Segmenting the memory will inadvertedly result in more data duplication so you'll need more memory too. I can see some logic for segmenting stuff like sound memory, but splitting main mem and graphics would seem like a step back - especially after 3 out of 4 consoles this generation used unified approach.
News about Rambus and the PS3
- Thread starter McFly
- Start date
megadrive0088 said:
Who knows? Micron hates Rambus so I'm sure they would really love to trump XDR (Yellowstone) and make money from Microsoft at the same time. Already there is the GDDR 2 and 3 standards which makes it seem like anyone can propose a standard. I guess it boils down to if Microsoft has a need for something better than DDR II. DDR II should be affordable by the time next gen consoles launch. They could go out and create a new standard called GDDR-X (the x represents the xbox console). The GDDR-X standard could end up being like a GDDR-4 or DDR III standard in disguise. Just slightly tailored for a Microsoft console.
If Microsoft does partner with Micron and Intel again, both Micron and Intel might want to use the console hardware to set standards that will cross polinate into the PC sector.
Link
Rambus marks finish line for new memory
By Michael Kanellos
Staff Writer, CNET News.com
July 10, 2003, 5:17 PM PT
Chip designer Rambus and its manufacturing partners said Thursday that their new high-speed memory will hit the market next year.
The XDR DRAM technology (formerly code-named Yellowstone) will transfer data at 3.2GHz initially, rising to 6.4GHz in the future--much faster than memory available now. In effect, that means XDR DRAM, which will be manufactured by partners Toshiba and Elpida, should ultimately be capable of transferring 100GB of data per second.
While XDR DRAM will likely face stiff resistance for acceptance in the PC market, it could find a home in consumer electronics and the communications industry, where memory based on Rambus' designs is used.
One of the early marquee devices to use the memory will likely be the next version of Sony's PlayStation. The Japanese giant has said it will couple XDR DRAM in the games consoles with the forthcoming Cell processor. Cell, a joint project between IBM, Sony and Toshiba, is expected to be used in the PlayStation 3.
Manufacturers Toshiba and Elpida stated that they expect to begin shipping XDR DRAM in 2004, with volume production coming in 2005.
Los Altos, Calif.-based Rambus designs chip interfaces, the portals that let chips connect to each other. The company designed RDRAM, a form of computer memory with a high-speed interface. While heavily touted, RDRAM proved to be expensive and unpopular. Even though RDRAM is faster than other types of memory on the market, few PCs come with it. Instead, PC makers and customers have opted for double data rate (DDR) DRAM.
Rambus has also sought to obtain royalties out of a number of high-tech companies through patent claims.
Rambus marks finish line for new memory
By Michael Kanellos
Staff Writer, CNET News.com
July 10, 2003, 5:17 PM PT
Chip designer Rambus and its manufacturing partners said Thursday that their new high-speed memory will hit the market next year.
The XDR DRAM technology (formerly code-named Yellowstone) will transfer data at 3.2GHz initially, rising to 6.4GHz in the future--much faster than memory available now. In effect, that means XDR DRAM, which will be manufactured by partners Toshiba and Elpida, should ultimately be capable of transferring 100GB of data per second.
While XDR DRAM will likely face stiff resistance for acceptance in the PC market, it could find a home in consumer electronics and the communications industry, where memory based on Rambus' designs is used.
One of the early marquee devices to use the memory will likely be the next version of Sony's PlayStation. The Japanese giant has said it will couple XDR DRAM in the games consoles with the forthcoming Cell processor. Cell, a joint project between IBM, Sony and Toshiba, is expected to be used in the PlayStation 3.
Manufacturers Toshiba and Elpida stated that they expect to begin shipping XDR DRAM in 2004, with volume production coming in 2005.
Los Altos, Calif.-based Rambus designs chip interfaces, the portals that let chips connect to each other. The company designed RDRAM, a form of computer memory with a high-speed interface. While heavily touted, RDRAM proved to be expensive and unpopular. Even though RDRAM is faster than other types of memory on the market, few PCs come with it. Instead, PC makers and customers have opted for double data rate (DDR) DRAM.
