PS2, AI and physic calculation

[Bohdy]

Correct me if i am wrong, but isn't memory access a major bottleneck for current Cpu's?

And besides, the Gekko is a very well designed chip, 1t-Sram or no and can certainly pack a punch.

I think you give it too little credif, faf.

Maybe you haven't stared long enough at the volcanic ash room in Wind Waker.

 

[London Geezer]

yes Bohdy, the Gekko is a pretty processor.

still, from what i've seen of HL2 (quite a lot), there's a whole lot of stuff happening in that game, not only pretty effects. we're talking about completely interactive geometry (looked like it in the videos, u never know though, they were demostrations after all and might be different in the final game), robust physics engine and all.

 

[Bohdy]

Call me hopelessly optimistic, but I expect Prime 2 to be comparable to HL2 visually.

Except for the complex physics, HL2 doesn't obviously have anything that I haven't seen in one or another curent GC game.

 

[London Geezer]

Call me hopelessly optimistic, but I expect Prime 2 to be comparable to HL2 visually.

Except for the complex physics, HL2 doesn't obviously have anything that I haven't seen in one or another curent GC game.

ERRR... well thats the thing... it's the complex physics that sets it apart from the rest matey...

"visually" even Halo2 might be better (i said might), but it would have NOWHERE NEAR the level of interactivity of HL2. nothing does at present time.

 

[Bohdy]

Well. as Valve recently re-affirmed, a 733 should be able to run it.

And I maintain that if a 733 can then Gekko can.

Of course, this is all theory, as even if a GC port was made the necessary optimisation would not be made

 

[Fafalada]

Correct me if i am wrong, but isn't memory access a major bottleneck for current Cpu's?

L2 caches and OOE make a world of difference there though.

And besides, the Gekko is a very well designed chip, 1t-Sram or no and can certainly pack a punch.I think you give it too little credif, faf.

Don't get me wrong, in general I am much more fond of PPC designs then x86, even if it's only for the instruction set preference.
But that doesn't change the fact Gekko is a 750cx core derivative with added 2way SIMD for fpu, and as such more or less on par with P3 running on the same clock.
Given the memory architecture it should outperform it, but not enough to make up for 300mhz difference.

Anyway maybe that min spec is referring to scaling down everything, including interactive elements - but that kinda defeats the point of arguing if you could run the same game on console spec cpus.

 

[Bohdy]

But that doesn't change the fact Gekko is a 750cx core derivative with added 2way SIMD for fpu, and as such more or less on par with P3 running on the same clock.

And what about the lockable D-cache, the Write Gather Pipe and the Data Quantisation? These things are supposed to be highlights of the Gekko, how much do they help?

And how useful is the Gekko paired-single capability compared to SSE?

 

[notAFanB]

But that doesn't change the fact Gekko is a 750cx core derivative with added 2way SIMD for fpu, and as such more or less on par with P3 running on the same clock.

And what about the lockable D-cache, the Write Gather Pipe and the Data Quantisation? These things are supposed to be highlights of the Gekko, how much do they help?

And how useful is the Gekko paired-single capability compared to SSE?

what about memory bandwidth, massive memory utilasation and non fixed form TnL in the (targeted) PC realm?

 

[AbbA]

Relative to what?

If you had to choose between XBox's celeron@733 and the Ps2's EE, in order to manage a physical engine like the havoc, what you would choose? And why?

If it were possible to quantify what "level of HL2" means, then it might be possible to give a reasonable answer.
Anyway I'm not too familiar with AI in HL2 to begin with, as for physics I can tell you that things Havoc does can be performed at good speeds but you need to be willing to go through hoops of platform specific code rearchitecting. It's debatable how often that's a reasonable solution though.

I say something like the havoc engine added to a good artificial intelligence, according to you is possible to see something similar on the EE?

Thanks!

 

[London Geezer]

we also have to remember that the console versions would run at 640x480 at 30fps. so that's a lot of performance gained in the process...

i thought we agreed on the fact that the more u use the VU's for non-graphical tasks, the worse the game looks...

we should get information on what the VU's were used in past FPS on ps2. like timesplitters 2 or the coming Killzone... then maybe we would have something to compare it to..

 

[Fafalada]

And what about the lockable D-cache, the Write Gather Pipe and the Data Quantisation?

Cache lock allows to use part of cache as SPRAM - not sure how much need there is for it though, given already lowlatency memory and large L2 cache. The reason why scratchpad has been used a lot on PS1 and now PS2 is that their MIPS cores both have tiny L1 cache and no L2 so it makes a dramatic difference in some situations.
Quantized data loads/stores are actually part of the extended SIMD I mentioned, and yeah they are cool, they simplify use of compressed data and in effect tie in with the Gekko being needed for dynamic geometry processing in the system.
These two also both need handcoding to take advantage of.

Don't know if I'd exactly call Write Gather Pipe a highlight though, afaik there's lots of other cpus that have it, including R59k (although it's called differently). It basically helps more efficient use of the bus by grouping data writes to memory.
Anyway of the above features the SIMD instructions make the most difference as far as I can tell.

And how useful is the Gekko paired-single capability compared to SSE?

IMO, they easily top SSE which lacks a couple of extra instructions to become really usefull.

If you had to choose between XBox's celeron@733 and the Ps2's EE, in order to manage a physical engine like the havoc, what you would choose? And why?

Can I take a 733mhz EE with 128kb L1 cache instead?

Well at the risk of sounding EE fanboy, I'd definately take that for pyhsics.
It's tough to say if it would perform dramatically better overall (there are still parts of code where there's no substitute for higher clock), but there are definately parts which lend themselves to VU optimization really nicely(I've gotten vectorised versions of some algoryhtms to run 3-4x faster clock-for-clock then P3&SSE). And besides, it just has that really rewarding feel tinkering with microcoding stuff that runs really cool.
If VU0 also had it's own capability to copy data in and out, I would also say it would be dramatically better at the said tasks.

I say something like the havoc engine added to a good artificial intelligence, according to you is possible to see something similar on the EE?

Wasn't there some info saying Killzone physics is supposed to be pretty good, not to mention they bragged about their AI too?