This is an article about the 65 nm process in the PS3:
http://pc.watch.impress.co.jp/docs/2004/1220/kaigai142.htm
http://pc.watch.impress.co.jp/docs/2004/1220/kaigai142.htm
nVidia building the PS3 GPU in its "entirety"
- Thread starter Joe DeFuria
- Start date
- Status
try to adjust the translated japanese version
*Hiroshige's Goto Weekly overseas news*
The PS3 media processor which is produced with 65nm process
--------------------------------------------------------------------------------
- The media processor of PS3 it produces with 65nm process
Hisashi 夛 it is good the wooden health person (SONY computer entertainment president and group CEO
First, it is necessary to presume whether production process of the PS3 media processor of NVIDIA how becomes.
A certain authorized personnel, calls the media processor of PS3 that it mass-produces with 65nm. Also the Cell processor which is the main CPU of PS3, at the time of PS3 appearance is presumed that it has become 65nm process. When we assume, that is, also 2 tips/chips mean to use leading process technology. When we assume that Microsoft Xbox2 (Xenon) uses 90nm process in the tip/chip, it means that process technology PS3 advances 1 generation.
This is the advantage of PS3 and, it is risk. Though, because new architecture suddenly trial manufacture -> those where it carries to mass production are the lith key with 65nm, Cell in the same way the media processor and first the possibility of starting trial manufacture with 90nm process are high.
According to release, production of the PS3 media processor, Fab2 and Toshiba and SONY of the SONY group being joint, has meant the tea which is managed s semi-conductor (OTSS) with to be done considerably. Fab2 90nm (2F) does mass production, 65nm (1F) has started. OTSS has prepared also 65nm with the 300mm wafer, at the time of real mass production, as for SCEI production the switch is possible with two Fab whose compatibility of process technology is high.
IBM, Embedded DRAM (eDRAM) it is included in the joint ownership process technology of Toshiba and SCEI. There is no sign to which at present, eDRAM has been recorded before the Cell processor. So, the media processor how being?
When of the character, memory zone most important GPU is thought, as for the media processor the possibility of placing eDRAM is high. In the first place, memory zone is guaranteed with eDRAM, the fact that performance is made is the proud pattern of SCEI. Also the patent application which is seen as the Cell processor of SCEI was visible, as designated eDRAM as prerequisite.
Though, there is also a hurdle in the adoption of eDRAM. As for that eDRAM and SOI (silicon-on-insulater) with coexistence of technology, as for these 2 technologies it is said that coexistence is difficult. 3 corporations have tackled also the engineering development which actualizes eDRAM on SOI, many dissertation has been issued, but there is an attitude which can really be mass-produced, you do not know whether or not is.
Because of that, to tell the truth as for the Cell processor and the PS3 media processor production process technology differs. According to information muscle, the Cell processor uses the SOI technology of 65nm, but as for the media processor you say that it is the bulk CMOS of 65nm. Although it is the expectation of the same Fab, that if the media processor is the bulk it presumes, that is, in order to use eDRAM which still is the hurdle in the coexistence with SOI coherence is agreeable.
- As for game machine the combination of eDRAM and high-speed external DRAM in main current?
If we assume that the PS3 media processor, uses eDRAM, video memory, is presumed that it becomes the combination of eDRAM and external DRAM. EDRAM can actualize very wide band, but capacity is limited. As for external DRAM as for zone it is limited than eDRAM vis-a-vis that but you maintain capacity largely. In case of 3D graphics, because necessary zone differs rather depending upon the data, it is presumed that segregation is possible.
For example, for GAMECUBE of ATI also system tip/chip "Flipper", mass memory of built-in (1T-SRAM of 3MB) with has combined external DRAM. As for 1T-SRAM of Flipper, as for memory cell itself eDRAM (0.18 μ m embedded DRAM process of the NEC electronics) is, but interface is high-speed memory of SRAM type. With Flipper, as a general rule the memory of this 3MB is allotted to the frame buffer and the Z buffer, and texture cash.
It is said that in funny thing, it becomes the memory constitution which and eDRAM and external DRAM combines also tip/chip for Xbox2 where ATI develops the core. If information and presumption of this are correct, for the game machine the memory of GPU means that eDRAM of built-in and combination of external DRAM become main current. As for that, for the game machine as for GPU, it means also the fact that it is restricted to the world where the Japanese semiconductor manufacturer whose eDRAM is proud directs.
