Playstation 3 e3 thread 3

[Paul]

Again, the point is that Vince saw 65nm as the sole target

Alot of us did.. I did. Based on everything coming out, the news that SNE, and SCE+Toshiba were building and upgrading fabs to .65nm specification to build Cell processor chips, the originall propossed patent plans of the Broadband Engine(Which was the original PS3 CPU to be built by Toshiba + SCE which would use Rambus interconnects proved in the SCE+Toshiba contract) all of this lead me to FIRMLY believe that PS3's IC's, both the BE and the propossed VS(Figure 6, the original PS3 before it got scrapped and nVIDIA came in) would be built ground up from .65nm.

But something happened.. Where is the Broadband Engine? Where is this TFLOPS class chip that was planned for the Playstation 3? Now you could argue if they would hit a TFLOPS with it or not, but that's not the point. The point is, what the fuck happened to it?

Now either the propossed BE+VS combo which was gonna be used in the PS3 was decided too risky(dev friendlyness, yields or whatever), or the plan for PS3 all along was to use a 90nm Cell processor and nVIDIA GPU is for you to decide.

I have come to the conclusion that SOMETHING happened leaving Sony to suddenly get nVIDIA in the game and ditch the whole Toshiba+SCE built BE+VS. Because there were detailed, detailed patents on the thing leading back to 2002, and even the BE contract between Rambus, SCE and Toshiba...

Now I'm not saying that the nVIDIA thing happened overnight, it didn't, it took place a year ago; but most definately PS3 got shifted from it's original embodiment.

Now whether or not the change was for the better we will never know, and I will devulge on this whole thing later but I'm a 17 year old with a gorgeous girlfriend and yea... I have an active social life...

 

[Legend]

of course there isn't a PS3 hardware yet. it is a year away from launch!
that is why they phrase "based on PS3 specifications" is used.
we won't probably see final devkits untill after or near the launch of PS3. this is the targeted performance, and whether the final PS3 product ends up slightly better or worse than those specs won't actually have a noticeable affect- at least, not in the early stages.

To Paul:
keep in mind that PS3 is a whole year away, so specs will change- including the out side design and contollers- maybe even a logo will appear.

 

[corysama]

... and Sony never showed 'false' demos for any of their gaming machines...

Let me tell you a little story about the PS2 E3 demos. The PS2 rasterizer is technically able to put out one poly per cycle and it runs at 150 MHz. That's a theoretical 150 million polys per second. Unfortunately, the vector hardware (which also runs at 150 MHz) takes longer than a single cycle to transform the vertices of the triangles. Super-simple tranform routines take more like 6 or 8 cycles. In practice most games squeeze between 10 and 20 million polys per second out of the machine.

So how did the demo FMV sequences from Tekken and Final Fantasy 8 running "in real time on PS2 hardware"? They pre-recorded the pre-transformed and pre-formatted command stream for the GS and had a PC stream it directly off of a hard drive into the rasterizer. Was it about 150 million polys per second? Yes. Was it "in real time on PS2 hardware"? Technically yes. Was it an honest demonstration? No. It was off by about 130 million polys per second -a 700% inflation of the truth.


In the opening shot of the Killzone video, the camera zooms out from the close-up on the sergeant's face an then pans left to show another soldier sitting on the deck. For the first few frames the soldier isn't visible when he should be. Some people have interpreted this pop-in as evidence on an culling bug common in real-time systems. However, if you look closely you'll see the the soldier is visible, it just that the ship is drawing on top of him. This kind of bug makes no sense in a real-time render which uses the Z-buffer to determine visibility. It does make sense if the sky, soldier and ship were rendered seperately and then composited in a video editing package. A simple mistake by the guy mixing the pre-rendered video streams could cause that.

This and the fire on the flame-thrower victims leads me to believe that the Killzone demo was rendered in Maya. I'm sure that the PS3 is physically capable of make fire that cool, but setting up a flame simulator that looks that good and runs so fast that it doesn't drag down an already overpowering scene? That is a PhD-level project. It would be trivial though to use Maya's built-in and extremely cool fire simulation systems. If you were on a tight time budget for an E3 demo, which would you do?

 

[Legend]

Tekken 5 is far better than PS2's Tekken demo. and GT4 is far better than the tech demo GT. in fact, seeing PS2 tech demos now is surprising. they aren't really that good now. this also happened with PSP; Death Jr. wowed people. now, the game is the uglies title on PSP (that soon).

 

[MechanizedDeath]

... and Sony never showed 'false' demos for any of their gaming machines...

