PS3 not so broken after all...
by James Steele · 06/06/2006 (3:17 am) · 33 comments
Before I start with this, I make no appologies for being a firm supporter of Sony's hardware efforts on the PS3. I realise that extremely powerful and complex hardware does not by default mean you'll have a great games. But at the same time, I don't see anything wrong with making advances in this area.
I can deal with all of the anti-Sony sentiment going on right now. I mean, it's only natural. Sony came along almost a decade ago, and changed the games industry beyond recognition. It started a trend towards catering for the mainstream consumer, making video games something cool to play and no longer just the realm of kids drooling infront of thier TV in a darkeneded bedroom.
All of a sudden, it wasn't okay to make niche games. We now had to worry about market penetration, production values, and god knows how many other details we never really had to worry about before. No longer could we make games that we thought would be fun for us to play, hoping that other would feel the same too.
That had been reveresed. Now we had to make games that the lowest common denominator found fun, and hoped that we might want to play them too. All because of Sony. All because of thier push towards main-stream consumerism. Damn them to hell. They've ruined our industry.
So why does it surprise me that I see articles like =http://www.theinquirer.net/?article=32171]this? At a first glance, Charlie Demrjian seems to know what he's talking about. But after a little close inspection, you find that the guy is quite blantantly writting an article based on a knee-jerk reaction. His opening paragraph says it all;
"earlier on the flight to Japan, my row-mate said 'if you think that's interesting, wait till you see this. Cell is hurting, badly'."
At this point, I think it's fair to say that ol' Charlie has already made up his mind that the PS3 is a piece of decicated panda doo. He's shown a slide with some bandwidth numbers on them, one of which is horribly low. So what does Charlie do? Does he investigate further? Does he bother to find out the context of the slide? Does he do his job to invesitgate this fully and report without bias?
Nope. Good ol' Charlie instead writes a sensationalist article titled; "PS3 hardware slow and broken".
Maybe if Charlie wasn't so eager to bash Sony, he would have done some research and found that;
* Sony always refers to VRAM and "Local Memory" when dealing with the graphics chip.
* The 16MB/s transfer speed refers to transfering data from VRAM to Main RAM.
* There are very few occasions when you would actually want to do this with the Cell.
* The RSX provides a nice high bandwidth from VRAM to Main RAM anyhow...so why the need to duplicate this functionality in the Cell? Isn't the hardware already expensive enough?
I tell you folks, I'm getting a little sick of this. I understand people no liking Sony, but c'mon...has it gotten so bad that reporters are now jumping on the bandwagon too?
[/rant]
I can deal with all of the anti-Sony sentiment going on right now. I mean, it's only natural. Sony came along almost a decade ago, and changed the games industry beyond recognition. It started a trend towards catering for the mainstream consumer, making video games something cool to play and no longer just the realm of kids drooling infront of thier TV in a darkeneded bedroom.
All of a sudden, it wasn't okay to make niche games. We now had to worry about market penetration, production values, and god knows how many other details we never really had to worry about before. No longer could we make games that we thought would be fun for us to play, hoping that other would feel the same too.
That had been reveresed. Now we had to make games that the lowest common denominator found fun, and hoped that we might want to play them too. All because of Sony. All because of thier push towards main-stream consumerism. Damn them to hell. They've ruined our industry.
So why does it surprise me that I see articles like =http://www.theinquirer.net/?article=32171]this? At a first glance, Charlie Demrjian seems to know what he's talking about. But after a little close inspection, you find that the guy is quite blantantly writting an article based on a knee-jerk reaction. His opening paragraph says it all;
"earlier on the flight to Japan, my row-mate said 'if you think that's interesting, wait till you see this. Cell is hurting, badly'."
At this point, I think it's fair to say that ol' Charlie has already made up his mind that the PS3 is a piece of decicated panda doo. He's shown a slide with some bandwidth numbers on them, one of which is horribly low. So what does Charlie do? Does he investigate further? Does he bother to find out the context of the slide? Does he do his job to invesitgate this fully and report without bias?
