And you can always tell the armchair-developers from the real deal when they tell you that "time consuming" is an A-OK trait for a development platform.
Is English not your first language? Because I have yet to see anyone post anything resembling that in this thread. Perhaps you can tell me your native tongue and I can try to explain in that language instead?
"Cache misses"? These are video games. How many cache misses are you expecting?
Greatly depends on what you are doing. Using SPEs for tesselation/vertex deformation, using them for shadow creation or lighting simulation, many different areas where you could see cache misses and create stalls if cache management was automatic.
Larrabee is a _GPU_, and it is not even pretending to be a CPU. This is _not_ the same as the Cell! I don't know how much more clear I can be.
You can think yourself as concise as you like, doesn't change anything. Cell's initial design goal was to remove the necessity for a GPU, much like Larrabee. In the abstract you can claim Larrabee is a GPU, but a bunch of modified P56 cores with some basic texture sampling hardware seems a lot more like a vector processor with some graphics functionality then what any reasonable person would consider a GPU. Larrabee needs to have an abstraction layer running just for rasterization, if that is what you consider to be a GPU then by all means, knock yourself out.
Writing multi-threaded code is hard on an SMP architecture. Writing it on something like the Cell, which is asymmetric, is even more difficult. Unless there is some sort of compelling performance reason for this, and best I can tell, there is not, it was a dumb design decision.
Since you can't see why anyone would use an AMP setup it must be wrong, yeah, brilliant display of logic there. The performance differences are rather clear and easy to demonstrate, run a physics simulation on the 360s CPU and on Cell, best if you use one that someone else handled by the sounds of it. It isn't remotely difficult to see where Cell is
significantly more powerful then its' SMP counterpart running certain types of code. Cell is, without a doubt, more powerful on a per watt or per transistor basis. There is debate on if it was worth the trade off given the increased development time required- the fact that Cell is more powerful isn't in question however.
You seem to not comprehend that "programming for the Cell" is not the same problem as "writing multi-threaded code". If the latter were so hard, the 360 would be experiencing myriad performance issues. Interestingly, it's not.
Where are you getting this logic from? The PS3 isn't experiencing myriad performance issues so it must be as straightforward to code for as the 360 using your exacting logic.
Because, um, Larabee is a GPU, and the Cell has already proven itself not to be? Honestly, did you even read what you were responding to before responding in a fanboy rage?
You break down then exactly why Larrabee is a GPU. Larrabee is closer to Cell's architecture then it is any GPU we have seen to date. Just look at the amount of abstraction require to get Larrabee to handle simple rasterization. Abrash published a nice article on it not that long ago. Maybe it's just me, but I have a hard time taking seriously that something is a dedicated device when its' most vocal supporter needs to explain how you emulate a GPU in order to get it to perform basic rasterization.
Ah, so the definition of "top-tier" is reduced to "people who write reasonably-performing PS3 games". That's a definition that only a fanboy could love.
A top tier developer can write code that performs well on any platform. If the most someone can handle is getting x86 code to perform well on a single core then they are far removed from it.
I feel like you didn't even comprehend what I wrote, which is starting to not surprise me. SPEs are hard to fill, because they're so specialized in what they do well.
Of course, who isn't aware of this? This is no different then PS2 was at all with its' vector units, which is something you keep dodging, why? With all of your self claimed expertise, why do you ignore the fact that Cell, for all of its issues, is still easier to deal with the EE was at this point in its life cycle? If you have nearly the level of experience you claim this should be well known to you. Cell *is* easier to develop for then what it replaced.
If programmers are finding that the ratio of SPEs to PPEs is too heavily weighted towards SPEs, it's time to adjust that ratio.
If Polyphony, NaughtyDog or Insomniac start saying it I would listen. I would take the word of people who have proven themselves utilizing manual coded vector units- but so far they think it works very well based on everything they have said.
Why do you think Sony has a magic crystal ball that told them that 7:1 is the right ratio?
You think of it bass ackwards is your problem. You think it is the hardware companies jobs to make your life easier. As a developer, your job is to utilize the hardware you are dealing with in the best possible manner. It is why we are currently seeing an increasingly large rift between platform exclusive titles in terms of what they offer.
Funny that your examples of "unmatched visuals" are games that haven't even come out yet. Let me know how reality matches up with bullshots.
KZ2 came out quite a while ago, GT5P came out a while before that and UC2 was playable at E3. Do you know anything about the PS3?
for the record, killzone 2 is out and looks great, but didn't come close to delivering what it promised, much like the original.
Playing the game and the old E3 trailer side by side, outside of better AA, what was the big difference?
Just looking at screen shots on IGN
Play the games side by side and get back to me. Hook up both your 360 and your PS3 side by side on comparable displays and try it out. It really isn't all that close(not huge, but the PS3 clearly has an edge on exclusive titles).
Facebook
Twitter