To preempt the inevitable, I am ACTUALLY a professional software engineer and I have worked on several 3D game engines.

That out of the way...

MikeB:

Regardless of validity of any your points, your side of this "Argument" screams propaganda, as everything is "best case." The fact of the matter is that from a game engine standpoint, TONS of the functionality is interdependant. Unlike something like this topic talks about (Password cracking) different pieces of the game engine rely on other pieces.

The IK engine can't function until the physics calculations are pretty much done. Certain multi-pass rendering can't be done without data from previous passes. AI functionality is dependant on physics to some degree and collision detection routines (advanced ones any way) need to wait until at least the initial steps of animations are done.

Where are the real world numbers? How many asynchronous threads need to wait and do nothing for other data during a single render pass? The fact is games really do not NEED that many concurant threads, and a ton of the processing power of the cell is wasted on gaming applications.

For applications that can run compleatly asynchronously or highly multithreaded applications, yes, the cell can out perform pretty much any other comercially available product.

On another note, you are talking about developers spending millions on R&D for games that will probably have them ending up losing money, when they can produce a pretty much accepted standards for this gen for a fraction of the cost. Game companies aren't some purely artistic entity recieving grant money to develop their masterpieces. Most of them are just barely scraping by and your criticism for them not devoting untold time and money to something that will probably be a horrible business decision is rediculous.

@ CrashMan

The provided data are best case scenarios in both PC CPU and Cell setups, thus a direct comparison is valid. And yes games design is more complex, but asynchronous game engine design should actually result in performance gains compared to more synchronous designs handling data simultaneously (less waiting!). It's all about smart design.

[edit:  never mind this part]

What do you think about my proposal a few posts back? I'm quite serious. I think it would benefit all parties.

Final-Fan said: [edit: never mind this part] What do you think about my proposal a few posts back? I'm quite serious. I think it would benefit all parties.

I agree, I would like to see such a thread being created. I'm not sure it would be particularly accurate or useful, but the idea certainly has potential and would avoid a lot of arguments and snippets being repeated in so many threads.

 

MikeB said: @ CrashMan Regardless of validity of any your points, your side of this "Argument" screams propaganda, as everything is "best case. The provided data are best case scenarios in both PC CPU and Cell setups, thus a direct comparison is valid. And yes games design is more complex, but asynchronous game engine design should actually result in performance gains compared to more synchronous designs handling data simultaneously (less waiting!). It's all about smart design.

No a direct comparison ISN'T valid as if you do purely linear game design, it won't be as fast as multi threading but it will be more efficient.  There will be NO wasted cycles.  

 In a synchronous scenario, both best and worst case, and naturallt the average case, all cpu cycles will be used.  In the average case for the cell processor, there will be tons of wasted cycles.  How can you claim a direct comparison is valid?

If you are comparing the two for generic processing power, yes a direct comparison is valid, but we are talking about a very specific area: game development.  In order to do a valid comparison, an average case HAS to be used as the best case for the cell isn't possible for games.

@ Crashman

 

Neither is it for using a single CPU.



Results will be dependent on game design. One of the early examples of Asymetric -Multi-Processing would be the Amiga where different processing tasks are performed simultaneously and independently requiring minimal intervention from the CPU. Sound processing was for example done by the Paula chip, also for example the blitter operates independently and signals when it is finished in a fully asynchronous multi-processing manner. This was a very powerful design which is much harder to emulate, even today emulating a 7 Mhz Amiga on a modern PC can result in a worse user experience with music stuttering and the OS not always responding to user input, like AmigaOS always responds without delay when clicking a button.

takeru51 said: Now all I have to do is use the power of t3h c3ll to travel back in time, so that I can be the one who cracks the Japanese code and be famous for winning WW2. Thank you PS3.

I hear Sony will be adding time travel to the list of the PS3's capabilities in an upcoming firmware update.  Of course, the same source also said Nintendo would start making good games too so I'm thinking that source may be a little unreliable.  

yes but these were definitely defined processors for game tasks not general use processing threads. You are dancing around the issue that in even a perfectly implemented game design, there will be waiting by the cell processor.

What we are talking about is the amount of time it takes to draw one frame of the game. After all the Wii COULD produce the same effects that the PS3 could and COULD create a screen shot that looks like *insert generic fanboy graphic example* but it wouldn't be able to do it in real time speeds (as many functions would need to be implemented in software, etc)

For a fair comparison you would need to compare the amount of time it takes to draw one frame in each environment.

The limiting factor for the PS3 is the frame will be drawn at a rate equal to the speed of the SLOWEST thread. Therefore you CAN'T use a generic best case for the PS3 in the game environment, it eronious, as it will never happen. All threads can't POSSIBLY be running continuously in a game environment.

And besides any one who knows any thing about efficiency comparisons knows you NEVER use the best case unless you are trying to fudge the results, you HAVE to use the average case for any sort of valid comparison.

Ok, this is CS 101, If you compare the best case of two sorting algorithms (lists are already sorted) then you will come to the conclusion that "all sorting algorithms are equally efficient" and maybe even conclude that some slower algorithms out perform some faster algorithms (bubblesort vs merge sort)

However, this is a rediculous argument as bubble sort is O(n^2) and merge sort is O(n log n) for average (and worse case incidentally)

Comparisons by best case in general or rediculous, but particularly in this case. You need REAL DATA if you are going to convince any one of even average intelligence.

Now in a environment that is inherantly multi-threaded (ie a PC OS) then the cell would be phenominal, but game do not have that nature and therefore can not be compared as such.

@ Crashman

Yes, there are always delays involved but also with regard to games designed for a single CPU solution. The fun is really that the SPEs each have memory available to them which can operate as fast as cache, it's design (systems on a chip) is potentially more powerful than just an ordinary multi-core CPU setup.

PS-She said: takeru51 said: Now all I have to do is use the power of t3h c3ll to travel back in time, so that I can be the one who cracks the Japanese code and be famous for winning WW2. Thank you PS3. I hear Sony will be adding time travel to the list of the PS3's capabilities in an upcoming firmware update. Of course, the same source also said Nintendo would start making good games too so I'm thinking that source may be a little unreliable.

Nice trolling there.