Nope. Good ol' Charlie instead writes a sensationalist article titled; "PS3 hardware slow and broken".
Maybe if Charlie wasn't so eager to bash Sony, he would have done some research and found that;
* Sony always refers to VRAM and "Local Memory" when dealing with the graphics chip.
* The 16MB/s transfer speed refers to transfering data from VRAM to Main RAM.
* There are very few occasions when you would actually want to do this with the Cell.
* The RSX provides a nice high bandwidth from VRAM to Main RAM anyhow...so why the need to duplicate this functionality in the Cell? Isn't the hardware already expensive enough?
I tell you folks, I'm getting a little sick of this. I understand people no liking Sony, but c'mon...has it gotten so bad that reporters are now jumping on the bandwagon too?
[/rant]
About the author
#22
I mean....maybe you could interest small publisher (or a big one) if you have something pretty cool running on actual hardware.
As for the XB360...the only way you'll get anywhere close to doing indie dev on that thing, is if you have 20k hanging around doing nothing. The XB360 test stations were fairl similar to the dev-kits, until recently. Microsoft have now made them purely test-stations that can't be run from the dev environment (somebody correct me if I'm wrong on this...I have second hand info).
06/06/2006 (1:18 pm)
Well Ray, any standard Linux distro comes with C/C++ compiler, and the PS2 kit also comes with a VU0/VU1 assembler, full EE and GS docs. I imagine the PS3 kits will be something pretty similar, although I'm not sure what sort of steps people would have to go through to get stuff published, or distributed.I mean....maybe you could interest small publisher (or a big one) if you have something pretty cool running on actual hardware.
As for the XB360...the only way you'll get anywhere close to doing indie dev on that thing, is if you have 20k hanging around doing nothing. The XB360 test stations were fairl similar to the dev-kits, until recently. Microsoft have now made them purely test-stations that can't be run from the dev environment (somebody correct me if I'm wrong on this...I have second hand info).
#23
you can really feel peoples reactions.
06/06/2006 (1:33 pm)
Sony seems like such a joke to me, they're slow up to bat this time around. Just by reading all these postsyou can really feel peoples reactions.
#24
At that time, console games were basically arcade re-makes that really sucked unless you were 10 years old. Sony changed all of that.
They definitely deserve their place in history. But the PS3 is joke, IMHO. It is late. It is over-hyped. It is probably going to complicate the development pipeline by 10 fold, so they game choices will be thin and will probably be refactored versions of PS2 games, much like the 360.
In the end, I hate the corporate nonsense that has taken over the game industry. I have always been attracted to creating games because of the creative nature of the art. But companies like Sony want to control what I can release, who I can release it to, how I can distribute it, etc.. Screw that. Screw them.
I am a PC gamer, so my opinion is skewed. IMO, if its worth playing, its worth playing on my PC. I realize that the PC market is not as big as the console market, but I don't care.
If you have $600 for a PS3, spend it on a good nVidia graphics card and make your PC worthy of the currently available games.
That's my 2 cents.
06/06/2006 (5:15 pm)
I remember back when the PS1 came out. The source code for Wolfenstein 3D was just released and Doom was all the rage. I was reading every book there was on 3D games. Then I bought a PS1. It was rendering triangle meshes with textured surfaces. I was amazed.At that time, console games were basically arcade re-makes that really sucked unless you were 10 years old. Sony changed all of that.
They definitely deserve their place in history. But the PS3 is joke, IMHO. It is late. It is over-hyped. It is probably going to complicate the development pipeline by 10 fold, so they game choices will be thin and will probably be refactored versions of PS2 games, much like the 360.
In the end, I hate the corporate nonsense that has taken over the game industry. I have always been attracted to creating games because of the creative nature of the art. But companies like Sony want to control what I can release, who I can release it to, how I can distribute it, etc.. Screw that. Screw them.
