ATI Talks Lithography
- Thread starter Vince
- Start date
Megadrive1988
Veteran
I think that only 128 MB of external main memory (Rambus XDR) in PS3 is absurd. The reported amount is 256 MB anyway. And I think even that is kinda skimping. I was banking on at least 512 MB, which may or may not happen depending on what developers want, and what Sony decides. PS-X / PS1 main memory was originally 1 MB but developers asked for 2 MB and Sony gave it to them. thus the PS-X/ PS1 had 3.5 MB of memory in total.
(2 MB main + 1 MB VRAM + 0.5 MB audio)
So, for PS3, there will be at least 256 MB of XDR main memory (external) plus 32~64 MB of eDRAM on the Cell-CPU. plus a further 32~64 MB of eDRAM on the GPU. Therefore the combined amount of eDRAM for PS3 will be between 64 and 128 MB (CPU+GPU). It will probably be cut down to more like 64 MB eDRAM total for the CPU+GPU though. in other words, the minmum figures I mentioned above. (they're not my own figures anyway)
So 320 MB grand total (64 MB eDRAM plus 256 MB external main memory) is the absolute smallest figure the PS3 will have.
Actually I am not even counting several MB worth of SRAM / Local Memory.
The above is probably what the real PS3 will have. below, is what I would LIKE the PS3 to have, and is therefore 'just for fun'. I give two sets of figures
semi-realistic memory configuration
512 MB ~ 1 GB main memory
128 MB eDRAM on CPU
128 MB eDRAM on GPU
impossible dream memory configuration
2-4 GB main memory
512 MB eDRAM on CPU
512 MB eDRAM on GPU
(2 MB main + 1 MB VRAM + 0.5 MB audio)
So, for PS3, there will be at least 256 MB of XDR main memory (external) plus 32~64 MB of eDRAM on the Cell-CPU. plus a further 32~64 MB of eDRAM on the GPU. Therefore the combined amount of eDRAM for PS3 will be between 64 and 128 MB (CPU+GPU). It will probably be cut down to more like 64 MB eDRAM total for the CPU+GPU though. in other words, the minmum figures I mentioned above. (they're not my own figures anyway)
So 320 MB grand total (64 MB eDRAM plus 256 MB external main memory) is the absolute smallest figure the PS3 will have.
Actually I am not even counting several MB worth of SRAM / Local Memory.
The above is probably what the real PS3 will have. below, is what I would LIKE the PS3 to have, and is therefore 'just for fun'. I give two sets of figures
semi-realistic memory configuration
512 MB ~ 1 GB main memory
128 MB eDRAM on CPU
128 MB eDRAM on GPU
impossible dream memory configuration
2-4 GB main memory
512 MB eDRAM on CPU
512 MB eDRAM on GPU
nonamer said:
Been trying to see myself what STI is using. From what I've seen, it's a derivative of IBM's 11S process, which is partially depleted.
This also fits in with Toshiba's work on an eDRAM-on-SOI cell that's capacitorless by manipulating the FBE that's commonly seen in PD-SOI. My thinking is Toshiba's R&D expanded on the 10S technology liecensed from IBM back in 2001. But, I'm not 100% sure... maybe 85%.