Let me tell you a little story about the PS2 E3 demos. The PS2 rasterizer is technically able to put out one poly per cycle and it runs at 150 MHz. That's a theoretical 150 million polys per second. Unfortunately, the vector hardware (which also runs at 150 MHz) takes longer than a single cycle to transform the vertices of the triangles. Super-simple tranform routines take more like 6 or 8 cycles. In practice most games squeeze between 10 and 20 million polys per second out of the machine.

So how did the demo FMV sequences from Tekken and Final Fantasy 8 running "in real time on PS2 hardware"? They pre-recorded the pre-transformed and pre-formatted command stream for the GS and had a PC stream it directly off of a hard drive into the rasterizer. Was it about 150 million polys per second? Yes. Was it "in real time on PS2 hardware"? Technically yes. Was it an honest demonstration? No. It was off by about 130 million polys per second -a 700% inflation of the truth.

Play TTT and watch that Tekken demo again. Then post the same thing. TTT on that schoolyard scene alone kills the demo, period. The demo was aliasing like crazy and had a dozen or so guys crowded around the fighters. The stage had a tanker truck and some other crude objects. The TTT schoolyard stage had a couple dozen people in the background and two fighters that look more detailed than the demo version. At the very least, TTT (a launch game) matched the Tekken demo. But IMO, it killed it.

The FF8 dance has no real equal in terms of scene setup. But look at the intro to RR5 (another launch game), or scenes from The Bouncer (realtime) or even TTT again. TTT had better models and much better bgs. That FF8 red herring is trolled out on a daily basis the last few weeks, but if anyone took a second to watch the demo again (on PS2.IGN) they'd laugh at how simple the scene was. The FF8 demo wasn't amazing for its level of detail so much as the attempt to recreate a stunning scene from FMV, and it was a good job. But IIRC, there were clipping problems, and the lighting was not very good as the models still had that "paper-thin" look to them. You compare it again to TTT, which has two highly-detailed character interacting on screen, and it's no contest.

Post some comparison screens if you want. I think Square had the FF8 demo pics on their site years ago, and it's easy enough to find high-quality TTT shots. TTT alone beat those two demos IMO. I don't know why some people still try to flog this dead horse. All PS2 demos were surpassed by early 2nd-gen. The exception is the head demo, and mostly b/c we haven't had any floating head moments in PS2 games. Feel free to correct me if I'm wrong here. PEACE.

 

[PC-Engine]

I remember some people were saying CELL would be built on 65nm with a quick process change to 45nm....yeah ok.

 

[MechanizedDeath]

I remember some people were saying CELL would be built on 65nm with a quick process change to 45nm....yeah ok.

I believed this last year. What a difference a year makes. Then again, I was only reading fragments of the Cell discussion last year, so I didn't pay attention to many of the details. Maybe there were tell-tale signs from back then, but it seemed like the doubters were just the usual suspects. :? Of course you would have guess this PC-E. You would have also guessed the PS3 being the PS2.5 had that turned out to be true too. PEACE.

 

[passby]

Thanks corysama for the post. It was very interesting, very good to know.

 

[Geeforcer]

I can't escape the feeling that NVIDA GPU orderd at the last possible moment.

 

[ANova]

I remember some people were saying CELL would be built on 65nm with a quick process change to 45nm....yeah ok.

While 45nm may not be achievable anytime soon, I wouldn't be surprised if Cell moved to the 65nm process for the PS3 revision.

 

[overclocked]

I can't escape the feeling that NVIDA GPU orderd at the last possible moment.

Yes and i think this was the best thing that happened to PS3.

 

[Panajev2001a]

... and Sony never showed 'false' demos for any of their gaming machines...

Let me tell you a little story about the PS2 E3 demos. The PS2 rasterizer is technically able to put out one poly per cycle and it runs at 150 MHz. That's a theoretical 150 million polys per second. Unfortunately, the vector hardware (which also runs at 150 MHz)

Cut it right there... 150 MHz...

First, the GS could not output 150 MPolygons/s even if it was dreaming: its triangle set-up rate is limited to 75 MPolygons/s (1 polygon = 1 vertex in their lingo). The only thing it can render at that rate are point-sprites.

Second, the VU's run at 300 MHz.

Third, the fastest perspective transform on VU1 is 5 cycles per vertex and on VU0 is 7 cycles per vertex (you are limited by FDIV speed).

Fourth, real-time, interactive games on PlayStation 2 got all of those demos beat one by one.