Depends how they code for it. I think they could isolate SPU's to certain activities rather than using the distributed model.
Use Vector Unit Zero ... It's... a miracle of technology!
- Thread starter joe75
- Start date
Panajev2001a
Veteran
joe75 said:
You need to read what othe rpeople post.
ihamoitc2005
Veteran
Minor correction
Yes, micromode is key to graphics performance since Vu0 has same geometry transform capability as Vu1.
Many companies hype games with talk of Vu0 usage claimes but final product not reflective of added performance. Only hard data on Vu0 usage is from many years ago and shows old ratchet and clank with 8% Vu0 and 56% Vu1 (27% of available vector unit power if used in micro-mode) and polygon draw rate was 7.5M/sec in PA analysis. No doubt many modern games are far superior in Vu0 and Vu1 usage. Two examples are Ghosthunter with very nice detailed environments real-time lighting and stencil shadows and on opposite end of the spectrum is Transformers with extremely high polygon count and nice bloom lighting effect.
Ghosthunter
http://eurogamer.net/article.php?article_id=54079
" Quite simply, there are times in it where I have never seen anything so impressive in a game."
Transformers
http://eurogamer.net/article.php?article_id=55753
"Standing on the top of the temple on the third level overlooking the rainforest's dense canopy of intricately detailed trees stretching right off into the distance - with zero pop-up - is not a sight many can have expected to see on this machine. The great thing is, should you single out one piece of the landscape, you can basically glide all the way there and watch in admiration as the whole scene gradually changes until you find yourself suddenly dwarfed by a huge canopy of massively detailed trees. What an achievement."
In regards to Splinter Cell, it was very poor port. Conversion document available on Gamasutra and suggests very mediocre engine. They revealed Splinter Cell PA analysis compared with a PA analysis from Ratchet and Clank and difference in efficiency is very obvious. Splinter Cell only = 16,000 polygons drawn per frame at 30fps (almost PS1 performance) while Ratchet and Clank = 125,000 polygons drawn per frame at 60fps. Of course the Ratchet and Clank PA analysis they provided for comparison in article was not best available ... not 125,000 polygons drawn per frame but just 90,000.
vliw said:In Macromode it's useless, micromode is the key and the maximum utilization rate as been on Jak 3 8%
Yes, micromode is key to graphics performance since Vu0 has same geometry transform capability as Vu1.
Many companies hype games with talk of Vu0 usage claimes but final product not reflective of added performance. Only hard data on Vu0 usage is from many years ago and shows old ratchet and clank with 8% Vu0 and 56% Vu1 (27% of available vector unit power if used in micro-mode) and polygon draw rate was 7.5M/sec in PA analysis. No doubt many modern games are far superior in Vu0 and Vu1 usage. Two examples are Ghosthunter with very nice detailed environments real-time lighting and stencil shadows and on opposite end of the spectrum is Transformers with extremely high polygon count and nice bloom lighting effect.
Ghosthunter
http://eurogamer.net/article.php?article_id=54079
" Quite simply, there are times in it where I have never seen anything so impressive in a game."
Transformers
http://eurogamer.net/article.php?article_id=55753
"Standing on the top of the temple on the third level overlooking the rainforest's dense canopy of intricately detailed trees stretching right off into the distance - with zero pop-up - is not a sight many can have expected to see on this machine. The great thing is, should you single out one piece of the landscape, you can basically glide all the way there and watch in admiration as the whole scene gradually changes until you find yourself suddenly dwarfed by a huge canopy of massively detailed trees. What an achievement."
In regards to Splinter Cell, it was very poor port. Conversion document available on Gamasutra and suggests very mediocre engine. They revealed Splinter Cell PA analysis compared with a PA analysis from Ratchet and Clank and difference in efficiency is very obvious. Splinter Cell only = 16,000 polygons drawn per frame at 30fps (almost PS1 performance) while Ratchet and Clank = 125,000 polygons drawn per frame at 60fps. Of course the Ratchet and Clank PA analysis they provided for comparison in article was not best available ... not 125,000 polygons drawn per frame but just 90,000.
mckmas8808
Legend
_phil_ said:
My bad I should have pointed that out. I meant what you meant. EA has programmed for 3 SPEs worth of data (if that explains it a little better). They have 4 other SPEs worth left for other stuff like physics, sound, etc if I remember correctly.