Rambus has also sought to obtain royalties out of a number of high-tech companies through patent claims.
Haha and a p4 should ultimately be capable of 1tflop . I don't go around saying it though. Lets be honest. The fasted they will put in the ps3 is mabye 4ghz mabye . Thats if yields are extremly good. Remember they need to have alot of ram for launch so this is going to be a big issue.
Brimstone said:
oh god... do we really need another thing named Something-X... or X-something....
i mean the last 2 years have been the worst, everything had an X attached to it.... only good thing was Xmen the movies.... and thats because they had the X first...
i mean, XXX
Jason X
the amount of games with X in them...
the amount of movies with X in them somewhere...
even rock bands (im sure i heard of some band x something...)
i mean X is the new 2....
Panajev2001a said:
Local storage and registers will to a large extent merely duplicate data already in eDRAM, which will to a large extent duplicate data in main RAM. Having all this on-chip storage is a BONUS, it's not a replacement for a high-capacity memory subsystem!
Blowing through 64MB of eDRAM, assuming the thing gets that much which isn't a certainty yet, happens FAST if you got 1Tflops of computing horsepower available... Things will be more problematic still if eDRAM is only half or a quarter of the speculated 64MB for the final Cell.Not all rendering; majority of textures and geometry will be in main memory.
Quite a bit yes, but for GPU especially it'll be duplicates of what's already in main RAM, in particular when it comes to textures, and we can't fit ALL of the data we need for a frame either. We can assume PS3 uses a minimum of 4x AA and 128-bit color, at HDTV res that chews up quite a bit of framebuffer space. Assume 1080 horizontal res at 16:9 aspect, I find we need appx 40MB for the floating-point backbuffer ALONE, unless I made a grave calculation error somewhere.
Add another ten megs for full-size 32-bit Z/stencil, plus another five for our display-sized 32-bpp buffers and we have just nine or so megs left for textures. You think a 2005 game can do with 9 megs of textures per frame?Still having the belief PS3 won't need lots of main RAM b/w? LOL!
Costs? What costs? Do you have an estimate of the "savings" in 2005 to implement a memory subsystem that will absolutely NOT!!!! be state-of-the-art by that time? Look at PS2. Every single raw spec was inflated to rediculous levels compared to competition. Do you honestly expect PS3 to launch in around two years' time with main memory that is slower than some of the solutions we already got TODAY? Man, I have to say you're OUT THERE!
It won't be any kind of SAVING, certainly not enough to enable Sony to put in a blue-ray player - WHICH WE KNOW THEY WILL DO ANYWAY. PS3 is the optimum launch vehicle to get the masses hooked on blue-ray, Sony will not miss this opportunity. Mark my words!
What makes you think it'll be only 256 megs? I expect no less than half a gig. PCs TODAY are generally delivered with half a gig as standard, even cheap ones. Top of the line systems come with a gig. In 2005, one gig will be mainstream. One gig will probably be listed as baseline requirement for longhorn, hehehe.
256MB is not enough for a next-gen system. In 2005, they won't MAKE memory modules that small to bring a multiple-channel memory setup to just 256MB.
Again, VISION, PEOPLE!
Also, let's not forget the power of propaganda. We know 66MPolys/s was bullshit when Sony revealed PS2 specs, but it was a massively high and suitably impressive figure. Of course people got affected by it. By giving PS3 a main memory that will be considerably lower in performance than what the competition will have at that time is not going to help them. Like I said, when it comes to the financial aspect, expect Sony to take it squarely in the nuts the first couple years or so. Cost will not be their primary concern just like it wasn't with the PS2. EE was a HUMONGOUS chip at .25u. Its die area was like Itanic-sized! Intel sells those things for like $5000, Sony selled theirs for less than a tenth. EE too was extremely large by the standards of that time both size-wise and transistor-wise, did that stop Sony? LOL! No.
They did it once, they'll do it again if it means trouncing the competition.