Vince quite funny . I don't see you complaining about the first person to troll your thread and post about sony products and processes in it. You only complain about me who responded about said company in regards to another post. Stfu . IF you have beef with me report it to dave. I don't see you complaining about anyone else that has highjacked your thread with regards to other products. If anyone "highjacked your thread it was Panajev2001a who brought up sony and its process which has nothing to do with ati.Vince said:
I don't see where you get this from. in 2005-2006 90nm is what ati will be using just like every other gpu maker in the pc sector. They don't have acess to greater fabs because they out source production. They also don't have billions to make thier own fabs. Perhaps ms will pay for a very advance fab and we might see 65nm. But there is no reason to do that. They don't need that advance process . It seems to me Ms will be going with 90nm for initial launch and use 65nm and 45 nm for cost savings later on and sony will be pushing 65nm and will move to 45nm for cost savings. Its hard to say which one will work best. But its most likely that the gpu on the xbox will become cheaper quicker than the cell chip. If ms chases the tech .nonamer said:This pretty much confirms my suspicion that ATI is the low-cost, low-risk GPU provider for Microsoft. If they are wait till 2005 for the 90nm, then frankly I don't see them very competitive in terms of power (probably comparable, like the NGC is to the XB, but you can't simply say the GC is equal/better). As it stands now, even if Cell is a lousy design, PS3 still has probably won the powergame already simply due to brute force design and process investments. I saw MS as a potential challenge, but unless a major change in their PR then I can't see MS seriously challenging Sony anymore.
Panajev, I'll bet you a $1000 that the PS3 will be out on 65nm. 45nm is waaaayyyy out there, like 2007/2008. There are major lithography problems at 45nm and leakage on that process could make SOI scream "mommy!"I am certain that S/T/I? will release at 65nm, maybe lose [lots of] money, and move to 45nm later when it is ready.
Vince, is Sony's SOI PD or FD?
I'm not sure I see much surprise in this thread at all since this is exactly how these thing have played out for a long time in the customer fab area.
Offtopic, but I just wanted to vent it.
There was an _very_ interesting article on arstechnica on the new SUN chip interconnect stuff. The same post included a pointer to an lengthy explanaition on Moore's law, the original paper. The chip interconnect is a holy grail, because that means they can use more of each wafer if they use more and smaller dies to create a larger processor. This isn't new, and IBM has had these grid-things on multiple die packages a long time. And upon reading the paper it is very evident that the way to break Moores law is to do exactly this, produce many many small dies, interconnect; and it will still be cheap. Have a wafer with only APU's; and it will be cheap since there is less fault rate with that. Oh, go read it. It is still on arstechnica. here. http://arstechnica.com/archive/news/1064605458.html
There was an _very_ interesting article on arstechnica on the new SUN chip interconnect stuff. The same post included a pointer to an lengthy explanaition on Moore's law, the original paper. The chip interconnect is a holy grail, because that means they can use more of each wafer if they use more and smaller dies to create a larger processor. This isn't new, and IBM has had these grid-things on multiple die packages a long time. And upon reading the paper it is very evident that the way to break Moores law is to do exactly this, produce many many small dies, interconnect; and it will still be cheap. Have a wafer with only APU's; and it will be cheap since there is less fault rate with that. Oh, go read it. It is still on arstechnica. here. http://arstechnica.com/archive/news/1064605458.html
What does MS have to do with the speculation that the PC R500 part may be built using a 110nm process? Haven't we already been over the points of IP licensing? It seems, as well, that if R500 is still scheduled for 2004 then there is, unless my maths fails me, another year before 2005?
The point being that semicon's using customer fabs are always behind, those that invest heavily in their inhouse processes, and this has been shown to be the case even when NVIDIA was pushing "the bleeding edge" (for them).
The point being that semicon's using customer fabs are always behind, those that invest heavily in their inhouse processes, and this has been shown to be the case even when NVIDIA was pushing "the bleeding edge" (for them).
jvd said:
The issue isn't the bandwidth (well, that is an issue, but a different one), it's the space. Personally, I think even 256MB of main RAM is too little, unless they are planning on going with some sort of memory model similar to the NGC. Even if we will have enough power to move towards more procedural methods, space can still be taken up rather quickly.
Megadrive1988
Veteran
I thought that the original target date for R500 was 2004. That was sometime back in 2002. now with both ATI and Nvidia saying that there will be a much longer time between major new architechures, the R500 is due in 2005.
And also, the R500 that is being worked on now, is based on the original concept of the R400. (with R420 / Loki being based on a souped up R3XX)
Which means the original R500 spec is now being moved to R600 ( right?)
