can determine whether or not they can be seen.
E.g. you have a soldier running across the landscape. He runs behind a hillside. Therefore you move the polys of the soldier before you can determine to cull him from your view.
Lazy8s said:
How do you determine what can be seen before you even determine where its at? This is directed towards your statement about "smart game based visibility culling". Do you have an example of how this works?
Lazy8s said:
But how is the transform power of the DC? Where is the set up done? The scene is made up of many animated characters so it's not necessarily fill rate limited. Would be interesting to find out the amount of polys in those scenes.
SH-4 can transform 10 million raw polys/sec. The PowerVR CLX has a triangle setup limit of 7 million polys/sec. Looking at those screens, at any given moment you only have a couple hundred models visible. The rest is so far you just need to use 2D sprites like lb said. Of course DC won't be able to light the scene at that quality level, but the rest is pretty trivial. There aren't that many particles from what I can see. For 300 characters each made up of 100 polygons and factoring in LOD, that's only 30K polys/frame.
Last edited by a moderator:
Ty:
PC-Engine:
Generalized tests in some cases can check if an object even has a chance to be visible based on its previous state or position, so transforming to find out specifically where or what it is might not even be necessary.
The issue was really more in regard to DC handling scenes with many objects in general than that particular scene of which no specifics are actually known. A Dreamcast would have to be tasked with lower, DC-level complexities in certain areas of performance of course.
PC-Engine:
Its peak performance is thought to be even higher, possibly. Maybe around 12M.
I think it's quite impressive, regardless of what some of the more vocal Sony haters say. Sure, there are some repeating textures, most models are the same, and it uses quite aggressive LoD, but the overall feel of the screenshots more than makes up for it. They look nearly next-geny at first glance, especially the first one.
PC-Engine said:Fun hearing all the SONY fb wetting their panties over it though.
Rome: Barbarian Invasion
http://games.tiscali.cz//images/romebarbarian/bnga.jpg
... PC CPU 2-3GHz,RAM 512-1GB,DX9 class GPU + 128MB vram ... 300 polys per NPC .. ehh.
Kingdom under Fire
... PC based CPU 733Mhz,UMA RAM 64MB,DX8 class GPU + UMA ram
Ikusagami
... poor PS2
... NOT very funny,isnt it.
Last edited by a moderator:
mckmas8808
Legend
The PS2 pic looks pretty good compared to those PC pics that you've shown.
http://watch.impress.co.jp/game/docs/20050819/ikusa10.jpg
http://watch.impress.co.jp/game/docs/20050819/ikusa09.jpg
You telling me that these two pics don't look nice for the PS2?
You telling me that these two pics don't look nice for the PS2?
seismologist
Regular
PC-Engine said:Why all the blurring? Is it to hide ugly repeating textures? Anybody got a video so we can see what the animation is like? I bet the LOD system shrinks the characters to like 10 polygons when the camera zooms out.
Fun hearing all the SONY fb wetting their panties over it though.
It's fun that on the PS2 we're getting a taste of something thought to be only possible on next gen consoles.
that sounded too much generalized ; ) and that matter is well complicated to allow for such generalizations. yes, generally you can do relatively cheap view-volume tests at the object level [ed: or any level for that matter], but when it comes to scene-wide occlusion, things are far from trivial. tons of early occlusion algorigthms have been developed throughtout the years, none of them too successful in the general case (tm). to the point where in some cases it is still more efficient to throw everything (front-to-back) at the gpu and let it sort it out through its per-pixel occlusion techniques, in which case transformation is unavoidable. unfortunately a filed of soldiers would be an example of the bad general case.Lazy8s said:
mind you, that'd be vertices, not triangles. and you don't really want to get that close to the theoretical peak. you never know if your words may not reach the ears of the publishers ; )
Last edited by a moderator:
Similar threads