*G*
well (again), Sony clearly said theoretical 66million (or 75million) flat shaded polygons per second. which PS2 can theoretically do. where's the bullshit? ... we even heard from people from these boards that theoretically using both the VU's the amount is almost double that. of course flat shaded and of course in not-real scenarios, but what the hell, microsoft was pulling out hundreds of millions of polygons figures out of their ass as well...
Panajev2001a
Veteran
I will answer you later in full...
I remember a console that launched in 2000 with 32 MB of main RAM which yelded 3.2 GB/s of bandwidth...
Uhm... no, not possible... who would have launched with ONLY 32 MB of RAM in mid 2000, especially with a memory which was not faster than the top of the line PC's main RAM ?
Nobody, I tell you... nobody
I remember a console that launched in 2000 with 32 MB of main RAM which yelded 3.2 GB/s of bandwidth...
Uhm... no, not possible... who would have launched with ONLY 32 MB of RAM in mid 2000, especially with a memory which was not faster than the top of the line PC's main RAM ?
Nobody, I tell you... nobody
PC-Engine said:
he was being sarcastic.................. u know.... console launched in 2000 with 32 meg of ram.... PS2 anyone?.... still GCN in 2001 with 24Meg is even more of a confirmation....and XBOX as well, although its 64Meg are (were) a lot at least compared to the other consoles, it still was very little compared to PC's at the time...
Panajev2001a
Veteran
Incorrect, the DC had 16 MB of main RAM...
PC-Engine said:Umm...I know...just in case people like YOU didn't know
Very pompous of you to assume that only yourself are capable of recognizing sarcasm
BTW DC had 24 MB of RAM in 98...
maybe i should use short simple sentences...
panajev was responding to Grall's claim that next gen consoles will have at least as much RAM as PC's sold at the same time.
panajev said that EVERY console released has always had MUCH less RAM than PC's sold at the same time...
CONFIRMATION: PS2 launching in 2000 with 32Meg and GC launching with 24.....
dont think anything said was entirely incorrect.....
what's with the attitude sweetie?
er....... it confirms panajev's argument...?
london-boy said:
I never claimed that.
True, which WASN'T as much a problem in the past, because if console X had only half as much RAM as the average PC at the time of launch, it was a difference of maybe a quarter gig at most. If PS3 launches with 256MB in 2005 the difference will be THREE QUARTERS of a gig or maybe more, and when devs already whine and bitch about too little RAM in today's consoles you wouldn't expect Sony to go and do the same mistake again.
I expect no less than half a gig main RAM, just like I said in my post. If PS3 has 2x64MB eDRAM, 256MB is just 2x of its on-chip memory pools. 512MB is a much more comfortable 4:1 ratio.
...And devs were dissatisfied with GC in particular. Doom3 on XB will alledgedly feature load delays several times PER LEVEL, and D3 isn't even a next-gen game engine. It leverages DX7 level of technology fer crissakes!
I ask people to THINK, and look at the PROGRESS of this industry. You will REALIZE YOURSELVES that 256MB is far too little for 2005 and on. PS3 isn't supposed to exist ONLY in 2005 you know, it has to be a VIABLE DEVELOPMENT PLATFORM FOR YEARS AHEAD, maybe all the way up til 2010. Anyone thinks 256MB will cut it in 2010? Don't make me laugh. 512MB won't either by then, but at least it will be better.
Comes back to people lacking vision. These guys say, blah-blah too expensive for 2005 blah blah (as if they really have a clue apart from what they read in that EEtimes article on costs of semiconductors or whatever, but I digress). Well guys, it MAY BE too expensive for 2005, but it won't be for 2007 and onwards. Win the war, not the battles and all that, remember?
Oh, and btw... I expect to see PS3 with a harddrive also. Blue-ray may be rewriteable and all, but its read/write speed is slow and access time atrocious. Also, data cannot be shared between multiple titles if everything is saved on the blue-ray disc. Once you remove the disc, you remove the data. Harddrives are much smarter that way.
