By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - PC Discussion - Why PS3/360 can not compete vs PC - GPU vs CPU Benchmarking

2 or 3 PPEs would make it essentially as easy to program as the 360


Three PPEs ARE a 360 CPU. Three power cores taped together. The basic problem is that you have only a limited number of transistors on a chip and need to use them as efficient as possible. The problem with general purpose cores is that they use much more transistors than are necessary for most repeating tasks. Because of this the CELL processor is much more efficient than a similar sized chip that has multiple standard cores put together. Its more complicated of course but at PS4 release developers should be used to it.

but I can see where game developers would have problems.


True and I am not sure if the Cell was the optimal CPU for a game console. On the other hand you only have to solve a problem once. So most game developers will have working solutions for using CELL SPEs for most of their problems by the time the PS4 comes out.

You are correct that game devs will have many things that are better done on a general purpose core, but I really doubt that programming the CELL would get easier if they would mix multiple PPE with multiple SPEs. At the moment you have one core distributing work. In your proposed architecture you would have multiple cores distributing work to multiple SPEs? Sounds like a deadlock disaster waiting to happen.




Around the Network

@Kyros: I didn't say all the cores would be distributing work, you could perfectly assign that to a single core (or a single thread which may not even occupy a core the whole time).

The transistor number problem applies, but that's outside the scope of my post. I started with the assumption (an assumption I'm not sure about) that the PS4 will use an architecture similar to the Cell. If I'm right, what will happen if that's the case is that the number of transistors they're able to use will be enough for more than 1 PPE and a few SPEs.

Remember that for the PS4, they will want to not only make PS3 developers happy but also make it as easy as possible for new developers to use their machine efficiently. If their architecture is neat enough, they can achieve both of those goals without even using a Cell-like architecture. I think they must have learned something from the PS2 and PS3 (both are hard architectures, but the PS2 didn't suffer much due to it since it was successful enough to force all developers to work as well as possible with it).

 



My Mario Kart Wii friend code: 2707-1866-0957

Kyros said:
You are correct that game devs will have many things that are better done on a general purpose core, but I really doubt that programming the CELL would get easier if they would mix multiple PPE with multiple SPEs. At the moment you have one core distributing work. In your proposed architecture you would have multiple cores distributing work to multiple SPEs? Sounds like a deadlock disaster waiting to happen.

 


Often, the SPEs are fetching work independently of the PPE.  It depends on how your application is structured, but it is not necessary for the SPEs to be entirely dependent on the PPE, other than the PPE initially starting the SPE threads.  In fact, it's recommended for the SPEs to initiate most of their own DMA transfers, because they have more slots.  One of the better ways to structure things is to queue up jobs for SPEs to do, and have the SPEs fetch those jobs.  This queue would already be controlled by a lock for thread safety, so a second PPE queueing up jobs wouldn't break anything.

IMO, if you're smart enough to make good use of the Cell as it is, then you'd be smart enough to be able to use more than one PPE effectively.



Griffin said:
HappySqurriel said:
slimeattack said:
If you want to see the difference between console and PC graphics, you need to have a good processor and a good graphics card. The latter costs equally much as a console.

 

I don't know about anywhere else, but in Canada you can pick up a mid-line Graphics Card (Geforce 8800GT) for about $150 (as low as $100 on sale), and a pretty good processor for about $150 (once again as low as $100 on sale), which enable you to play most new games at medium quality at a decent resolution ... certainly, it is not a complete system but people overestimate how expensive it is to produce a more powerful PC than the PS3.

 

Where are you finding at 8800Gt for 100-150.  I check out BestBuy and Future Shop each week and the 8800Gt 512mb has never gone below $200.  The flyers stlll have a 7600 card listed at $130.

