There's been some more sleuthing (Ha! Legitimately unintentional pun, we swear) on the Finnish magazine that had the Splinter Cell: Conviction screenshot scans from Friday. Apparently the interview has some spiffy details about the enemy AI, Sam Fisher's ability to interact with his environment, and most deliciously, Ubisoft giving the PS3 some flack.
First, they say that Conviction being exclusive allows them to accomplish things on the 360 that couldn't be done if the title was multiplatform. They even go so far to say that if the title was a PS3 exclusive, they still would not be able to achieve some of these feats. Ouch ... coming from a well respected developer like Ubisoft, that's gotta hurt.
41. Ok I see a lot of arguing about the architecture of the 360. Here is a blog I posted to one of my friends highlighting benefits of the 360's architecture over the PS3 and the positives of xbox's design. Though the info is accurate as I can tell, it is geared towards being pro 360 so keep that in mind. My friend was suppose to do a counter article but never did it so I cannot post a counter and do not feel like researching it at this time. Yes it's a lot but I'm an engineer student and love this stuff. I use citations as well so check out the articles listed if you want to learn more. Take it as it is as I do not outline many of the PS3's benefits. I will say this though, the cell will not fully realized in the PS3's lifetime. IBM, part creators of the cell, cannot write an operation to 100% efficiency of the architecture. They can get the SPE's to 98% efficiency using a optimized parallel Matrix Multiplication, which is extremely linear and very unlike the majority of code used in a game. The system as a whole is less efficient than the 98% stated. Designing a game to come close to said efficiency will never happen. It will certainly improve from where it currently is but designing such a game would take more time and money than any company would ever put into it. Multithreaded operations have been giving engineers and mathematicians are hard time for the last 20 plus years and the cell is harder to work with than what they have been dealing with. It will get better but I believe cell is ahead of it's time. On to the reason I posted. Get ready for a long one.
I think the microprocessors are the most misleading aspect of each machine so I will go through those first. The one thing I hear over and over is the Floating Point Performance of the cell, 204 GFLOPS. Yes this is an impressive number but a few things must be taken into consideration. That number is for a full Cell processor where the1 Power Processor Element (PPE) and ALL 8 Synergistic Processing Elements (SPE's) are active. In the PS3 1 SPE is disabled to improve yields in the event one doesn't work and another SPE is dedicated solely to running the OS of the PS3 and in some cases the OS can and will borrow an extra SPE. Each SPE contributes approximately 25.6 GFLOPS of performance, so factoring the loss of two SPE's we arrive at 152.8 GFLOPS of theoretical performance ( dpad.gotfrag.com/portal/story/35372/?spage=4 ). That's an almost 25% performance hit right of the bat, ouch. Not to mention that IBM has yet to get hit that 204 GFLOPS with a full cell processor even using specialized coding designed for the sole purpose of making that number as high as possible. This meaning of course games will never come close to such theoretical values, though the PS3 should definitely outperform the 360 (115 GFLOPS) in this area. Floating point performance also is not as important as one might think. The PS2 had twice the power of the xbox's celeron in this regard and it certainly didn't seem to hurt the xbox graphically or otherwise now did it. An interesting quote, "Based on measurements of running next generation games, only ~10-30% of the instructions executed are floating point. The remainders of the instructions are load, store, integer, branch, etc." - www.rojo.com/story/My3j9I-kR-N8Jxw_. MS is aiming for a new kind of coding, namely procedural synthesis, to make up a lot of their games and is getting away from these inefficient and archaic means of programming certain things, textures especially. Such methods or much more efficient on space and has the benefit of creating a more dynamic and realistic background, though I'll explain more on this later.
This though is one of the reason, as well as MS is using more advanced compression schemes ( just like they are in HD movies, VC-1 vs MPEG2) that the PS3 will definitely need a bluray drive, whereas very few games, unless ridden with a good amount of High Def CGI scenes or poor coding should go over one DVD. The Final Fantasy series would probably use at least two DVDs. As you can see I listed integer performance as important, which is good for the 360 as it has over 3x the integer performance of the cell. See those loads and store times I listed up there? The cells SPE's do not have direct access to the memory which makes data storage tight especially when you consider the meager 512kb cache being shared by the PPE and SPE's is used up. Granted there is a performance hit for using the RAM I would suppose it would be nice to have the option. The 360 has 1 MB of cache shared between the 3 cores, which is cache coherent allowing any core or the GPU to pull data from the cache, whereas the PS3 has 8 processors sharing that 512 KB of cache. This will make storing data much more difficult and certainly be a nice challenge for programmers. Game programming for the cpu in general relies much more on the calculation of data, running usually nonlinear code, AI or collision detection than the type of coding the Cell was designed for. Granted the cell does have an on board memory controller which is very nifty and a saving grace of the cell, the latency obtained by there cache setup is going to make it much less efficient than the 360 when it comes to accessing needed data for processing and for the type of code used in games is going to slow down the processor significantly as it cannot read and write the data to efficiently to the cache as it will be filled to the brim in many instances.