DeathKnight
Regular
My sentiments exactly... lolBenQ said:
Bingo. After reading the PS2 hardware analysis PDF from 2003 it kind of strikes home: Even in 2003, 3 years after the PS2 launch, there was ongoing discussions on how to best use the architecture.Shifty Geezer said:
5 years is a very short window to get a new architecture, get comfortable with it, and release a quality product. You are looking at 2, maybe 3 new games (maybe a couple sequals tossed in with tweaks) built for the system from each dev team in that period--and if 1 is a launch title, that means it probably did not get into the hardware much.
I am sure with Xenon/CELL we will be having this same discussion in 2008 to a degree. Finding out what works best on each architecture and how it can benefit games is exciting. We may learn that threads, as unlikely as it sounds, is good on the SPEs. Or we may find an applet approach that appears popular for SPEs may work well on Xenon.
Or something completely different!
That is the nice thing with robust hardware--it kind of allows each dev team figure out their specific problems for their game design and then use the hardware how they see fit to resolve the issue. CELL and Xenon should offer up a lot of unique answers about the future of multithreading--because if there is anyone who can get the most out of CPUs and can give a good outlook on the future and future needs of CPUs it is definately game devs.
I think although he does give alot of credit to his game developement (what developer wouldnt ) , he is probably just proud of the work they have done on a system that is considered difficult to program for. As a average programmer I know that when i can make gains from hand coding for desired effect and performance gains (to which some thought were not achievable) there is a little pride there
Granted from the outside modest gains are pretty much the norm but as a game develops, the hours your team puts into hand coding a game squeezing every bit of resources to meet your goals (and even in some cases surpasing them) is a satisfying feet indeed. Cant blame him for having a little pride given the accomplishments that they "appear" to have made with a system like the PS2.
) , he is probably just proud of the work they have done on a system that is considered difficult to program for. As a average programmer I know that when i can make gains from hand coding for desired effect and performance gains (to which some thought were not achievable) there is a little pride there Granted from the outside modest gains are pretty much the norm but as a game develops, the hours your team puts into hand coding a game squeezing every bit of resources to meet your goals (and even in some cases surpasing them) is a satisfying feet indeed. Cant blame him for having a little pride given the accomplishments that they "appear" to have made with a system like the PS2.
That number doesn't make sense anyway.vliw said:
First time I wrote my skinning code to run on VU0 it was very slow and it was spending
more than half frame just to skin the main character. Can I say I was doing better than Naughty Dogs just cause the VU0 was running 50% of the time?!
Having analized a lot of games on PA I can say there are quite a number of games doing some kind of work on VU0 and it's not so uncommon to see VU0 running for more than 8% of a frame (at full throttle..)
ciao,
Marco
p.s. ND and Insomniac games make a good use of VU0..
Definitely. There aren;t many places to explore anymore. Most of the world has been explored already. There's nothing much of interest in space worth exploring even if we could. But ideas and technology is one such area where there can be lots of unknowns and people trying new stuff never before tried. That's what I personally find interesting in these hardwares.Acert93 said:
nAo said:That number doesn't make sense anyway.
First time I wrote my skinning code to run on VU0 it was very slow and it was spending
more than half frame just to skin the main character. Can I say I was doing better than Naughty Dogs just cause the VU0 was running 50% of the time?!
p.s. ND and Insomniac games make a good use of VU0..
Mmm...i will look, tell me one game that is supposed to use well the VU0.
Bye
Macromode is 4-10x faster then using flat R5900 FPU code.vliw said:
There was early PS2 stuff that ran entirely off VU0 in macro (no VU1 usage), and somehow still managed to impress people - graphically.
Fact is that VU0 can't do squat without R5900 intervention, so Micro mode savings are smaller then people would have you believe. And like nAo said, utilization numbers by themselves are meaningless.
If we want to talk what is truly underused, IPU is miles ahead of anything else in PS2.
overclocked
Veteran
Have some of you guys tried/played ACE-COMBAT4/5 and if what have the PA reported in these games.
Similar threads