Actually, when we assume that eDRAM of bulk is loaded, the foundry of Xbox2 (foundry), the Japanese semiconductor manufacturer which has high-level eDRAM technology becomes most major candidate. With Flipper, as for Nintendo Co. you used NEC as the foundry partner. Because as for this, the NEC electronics has the process of logic &eDRAM of high performance. When of such details are thought, the tip/chip of Xbox2 which the same ATI develops, the possibility the area of NEC which takes charge of the production of Flipper producing is high. By the way, it is presumed that from the next generation GPU of ATI "R520" 90nm process, architecture for Xbox2 of the same generation also the tip/chip is 90nm process. NEC has offered 90nm process "UX6" already.
- If 65nm process constitution of 24 pipes or more possibly
So, when we assume, that the PS3 media processor, is mass production with 65nm, it probably can insert how many functions? Graphics Synthesizer of PS2 (GS), the first announcement time they were 279 squares mm at 0.25 μ m, but it shrank to period relatively shortly. This time, when you think of that it is restricted to 65nm for a while, the possibility the die/di becoming small is somewhat higher than the first generation GS. Though, because wafer size from 200mm increases to 300mm, that much, is increase of the die/di withstands, (it meaning that cost per wafer goes down). So, when you think, perhaps with 65nm 200 square mm questions the proper hitting.
In case of NVIDIA, with the 130nm process of IBM, the die/di of approximately 300 squares mm 2 hundred million 2,200 ten thousand transistors are placed in NV40 (the semiconductor itself), the 32bit precision 16 pixel pipe is formed. When it becomes 65nm, 2 generations being the case that it is refined from 130nm, if, at the time of 65nm the die/di of 200 squares mm, in regard to calculation if 6 100010000 transistor weaknesses, 300 squares mm it means to be able to load above approximately 8 hundred million transistors as logic &SRAM (some the possibility of decreasing is higher than this in actuality). Assuming, if it cut the half of the die/di in eDRAM, in regard to calculation pipe constitution of 1.5 times or more (24 - 32 pipes) it means to be able to take. Actually as for the pipe of raster operation being so, it is overkill, but constitution of the Shader side which is operational unit is a possibility of 24 - 32 pipes.
You must add to calculation in other things, as for, for PC GPU is unit of the around interface which differs. It is seen that it is integrated in the PS3 media processor, interface of interface "Redwood" between the tip/chip and XDR DRAM because the circuit and the like of timing adjustment in order to absorb the skew is had, is takes the die/di. In addition, if we assume if, it integrates also the interface around the southern bridge tip/chip, (the possibility of becoming another tip/chip it is high), it is necessary also for that area to add. But, like PCI Express x16 if there is no large-scale serial interface which needs the enormous area, it does not influence probably will be that much largely.
- When it calculates backward from the development cycle of NVIDIA
Process technology, estimating in the adult closing, when we assume that pipe constitution of 24 or more takes, as for NVIDIA it probably will make what kind of GPU core.
In order to presume that, it is necessary for NVIDIA to be able to do how many development periods and to cut the resource in the PS3 media processor and to know high.
First, development period with the schedule which never is room is improbable. As for the development cycle of GPU of NVIDIA approximately 18 - 22 months. When it is typical pattern, starting from product specification decision, it becomes as follows.
After the specification decisive approximately 3 - 4 months of micro architecture
RTL (Register Transfer Level) after the completion approximately 6 - 8 months of formation
Tape out approximately 10 - 18 after the months
After the start approximately 18 - 22 months of mass production
Because as for those where it has the gap, NVIDIA in the past has reached the point where it starts to be somewhat longer than the cycle which has been explained, recently. For example, because NV40 has started development from 2002 first half, nearly approximately 2 years it means to be required. In other words, if new architecture GPU, the schedule which was seen long is more accurate.
When even then, you compare with CPU of cycle of 4 years or more, extremely it is speedy, but when you compare with the chip set which is designed in 12 months it is rather long. When this is fitted in PS3, how becoming? As for NVIDIA the cooperative relationship with SCEI said 2 years, but a certain authorized personnel, says that even at that time there was complication.
Actually, when we assumed that, modification of the specification of XDR DRAM which is loaded onto PS3, according to the imagination is influence of the design of the media processor of NVIDIA, those where architecture of the around memory of the media processor became firm mean between to 2004 July from 2003 July. Whether it has the interface of DRAM memory of external directly, or does not have, the high level design of the media processor receives considerable influence. So when you think, after specification of media processor becoming firm, being long, only 1 year half it means not to be passed. If we assume that is, it means the level whether or not finally tape out it does.