On to branch prediction which is going to be quite an extensive bit of code used for any game. Both processors actually inefficient at this compared to the one is your pc, unless it's a dinosaur (Pentium 1 was the last strict order cpu). When an error in data handling occurs, which is inevitable, the PS3 is going to have a more difficult time dealing with said scenario. This just creates more latency and is going to further hinder the performance. One of the worst parts of both processors is they are not out of order execution processors. This hurts AI and other logic oriented coding more than anything and makes both much much harder to write code for as the programmers have to delegate instructions meticulously to the cpu otherwise you are going to get a huge latency issue which corresponds to a huge performance hit which would be worse on a PS3 SPE. The 360 does have a specialized 128 vmx (designed for MS) unit and a 4kb branch history table that at least keep it from being a full out strict unit cpu and is something that will greatly benefit the programmers and should dramatically increase performance compared to the cell when dealing with loops and if statements and the likes of that. Here is a link to some pretty graphs and stuff if you don't believe me on what I'm saying xbox360.ign.com/articles/617/617951p3.html. One more advantage of the 360's cpu's cache, it can be accessed by the gpu which means that RAM (huge bottleneck) can be bypassed completely and makes procedural synthesis (PS) run extremely efficiently. Both cores are going to be difficult for programmers as multi threaded applications development has been going on for over 20 years by some very smart people with not a whole lot of breakthroughs in getting a linear benefit out of them with besides a few exceptions like media encoding. Gaming requires lots of logic, branch predictions and out of order computing that both processors aren't particularly great at, though the 360 should have a decisive victory ( Assassins Creed developers reported the 360 version will have better AI even though it was originally a PS3 exclusive, check www.joystiq.com for link ). The 360 processor is also geared more towards Procedural Synthesis and I'll explain to you how awesome this in after I get down with the GPU as they are related.
Ok, onto the GPU. This is where things get interesting. Now while CPU admittedly does require a good deal of speculation and drawing conclusions based on what I know (and others) about programming and hardware, there is a lot more info on the GPU and can easily be correlated with others produced my NVIDIA and ATI, though more so NVIDIA. A lot of tech journals and blogs and even a NVIDIA spokesperson ( www.theinquirer.net/?article=25838 ) have said the RSX GPU is just a less powerful 7800GTX. This means that you are already getting dated technology in your PS3 as since NVIDIA has launched a 7900 series and currently offer a 8800 series GPU. So your the PS3 is only two generations behind what is currently available for PC the second it debuted. They even recently change the core frequency from 550 MHz to 500 MHz, the same as the Xenos and the Xenos was more powerful than the RSX before it took the hit in clock speed. The Xenos GPU on the other hand uses technology that is yet not available for any GPU on the market today and even some that will not make it into this upcoming generation, like its eDRAM. Talk about cutting edge eh? The PS3 uses DX9 and Shadermodel 3.0 capabilities just like it's 7800 counterpart where as the Xenos uses many DX10 features (unified shader support, shared ram etc ) and Shadermodel 4.0 abilities and the only thing keeping it from being designated as such is missing a few PC abilities not needed for a console. Advanced features of such architecture include a unified architecture for one. The Xenos has direct access to the CPU's 1 MB of cache and RAM, though the unique ability to bypass the RAM as mentioned above, a bottleneck of consequential magnitude I assure you, makes rendering and procedural synthesis run much more efficiently as well as general communication between CPU and GPU takes a huge jump in speed. Not to mention that the 360 has custom D3D and Dot product instructions and also the Xenos handles vector5's or 5 operations per cycle where as the RSX uses verctor4's or four operations per cycle. You add that up and the Xenos is a beast compared to the RSX in vector and computations in general and more flexible as it can offload certain calculations that would normally be designated for the CPU. Many of these features are only available currently on the 8800 NVIDIA cards and upcomg R600 core from ATI.
Onto ALU's (arithmetic logic unit) of the GPU. Responsible for doing vector calculations, integer arithmetic, and bitwise logic operations. The RSX has 52 ALU's (24 pixel shading pixel pipes capable of two operations per cycle and 4 vertex pipes capable of 52) to the Xenos 48. However the Xenos is a unified architecture which means that any of the ALU's can be
used for either vertex orpixel shader operations where the RSX has twelve of thirteen ALU's devoted to pixel and rest to vertex. So if a scene requires mostly vertex shaders the pixel shaders of the RSX must remain idle where the Xenos can go full out. The best part of it is the developers do not have to designate what pipes do what as the Xenos does it for them, no special programming required. The Xenos can run at full steam no matter what scene you throw at it unlike it less competent counterpart RSX. Let us do some math shall we. Lets look at how much GLOPS each GPU can do. It is going to be more relevant here than it is on CPU front. PS3 would be 52 ALU ops * 4 floats per op *2 (madd) * 500 MHz = 208. GFLOPS which is less than the Xbox 360's 48 ALU ops * 5 floats per op * 2 (madd) * 500 MHz = 240 GFLOPS. MADD stands for multiply add so you know. But wait there is more of course. For shader operations per clock, Sony is most likely counting each pixel pipe as four ALU operations (co-issued vector+scalar) and a texture operation per pixel pipe and 4 scalar operations for each vector pipe, for a total of 24 * (4 + 1) +(4*4) = 136 operations per cycle or 136 * 500 = 68.0 GOps per second. Given the Xbox 360 GPU's multi-threading and balanced design, you really can't compare the two systems in terms of shading operations per clock. However, the Xbox 360's GPU can do 48 ALU operations (each can do a vector5 and scalar op per clock), 16 texture fetches, 32 control flow operations, and 16 programmable vertex fetch operations with tessellation per clock for a total of 48*2 +16 + 32 + 16 = 160 operations per cycle or 160 * 500 = 80 GOps per second. In both instances you get some serious benefits with the 360. Again, the Xenos is quite possibly the most efficient GPU in existence because of its unified shaders and it is realistic to expect it to run very close to 100% efficiency while the RSX will never be able to do such. So in real world performance the gap in above numbers should definitely widen. When you take into account that the GPU and CPU can communicate directly over RAM and even better yet cache, you are looking at an incredibly fast and efficient architecture.