I am a PC gamer, so my opinion is skewed. IMO, if its worth playing, its worth playing on my PC. I realize that the PC market is not as big as the console market, but I don't care.
If you have $600 for a PS3, spend it on a good nVidia graphics card and make your PC worthy of the currently available games.
That's my 2 cents.
#25
* Wii
- supposedly has a cheap devkit
- marketed for casual gamers
- deemphasis on graphics
* 360
- live arcade
- connections to GG
- mostly appeals to hardcore gamers
- shiny graphics
* PS3
- probably too expensive for casual gamers
- shiny graphics
- might be hard to program for
I think the Wii and the 360 are probably more ideal for indie devs. Even if Sony wants to help indies out somehow, at $600 will the biggest market for indie games even have a PS3?
06/06/2006 (7:35 pm)
The site ate my post, so I'll put it in bulleted list form :)* Wii
- supposedly has a cheap devkit
- marketed for casual gamers
- deemphasis on graphics
* 360
- live arcade
- connections to GG
- mostly appeals to hardcore gamers
- shiny graphics
* PS3
- probably too expensive for casual gamers
- shiny graphics
- might be hard to program for
I think the Wii and the 360 are probably more ideal for indie devs. Even if Sony wants to help indies out somehow, at $600 will the biggest market for indie games even have a PS3?
#26
I find your comment a bit strange, given that Microsoft and Nintendo have just as restrictive systems in place on what you can what you can realease, who you can release it to and where. I actually have to wonder what people base thier opinions on sometime.
Don't get me wrong, it's your opinion and you're entitled to it, but your reasons for disliking Sony baffle me when all the other manufacurers behave in the same way.
@Drew
I don't think the PS3 will be so hard to code for. The only thing that's going to be difficult to get a hang of, is the multi-threaded nature your code on the SPU's. Even then, that's not exactly difficult to grasp, as it mainly involves planning out your code before hand.
On the graphics side, the PS3 uses OpenGL ES and nVidia's fragment shader tools. This is a big change from the PS2, where you had to write your transform code in assembler and also create your own DMA chains by hand, which often resulted in hard to tace crashes if you didn't do it preoperly.
The Wii is nothing more than Gamecube II. The only real thing that's changed is the controller. I'm actually insulted that Nintendo feel that a higher clock speed and a fancy controller are a good enough reason to charge more cash, and are touting it as something new and special.
06/06/2006 (10:47 pm)
@SteveI find your comment a bit strange, given that Microsoft and Nintendo have just as restrictive systems in place on what you can what you can realease, who you can release it to and where. I actually have to wonder what people base thier opinions on sometime.
Don't get me wrong, it's your opinion and you're entitled to it, but your reasons for disliking Sony baffle me when all the other manufacurers behave in the same way.
@Drew
I don't think the PS3 will be so hard to code for. The only thing that's going to be difficult to get a hang of, is the multi-threaded nature your code on the SPU's. Even then, that's not exactly difficult to grasp, as it mainly involves planning out your code before hand.
On the graphics side, the PS3 uses OpenGL ES and nVidia's fragment shader tools. This is a big change from the PS2, where you had to write your transform code in assembler and also create your own DMA chains by hand, which often resulted in hard to tace crashes if you didn't do it preoperly.
The Wii is nothing more than Gamecube II. The only real thing that's changed is the controller. I'm actually insulted that Nintendo feel that a higher clock speed and a fancy controller are a good enough reason to charge more cash, and are touting it as something new and special.
#27
06/08/2006 (12:59 am)
Drew, you forgot to add 'F*cking Awesome overpowered Supercomputer' under the PS3 title ;)
#28
06/08/2006 (9:36 am)
What I find amusing, is that by the time the PS3 comes out, you will be able to build desktop PC's that are at least twice as powerful as the PS3. ;)
#29
If not, then kiss goodbye to having time-crtitical parallel processing in one application :op
I somehow doubt that any of the multi-core CPU's currently on the release schedule for November come anywhere close to a single Cell-CPU.