- The possibility of pouring the development resource into PS3
However, at the time of 2004 May E3, when the sample tip/chip of the media processor already has operated, it conveyed a certain information source. When we assume, if that much to be quick the tip/chip did, the PS3 media processor, unless there is diversion from considerable, the existing design, means difficult thing.
But, there is also an evidence which is contrary to this. First, as for NVIDIA you say that are SCEI and cooperative relationship, but never cooperating, development "it did," that the tip/chip past in shape you do not say. From this, as for the PS3 media processor of the version which at least is loaded onto PS3, the possibility of not having completed yet is high. Substituting with the architecture close tip/chip, perhaps with present condition, it has tested.
Looking at the road map of NVIDIA, the possibility the PS3 media processor, without being the remodelling edition of existing, GPU being new architecture is high. Because it is clear for the same company to have cut the big development resource in the PS3 media processor.
NVIDIA with typical pattern throws 1 time new architecture GPU to 1, after that half year, throws revised edition architecture. If you refer to previous life generation, GeForce FX 5800 (NV30) with new architecture, GeForce FX 5900 (NV35) it becomes revised edition architecture. Actually, with NV35, inserting the hand inside the pipeline rather, it made NVIDIA and changed.
However, as for 2004 new architecture GeForce 6800 (NV40) although it appeared, the revised edition (NV47) throwing, somehow is improbable. Furthermore, as for ATI Technologies to throwing R520 of the next generation to 2005 first half, as for the counter product of NVIDIA there is no same timing. In other words, the development cycle of NVIDIA is the case that at least half year or more it has been delayed.
This has shown the fact that it has cut the development resource whose NVIDIA is enormous in the PS3 media processor. When you see just this, the PS3 tip/chip as former new architecture GPU means that the possibility of becoming, big architecture reformation of the same level is high. If we assume that is, as for NVIDIA the resource which develops GPU of 1 generation the next being packed, there is also a possibility of aiming for radical architecture reform.
(2004 December 20th)
*Hiroshige's Goto Weekly overseas news*
The PS3 media processor which is produced with 65nm process
--------------------------------------------------------------------------------
- The media processor of PS3 it produces with 65nm process
Hisashi 夛 it is good the wooden health person (SONY computer entertainment president and group CEO
First, it is necessary to presume whether production process of the PS3 media processor of NVIDIA how becomes.
A certain authorized personnel, calls the media processor of PS3 that it mass-produces with 65nm. Also the Cell processor which is the main CPU of PS3, at the time of PS3 appearance is presumed that it has become 65nm process. When we assume, that is, also 2 tips/chips mean to use leading process technology. When we assume that Microsoft Xbox2 (Xenon) uses 90nm process in the tip/chip, it means that process technology PS3 advances 1 generation.
This is the advantage of PS3 and, it is risk. Though, because new architecture suddenly trial manufacture -> those where it carries to mass production are the lith key with 65nm, Cell in the same way the media processor and first the possibility of starting trial manufacture with 90nm process are high.
According to release, production of the PS3 media processor, Fab2 and Toshiba and SONY of the SONY group being joint, has meant the tea which is managed s semi-conductor (OTSS) with to be done considerably. Fab2 90nm (2F) does mass production, 65nm (1F) has started. OTSS has prepared also 65nm with the 300mm wafer, at the time of real mass production, as for SCEI production the switch is possible with two Fab whose compatibility of process technology is high.
IBM, Embedded DRAM (eDRAM) it is included in the joint ownership process technology of Toshiba and SCEI. There is no sign to which at present, eDRAM has been recorded before the Cell processor. So, the media processor how being?
When of the character, memory zone most important GPU is thought, as for the media processor the possibility of placing eDRAM is high. In the first place, memory zone is guaranteed with eDRAM, the fact that performance is made is the proud pattern of SCEI. Also the patent application which is seen as the Cell processor of SCEI was visible, as designated eDRAM as prerequisite.
Though, there is also a hurdle in the adoption of eDRAM. As for that eDRAM and SOI (silicon-on-insulater) with coexistence of technology, as for these 2 technologies it is said that coexistence is difficult. 3 corporations have tackled also the engineering development which actualizes eDRAM on SOI, many dissertation has been issued, but there is an attitude which can really be mass-produced, you do not know whether or not is.