*G*
[/b]
grall baby, i wasnt agreeing with Panajev... i was merely explaining to our little PCengine what was said....
i also think ps3 will have a hard drive for caching purposes, thinking BlueRay will sort out all of ps3's problems is very naive... still, some people dont even think ps3 will have BlueRay......
dont know, some people still think PS3 will launch with like 128M of Ram, a CPU and GPU not based on CELL, be little more than a Xbox.. i could go on and on and on.....
still, just the fact that ps3 will be around for years after 2006 doesnt mean much, look at ps2.... it lives and it will live even after ps3 launches, with its tiny little 32M Ram and all the rest......
having vision is good (i've got 20/20 BTW) (sorry i was trying to fit this one in since the beginning of the thread) but raising your expectations too high might prove fatal.. look at what happened to poor little Chap...
i also think ps3 will have a hard drive for caching purposes, thinking BlueRay will sort out all of ps3's problems is very naive... still, some people dont even think ps3 will have BlueRay......
dont know, some people still think PS3 will launch with like 128M of Ram, a CPU and GPU not based on CELL, be little more than a Xbox.. i could go on and on and on.....
still, just the fact that ps3 will be around for years after 2006 doesnt mean much, look at ps2.... it lives and it will live even after ps3 launches, with its tiny little 32M Ram and all the rest......
having vision is good (i've got 20/20 BTW) (sorry i was trying to fit this one in since the beginning of the thread)
but raising your expectations too high might prove fatal.. look at what happened to poor little Chap... zidane1strife
Banned
For a while… pcs will slowly move to better and better memory… and after a while only the box will be using it.
With embedded ram the ps3 gpu will have far more b/w than any non-embedded pc solution for the following years… heck if the GS is an example of what’s to happen… It might even surpass the 2010 pc gpu’s b/w… assuming they don’t feature embedded solutions of course.
Well, we don’t know… It could be a real shocker.
I think when all is said and done 512MBs-1Gig-the vertex data alone should take quite an insane amount of memory-(unless it’s really expensive, if so it will be less)
When the ps2 launches we’ll be at several gigs of ram, so it’d certainly be significantly less.
I second, say a gta type game, you could have 100+ different models, with each being 100k+polys at highest lod, 100s of those will be on the streets(at lower lods), but the data for them and for all the 1000s of objects hlods should be kept there for quick access.
Panajev2001a
Veteran
Sorry mr. Senior MPU Engineer with DEC experience... but if you are telling me that 512 pins would not make the motherboard more expensive than 256 pins I do not know what to tell you...
You have logic points, which I might or not disagree with, I do not think there is a need for trying to flame people like that...
But, back to the topic at hand...
Well, who tells you that they will both have 64 MB of e-DRAM ? I hope they can pack that much, but seeing the optimistic numbers they put in their CMOS5 ( 65 nm ) related PRs I think 32 MB might be more probably ( the CPU would also have 4 MB of SRAM accounting all the Local Storages though ).
32 + 32 = 64 = again your nicer 4:1 ratio...
There is one concept you have to catch and I know you are more than smart enough to have already thought about it... who is saying that we cannot keep the increase in bandwidth needed for the increasing polygon data and texture data cannot be contained ? Who is saying that most of the e-DRAM and the local SRAM is necessarily storing for the most part merely a sub-set of the external RAM ?
You are treating it too... uhm... how do we ignorant people call it... "inclusively" perhaps ?
I see with the processing power and on chip bandiwdth Cell has that we might see for the first time micro-polygon ( REYES-like ) based renderers gracing real-time 3D on game consoles... I plan to see more enphasis on Shader Programs and procedural texture creation on the fly rather than just immaging the current paradigm multiplied by 10x... and then you nag on me for lacking vision, with all the respect "hello pottle, meet kettle".
The increased processing power might let us finally do things without TONS of pre-calculated tricks... we could at least reduce their usage in real-time 3D applications.
How would data be processed in this kind of architecture if we look at the big picture...
We would have streams of data that would help refill CPU and GPU dynamically ( unless each frame we consume and waste more data from e-DRAM, that we are not going to re-use, than what main RAM can provide us we should be fine )... let's think on vertex data...