06/08/2006 (11:49 am)
Good luck with that Ray :) Are Intel and AMD suddenly going to manufacture dual-issue cores instead of sticking with the old out-of-order- execution business? Are they going to create a memory controller that the programmer has full control over? If not, then kiss goodbye to having time-crtitical parallel processing in one application :op
I somehow doubt that any of the multi-core CPU's currently on the release schedule for November come anywhere close to a single Cell-CPU.
#30
Supposedly AMD and intel are coming up with newer, four core designs pretty soon. Intel is planning to only release dual core CPU's in the very near future. So at this point there is no reason why a desktop PC wouldn't have dual cores, if you built it for gaming.
06/08/2006 (12:01 pm)
The video cards available for PC's are already more power than what the PS3 comes with. There have been a lot of arguments about how well the new multicore designs for consoles perform. The CPU's used inside of the Xbox and especially the Cell are nothing like a modern CPU inside of a PC. The Cell processor for example doesn't do branch prediction, instruction caching or any number of other optimizations an standard issue AMD or Intel chip does out of the box. You will get a dramatic decrease in the number of instructions per cycle on those platforms. I would like to see actual numbers comparing the Cell to a high end, dual core CPU. Not theoretical figures where they store all of the numbers they have to do calcuations on, within the small memory space that's allocated to each processor in the Cell. I am talking about real life applications, for example a game.Supposedly AMD and intel are coming up with newer, four core designs pretty soon. Intel is planning to only release dual core CPU's in the very near future. So at this point there is no reason why a desktop PC wouldn't have dual cores, if you built it for gaming.
#31
The Cell does infact do a form of branch prediction. I'm not sure about the PPE, but the SPU's do actually cater for branch-hints to avoid stalls. It's up to the programmer to tell the processor which is the most likely branch, and even if this isn't used; the stall is only six cycles unlike the extremely nasty mimimum stalls on the x86 compatible processors (especially the Intel IA-32's).
You don't by any chance write for The Inquirer under the pseudname of Charlie Demrjian? I find it hard to believe that I've stumbled across two people who will publicly make uninformed claims about hardware, without first doing some research, in less than a week.
And why not quote figures where the data is is held in SPU local store Ray? After all, the Local Store functions the same as L1 & L2 cache on the x86 family, and I'll bet a lot of quoted figures from Intel and AMD only time stuff while it's in cache.
If you want to talk about bus speeds though...each SPU has it's own memory controller running on a bus at something around 3.2Ghz on a 128-bit bus. I'm pretty sure that, like the most recent Intel IA-32 and -64s, cache refresh is a non-issue.
I remember an interesting talk with a former colleauge who's lucky enough to be working on PS3. Right now, they're not using every SPU due to inexperience in working in such a parallel processing envrionment. His idea was to reserve one or two spu's, along with thier associated local store memory, and write a software cache controller to complement the PPE's existing cache (as the local store memory is directly accessable by the PPE too)
This way he could work with two completely different areas of memory, safe in the knowledge that his reads are unlikely to screw up the cache lines and cause unnecsasary refresh on the PPE.
But really...this is getting off topic quite a bit. I'm pretty sure a CPU debate could go on for eons without anybody agreeing, no matter what figures were at hand. And as we all know....you can do a lot with a crap CPU, if you get your algorithm optimised for it properly who really cares about ultra low-level optimisations?
06/08/2006 (1:46 pm)
I have to point out that branch-prediction and out-of-order execution are not really optimisations Ray. If anything they are band-aid sollutions for a poorly thought out micro-architecture with extremely deep instruction pipes. The Cell does infact do a form of branch prediction. I'm not sure about the PPE, but the SPU's do actually cater for branch-hints to avoid stalls. It's up to the programmer to tell the processor which is the most likely branch, and even if this isn't used; the stall is only six cycles unlike the extremely nasty mimimum stalls on the x86 compatible processors (especially the Intel IA-32's).