Because of that, to tell the truth as for the Cell processor and the PS3 media processor production process technology differs. According to information muscle, the Cell processor uses the SOI technology of 65nm, but as for the media processor you say that it is the bulk CMOS of 65nm. Although it is the expectation of the same Fab, that if the media processor is the bulk it presumes, that is, in order to use eDRAM which still is the hurdle in the coexistence with SOI coherence is agreeable.
- As for game machine the combination of eDRAM and high-speed external DRAM in main current?
If we assume that the PS3 media processor, uses eDRAM, video memory, is presumed that it becomes the combination of eDRAM and external DRAM. EDRAM can actualize very wide band, but capacity is limited. As for external DRAM as for zone it is limited than eDRAM vis-a-vis that but you maintain capacity largely. In case of 3D graphics, because necessary zone differs rather depending upon the data, it is presumed that segregation is possible.
For example, for GAMECUBE of ATI also system tip/chip "Flipper", mass memory of built-in (1T-SRAM of 3MB) with has combined external DRAM. As for 1T-SRAM of Flipper, as for memory cell itself eDRAM (0.18 μ m embedded DRAM process of the NEC electronics) is, but interface is high-speed memory of SRAM type. With Flipper, as a general rule the memory of this 3MB is allotted to the frame buffer and the Z buffer, and texture cash.
It is said that in funny thing, it becomes the memory constitution which and eDRAM and external DRAM combines also tip/chip for Xbox2 where ATI develops the core. If information and presumption of this are correct, for the game machine the memory of GPU means that eDRAM of built-in and combination of external DRAM become main current. As for that, for the game machine as for GPU, it means also the fact that it is restricted to the world where the Japanese semiconductor manufacturer whose eDRAM is proud directs.
Actually, when we assume that eDRAM of bulk is loaded, the foundry of Xbox2 (foundry), the Japanese semiconductor manufacturer which has high-level eDRAM technology becomes most major candidate. With Flipper, as for Nintendo Co. you used NEC as the foundry partner. Because as for this, the NEC electronics has the process of logic &eDRAM of high performance. When of such details are thought, the tip/chip of Xbox2 which the same ATI develops, the possibility the area of NEC which takes charge of the production of Flipper producing is high. By the way, it is presumed that from the next generation GPU of ATI "R520" 90nm process, architecture for Xbox2 of the same generation also the tip/chip is 90nm process. NEC has offered 90nm process "UX6" already.
- If 65nm process constitution of 24 pipes or more possibly
So, when we assume, that the PS3 media processor, is mass production with 65nm, it probably can insert how many functions? Graphics Synthesizer of PS2 (GS), the first announcement time they were 279 squares mm at 0.25 μ m, but it shrank to period relatively shortly. This time, when you think of that it is restricted to 65nm for a while, the possibility the die/di becoming small is somewhat higher than the first generation GS. Though, because wafer size from 200mm increases to 300mm, that much, is increase of the die/di withstands, (it meaning that cost per wafer goes down). So, when you think, perhaps with 65nm 200 square mm questions the proper hitting.
In case of NVIDIA, with the 130nm process of IBM, the die/di of approximately 300 squares mm 2 hundred million 2,200 ten thousand transistors are placed in NV40 (the semiconductor itself), the 32bit precision 16 pixel pipe is formed. When it becomes 65nm, 2 generations being the case that it is refined from 130nm, if, at the time of 65nm the die/di of 200 squares mm, in regard to calculation if 6 100010000 transistor weaknesses, 300 squares mm it means to be able to load above approximately 8 hundred million transistors as logic &SRAM (some the possibility of decreasing is higher than this in actuality). Assuming, if it cut the half of the die/di in eDRAM, in regard to calculation pipe constitution of 1.5 times or more (24 - 32 pipes) it means to be able to take. Actually as for the pipe of raster operation being so, it is overkill, but constitution of the Shader side which is operational unit is a possibility of 24 - 32 pipes.
You must add to calculation in other things, as for, for PC GPU is unit of the around interface which differs. It is seen that it is integrated in the PS3 media processor, interface of interface "Redwood" between the tip/chip and XDR DRAM because the circuit and the like of timing adjustment in order to absorb the skew is had, is takes the die/di. In addition, if we assume if, it integrates also the interface around the southern bridge tip/chip, (the possibility of becoming another tip/chip it is high), it is necessary also for that area to add. But, like PCI Express x16 if there is no large-scale serial interface which needs the enormous area, it does not influence probably will be that much largely.