Now we'll talk about eDRAM. This is something else that the Xenos has and will not be used on PC's for a while as reported that it is hard to do on a non-closed system and is best suited for consoles. The Xenos has 10mb of eDRAM connected to the GPU through a 256 GB/s frame buffer bus (Higher than the 8800's and 5x that of the RSX). Damn is right. For starters this allows 4xAA (Anti Aliasing which eliminates jagged lines and produces an overall sharper image), HDR (High Dynamic Range lighting, which creates more realistic lighting effects), and Z buffering (determines depth of objects) to all be done with almost zero impact on performance. With such a huge bus the eDRAM can take the image, enhance it and push it wherever it need to go with literally almost no negative side effects and in fact frees up RAM for the GPU and CPU further widening it's lead over the PS3 in this regard. All 360 games are required to take advantage of this. So obviously the 360 has more bandwidth than the PS3. It also uses it more efficiently. For example to get 720P with 4XFSAA on traditional architecture would require 28MB worth of memory. On the 360 only 16MB is required. There are also features in the 360's Direct3D API where developers are able to fit 2 128x128 textures into the same space required for one, for example. So even with all the memory and all the memory bandwidth, they are still very mindful of how it's used. ( dpad.gotfrag.com/portal/story/35372/?spage=8 ). The Xenos has so many DX10 features that make it a much more advanced GPU. The RSX was out of date a while ago where there is still nothing like the Xenos in the pc market.
The 360 also has more memory for textures. The RSX directly has 256 MB of ram to work with and another 256 or XDRAM that can be accessed if need but at a performance penalty. The PS3 OS take up 64MB of XDRAM used by the CPU and 32MB of the GDRAM used for by the GPU, whereas the Xenos can use the full 512 MB of RAM with the exception of the 32 MB being used by the 360's OS. Besides the fact the 360 is more efficient with RAM as said above, it has more of it to work with and does not get that slow down if it need to access more than 256, which is the case in Gears of War. The 360 has much more texture memory to work with even if it wasn't more conservative with it.
Bluray is another feature that has been touted by PS3 fans. The 2x Bluray drive has a constant read speed of 9MB/s because it is CLV (Constant Linear Velocity) where as the 12x DVD drive average read speed wit be 12.1 MB/s at least. 16MB/s for outer DVD and 8.2MB/s ( (16 + 8.2)/2 = 12.1 . So the 360's drive is over 33% faster which will help drive load times down. Not to mention that the 360 has more cache to stream content greatly improving load times and making them less frequent and on to of that if you have a full 25GB bluray disk you have a shit ton of loading to do. No load times in gears of war and I loved it.
Final topic, Procedural Synthesis. This is the technology that if implemented properly will truly make the 360 leaps and bounds above the PS3. Instead of storing textures and reoccurring objects like trees and buildings, it can generate them on the fly and dynamically with special algorithms. So my version of the game will look different than yours due to the dynamic rendering of the objects. This cuts down or needed storage capacity tremendously. Just look at this game .kkrieger ( en.wikipedia.org/wiki/.kkrieger ), it only take up 97KB of data. That is insane compared to what current methods would need. For another look at what it can do read about this game SPORE ( en.wikipedia.org/wiki/Spore_%28video_game%29 ). The foliage in Oblivion is all rendered via PS and looks incredible. This technique holds so much promise to revolutionize graphics. MS has filed for several patents for techniques for using the method which apply to the 360 and their next gen system. This will make games take up less space, look better and more varied and create a more unique experience for gamers. The way the CPU works, the GPU runs, and the way they communicate with each other are aimed at taking advantage of PS. That's likely one reason why the Xenos can do so much calculating for a GPU. Current ATI x1000 series GPU's can and are used for folding at home ( Calculates what a protein will do under certain circumstances and was until recently only done with CPU's). Once developers can get at least decent with PS we are in for a treat and will see things much beyond what a PS3 can bring us as far as what I know tells me, but I don't know it all for sure.
Posted at 12:37AM on May 8th 2007 by jaysins