You don't by any chance write for The Inquirer under the pseudname of Charlie Demrjian? I find it hard to believe that I've stumbled across two people who will publicly make uninformed claims about hardware, without first doing some research, in less than a week.
And why not quote figures where the data is is held in SPU local store Ray? After all, the Local Store functions the same as L1 & L2 cache on the x86 family, and I'll bet a lot of quoted figures from Intel and AMD only time stuff while it's in cache.
If you want to talk about bus speeds though...each SPU has it's own memory controller running on a bus at something around 3.2Ghz on a 128-bit bus. I'm pretty sure that, like the most recent Intel IA-32 and -64s, cache refresh is a non-issue.
I remember an interesting talk with a former colleauge who's lucky enough to be working on PS3. Right now, they're not using every SPU due to inexperience in working in such a parallel processing envrionment. His idea was to reserve one or two spu's, along with thier associated local store memory, and write a software cache controller to complement the PPE's existing cache (as the local store memory is directly accessable by the PPE too)
This way he could work with two completely different areas of memory, safe in the knowledge that his reads are unlikely to screw up the cache lines and cause unnecsasary refresh on the PPE.
But really...this is getting off topic quite a bit. I'm pretty sure a CPU debate could go on for eons without anybody agreeing, no matter what figures were at hand. And as we all know....you can do a lot with a crap CPU, if you get your algorithm optimised for it properly who really cares about ultra low-level optimisations?
#32
AMD Opteron (Deerhound) is expected to be released in H2 2007 on a 65nm process and running on a Socket F platform. Deerhound will be a 4 core CPU featuring a shared L2 cache and interfacing to Dual Channel Registered DDR2 memory.
Intel Xeon MP (Tigerton) CPU is expected to be released in 2007 as part of the Caneland platform. Tigerton is the replacement of the now abandoned Whitefield CPU and is expected to be a 4 core CPU possibly based on the Merom core and using Intel's CSI bus - akin to AMD's HyperTransport. Like Whitefield, Tigerton is being designed in Intel's design centre in India.
Intel Tukwila (formally known as Tanglewood, possibly cancelled) IA64 CPU is expected to be introduced in 2007. Tukwilla is the multi-core (4 core) successor to Shavano (now probably successor to Montecito) built on a 65nm process. Tukwila is expected to contain 32Mb L3 cache.
06/09/2006 (8:34 am)
Ray, the majority of the 4-core designs are not scheduled for relase until the 4th quarter of 2007. The PS3 comes out November 2006. AMD Opteron (Deerhound) is expected to be released in H2 2007 on a 65nm process and running on a Socket F platform. Deerhound will be a 4 core CPU featuring a shared L2 cache and interfacing to Dual Channel Registered DDR2 memory.
Intel Xeon MP (Tigerton) CPU is expected to be released in 2007 as part of the Caneland platform. Tigerton is the replacement of the now abandoned Whitefield CPU and is expected to be a 4 core CPU possibly based on the Merom core and using Intel's CSI bus - akin to AMD's HyperTransport. Like Whitefield, Tigerton is being designed in Intel's design centre in India.
Intel Tukwila (formally known as Tanglewood, possibly cancelled) IA64 CPU is expected to be introduced in 2007. Tukwilla is the multi-core (4 core) successor to Shavano (now probably successor to Montecito) built on a 65nm process. Tukwila is expected to contain 32Mb L3 cache.
#33
Intel Harpertown is the server version of Yorkfield, due to be released around 2009. Like Yorkfield, Harpertown is expected to be build on a 45nm process and contain 8 cores in a multi die package with 12MB of shared L2.
06/09/2006 (8:36 am)
And here is the 8 coreIntel Harpertown is the server version of Yorkfield, due to be released around 2009. Like Yorkfield, Harpertown is expected to be build on a 45nm process and contain 8 cores in a multi die package with 12MB of shared L2.
Associate Ray Gebhardt