What would happen to our polygon data ( let's think about using subdivision surfaces... and yes we could do deferred processing [depth sort the surfaces and subdivide only the visible ones] to reduce the amount of "slicing and dicing" needed in the next stage ) ?
Well our stream would be loaded by the CPU and the result of processing would be a stream that would probably be decisively bigger than the incoming one ( at the end that processed stream would end up sent to the GPU through a separate chip-to-chip connection, probably Redwood based )... the transformed stream of micro-polygons cannot be sent yet to the GPU, quite a processor intensive part... Shading
With this kind of rendering approach ( micro-polygons + enphasys on Shading programs and not pre-done textures ) we will actually mitigate the increased enphasys on external rendering bandwidth provided that we have enough e-DRAM and SRAM on the chips to allow a good amount of space for processing data locally...
Think about what people said about normal maps and lighting... why do we need those pre-calculated maps if we can do those calculations in real-time ?
Why do we need tons of texture layers if Shaders can be used to procedurally create several of them ?
Why do we waste space with ultra detailed polygonal meshes in main RAM if we can send "space" optimized ones ( NURBS, sub-division surfaces ) and we can "expand" them on chip ?
With 32 MB of e-DRAM on CPU and 32 MB of e-DRAM on GPU ( plus GPRs and LS ) and 1 TFLOPS of processing power you should be able to move to such a rendering approach...
BTW, you point out that the CPU for example will hold data it buffers from memory thanks to the e-DRAM ( that the CPU can freely address ) then you fail to connect it to the fact that it would also mean reduced main RAM contention with the GPU...
PlayStation 3 will probably once again use a ~hybrid UMA approach as the PlayStation 2 used, but if you notice the supposed local memory on each processor has grown a lot ( and the main RAM would still be 8x faster at 25.6 GB/s )... the point of Cell and the use of e-DRAM was to minimize the bottleneck caused by external memory speed and the cost involved in pushing it to speeds that would be acceptable...
Also regarding complaints of current developers... Texture Compression says anything to you ?
The CPU ( 1 TFLOPS class beast ) would have 1,024 GPRs ( 16 KB of space ), 4 MB of SRAM based Local Storage and 32 MB of e-DRAM...
The GPU ( if based on Cell like the patent theorized ) would have 32-64 MB of e-DRAM ( IMHO they could get away with 32 MB ) and 512 GPRs ( 8 KB of space ) and 2 MB of SRAM based Local Storage...
The total would be... 70+ MB and if we think about 64 MB on the GPU this climbs to 104+ MB and if we think the CPU too has 64 MB of e-DRAM this total grows to 136+ MB...
136.024 MB transferring data internally at speeds like 100+ GB/s ( for 1,024 bits e-DRAM that would require a clock-speed of at least 780 MHz which is NOT low for e-DRAM )... I do not think that would be a small amount...
136 MB + 256 MB = 392 MB
I would not call that amount bad at all...
Let's think that they can push e-DRAM to 100 GB/s... 25.6 GB/s would be less than 1/4th...
Even thinking about VRAM bandwidth for the GPU... 100 GB/s would still be more than 2x the bandwidth the current GS has and the amount of memory would be 16x over the current GS.
The current GS gets its data ( its only connection to the rest of the machine ) through a 64 bits pipe yelding 1.2 GB/s
This is 40x less than the total bandwidth available on the GS.
1/4th or 1/40th... I think I prefer 1/4th
But, let's be fair... 25.6 GB/s is split between two chips as the 3.2 GB/s on PlayStation 2 was as well ( and to make things worse the EE did not have e-DRAM and as much Local Storage to be less dependent on that bandwidth )...
100 GB/s / 12.8 GB/s = 8
So we have 1/8th now, but 1/8th is still better than 1/40th
Panajev2001a
Veteran
Slow speed ? 9 MB/s ( sustained speed, non burst speed ) available in 2003 and 18 MB/s projected for 2005 ?
You have a different idea of slow from the one I have...
We can still use main-RAM to buffer data that will need to be saved in the Blu-Ray disc anyways...[/quote]