That is my current understanding anyway, based on everything I've read on B3D and elsewhere. obviously what I said above could be totally off, and I have no way of knowing.
And also, the R500 that is being worked on now, is based on the original concept of the R400. (with R420 / Loki being based on a souped up R3XX)
Which means the original R500 spec is now being moved to R600 ( right?)
That is my current understanding anyway, based on everything I've read on B3D and elsewhere. obviously what I said above could be totally off, and I have no way of knowing.
jvd said:I don't see where you get this from. in 2005-2006 90nm is what ati will be using just like every other gpu maker in the pc sector. They don't have acess to greater fabs because they out source production. They also don't have billions to make thier own fabs. Perhaps ms will pay for a very advance fab and we might see 65nm. But there is no reason to do that. They don't need that advance process . It seems to me Ms will be going with 90nm for initial launch and use 65nm and 45 nm for cost savings later on and sony will be pushing 65nm and will move to 45nm for cost savings. Its hard to say which one will work best. But its most likely that the gpu on the xbox will become cheaper quicker than the cell chip. If ms chases the tech .nonamer said:This pretty much confirms my suspicion that ATI is the low-cost, low-risk GPU provider for Microsoft. If they are wait till 2005 for the 90nm, then frankly I don't see them very competitive in terms of power (probably comparable, like the NGC is to the XB, but you can't simply say the GC is equal/better). As it stands now, even if Cell is a lousy design, PS3 still has probably won the powergame already simply due to brute force design and process investments. I saw MS as a potential challenge, but unless a major change in their PR then I can't see MS seriously challenging Sony anymore.
Panajev, I'll bet you a $1000 that the PS3 will be out on 65nm. 45nm is waaaayyyy out there, like 2007/2008. There are major lithography problems at 45nm and leakage on that process could make SOI scream "mommy!"
Vince, is Sony's SOI PD or FD?
Great fabs are available. Last I heard IBM is in the fab market, TSMC will probably have 65nm available by 2005 (more like 2006 for real volume). Don't know about UMC though. The only limiting factor is the designing of the GPU, not the process. With MS's huge resources, they could easily pay more for something better than what's ATI offering.
The second half of what you're saying is what I'm saying. ATI is cheap, low risk, not a lot of power relatively. Obviously going 90nm will be much cheaper than aiming for 65nm, but it's not as impressive. It's the Gamecube situation. It may be different from XB2, but it's not something you would talk about in a forum like this.
Panajev2001a
Veteran
Nonamer said:
Heh.
I understand that it might come out in 65 nm, but do not expect the chips will stay in 65 nm for very long as they might border on the hardly economically viable ( 280-290+ mm^2 ).
It won't be a problem manufacturing PlayStation 3 consoles with 65 nm chips, sure it will be very expensive, but if they launch Q4 2005 or Q1 2006 then they do not have to wait very long for the 45 nm die shrink: if you add to that the fact they might be using the 45 nm SOI manufacturing process with capacitor-less e-DRAM then you can see how much smaller a 45 nm SOI CELL chip could be ( also knowing that the e-DRAM portion is not an insignificant part of the cip's area ).
As coincidence would have it, there's a article on eet.com about next-gen processes: http://www.eet.com/semi/news/OEG20030929S0046 Very interesting as it pretty much wraps up some of the biggest problems going into 65nm and 45nm.
Panajev2001a
Veteran
nonamer said:
Thanks for the link, I do expect STI to have CELL chip die-shrinks by Q1-Q2 2007 which is about 3 years after they start mass production ( Half-to-late 2004 ).
I still suspect that at 65 nm the PlayStation 3's CELL chips will be quite huge.
Not a problem if you launch Q4 2005 or Q1 2006 ( Q1-Q2 2006 for the North American launch, maybe ).
Don't you agree ?
( I am glad that litography fans like Vince and you are here in the console forum
)Similar threads