 

8800 GT for $120:

http://www.newegg.com/Product/Product.aspx?Item=N82E16814150278

Took me less than a minute to find.



fkusumot said:

Makes sense to mention the Cell as that's something the RSX has been adapted for. The Cell is excellently suited to pre-process and post-process RSX related stuff. I think the PS3 paves the way for a Cell based PS4, I think game engines will be taking advantage of the PS4 specs much faster. Developers say there's a huge amount of untapped potential, for example Housemarque claims a 50% gain in graphics performance for their next PS3 game compared to Super Stardust HD (1080p @ 60 FPS, over 20,000 physics based colliding objects and a 100,000 particle effects on screen at once). I think the PS3 should not sell too much until the hardware costs are brought down and mass appealing games like Gran Turismo 5 or Final Fantasy XIII are available. The software profits can then compensate investment costs. The install base will increase like you can see at a healthy pace and production costs will come down further through improved middleware, this roadmap is perfectly sustainable for Sony. Creating simpler games distributed on the PSN is a cheap alternative for smaller companies. The Cell will not be maxed out for many years to come, more code will be moved over to the Cell's SPUs and there's endless room for optimisations. For games which push the PS3 far more technically wait for games like Resistance 2 and Killzone 2, a lot of advancements will have been made in terms of technology. It's the most impressive launch title for the US, but SPU usage was still limited compared to where they are now. Nice touches include very nice window breaking and some cool physics, good weather effects, super solid framerate, diverse weapons, bodies piling up, etc. Basically for yielding near optimal results the lack of branch prediction hardware for the SPUs does not stand in the way. The approaches needed to get great results out of the SPUs are needed on other processors as well to use them more efficiently. Branch hints and branch prediction serve the same purpose which is to tell the CPU ahead of the branch execution time what code to fetch. Some code is very unpredictable for the branch predictor, then there's the advantage towards the manual approach best used for the SPUs, resulting into gains on for other CPUs as well. The Cell in the PS3 will be mostly used for game and multi-media related functions, for double precision performance new versions of the Cell will become available. The point is that the SPUs are well suited for double precision despite this being sub-optimal. It's a challenge mostly from a game engine design perspective, but technically from a hardware perspective it can be done and the Cell is more suitable for this than other CPUs (note the near linear performance increase while tapping more SPEs).The SPUs are full processors and the SPEs are like little system themselves within which the SPU acts as a CPU. Just because the SPUs are really fast at DSP, GPU or vector unit related tasks does not inhibit the other potentials. It requires redesign, some stuff you currently do manually on the SPUs to achieve maximum efficiency which translates into better cache hits on other CPUs as well. You should write your code as parallel as possible (8 hardware threads and maybe additional software threads where this suits your game engine) and for ideal performance you stick to half- or single precsion formats (but the Cell can outperform other top CPUs at double precision as well, due to more processors on the chip). You can write any kind of code which can run on the PPU for the SPUs. However within the Cell design the PPE is meant to be the general overviewing manager and the SPEs the allround experts. We will see better super computers and more powerful efficient devices based on the Cell. IMO the inclusion of Blu-Ray technology was absolutely the right decision for Sony looking from the grand and long term perspective.
There are seven SPEs in use on the PS3. One is used by the OS (uses which won't draw performance from the other SPEs or PPE). I would rather compare the PPE to a manager who does do some additional work within a company himself apart from just telling others what to do (like in the real world, telling others what to do isn't very (proccessing) intensive work, compared to what needs to be done). The SPEs in the PS3 are like 7 talented hard working employees, who can work independently once they are told what to do. The SPE is a self-contained vector processor which acts as an independent processor. So, each SPE can perform multiple operations simultaneously with a single instruction. The Cell should be able to deliver much improved, AI, Physics, Procedural Synthesis, etc, nomatter what you may have read elsewhere.

The Cell processor offers innovative technology potentially allowing unrivalled performance gains. IMO the Cell processor is the most interesting part of the PS3, without it I would have been a lot less interested. - MikeB

 

 

That's the longest paragraph I've ever seen in my life.