- When it calculates backward from the development cycle of NVIDIA
Process technology, estimating in the adult closing, when we assume that pipe constitution of 24 or more takes, as for NVIDIA it probably will make what kind of GPU core.
In order to presume that, it is necessary for NVIDIA to be able to do how many development periods and to cut the resource in the PS3 media processor and to know high.
First, development period with the schedule which never is room is improbable. As for the development cycle of GPU of NVIDIA approximately 18 - 22 months. When it is typical pattern, starting from product specification decision, it becomes as follows.
After the specification decisive approximately 3 - 4 months of micro architecture
RTL (Register Transfer Level) after the completion approximately 6 - 8 months of formation
Tape out approximately 10 - 18 after the months
After the start approximately 18 - 22 months of mass production
Because as for those where it has the gap, NVIDIA in the past has reached the point where it starts to be somewhat longer than the cycle which has been explained, recently. For example, because NV40 has started development from 2002 first half, nearly approximately 2 years it means to be required. In other words, if new architecture GPU, the schedule which was seen long is more accurate.
When even then, you compare with CPU of cycle of 4 years or more, extremely it is speedy, but when you compare with the chip set which is designed in 12 months it is rather long. When this is fitted in PS3, how becoming? As for NVIDIA the cooperative relationship with SCEI said 2 years, but a certain authorized personnel, says that even at that time there was complication.
Actually, when we assumed that, modification of the specification of XDR DRAM which is loaded onto PS3, according to the imagination is influence of the design of the media processor of NVIDIA, those where architecture of the around memory of the media processor became firm mean between to 2004 July from 2003 July. Whether it has the interface of DRAM memory of external directly, or does not have, the high level design of the media processor receives considerable influence. So when you think, after specification of media processor becoming firm, being long, only 1 year half it means not to be passed. If we assume that is, it means the level whether or not finally tape out it does.
- The possibility of pouring the development resource into PS3
However, at the time of 2004 May E3, when the sample tip/chip of the media processor already has operated, it conveyed a certain information source. When we assume, if that much to be quick the tip/chip did, the PS3 media processor, unless there is diversion from considerable, the existing design, means difficult thing.
But, there is also an evidence which is contrary to this. First, as for NVIDIA you say that are SCEI and cooperative relationship, but never cooperating, development "it did," that the tip/chip past in shape you do not say. From this, as for the PS3 media processor of the version which at least is loaded onto PS3, the possibility of not having completed yet is high. Substituting with the architecture close tip/chip, perhaps with present condition, it has tested.
Looking at the road map of NVIDIA, the possibility the PS3 media processor, without being the remodelling edition of existing, GPU being new architecture is high. Because it is clear for the same company to have cut the big development resource in the PS3 media processor.
NVIDIA with typical pattern throws 1 time new architecture GPU to 1, after that half year, throws revised edition architecture. If you refer to previous life generation, GeForce FX 5800 (NV30) with new architecture, GeForce FX 5900 (NV35) it becomes revised edition architecture. Actually, with NV35, inserting the hand inside the pipeline rather, it made NVIDIA and changed.
However, as for 2004 new architecture GeForce 6800 (NV40) although it appeared, the revised edition (NV47) throwing, somehow is improbable. Furthermore, as for ATI Technologies to throwing R520 of the next generation to 2005 first half, as for the counter product of NVIDIA there is no same timing. In other words, the development cycle of NVIDIA is the case that at least half year or more it has been delayed.
This has shown the fact that it has cut the development resource whose NVIDIA is enormous in the PS3 media processor. When you see just this, the PS3 tip/chip as former new architecture GPU means that the possibility of becoming, big architecture reformation of the same level is high. If we assume that is, as for NVIDIA the resource which develops GPU of 1 generation the next being packed, there is also a possibility of aiming for radical architecture reform.
(2004 December 20th)
Megadrive1988
Veteran
need more translation
The translation has a great start.
But nothing solid there. The article is speculation(ok, 'educated guesses with access to bits of industry news') not solid info, though we may weigh it according to the credentials of the writer. But I don't know who he is...
To summarise, he speculates that both CPU and GPU are fabbed using 65nm, and that the GPU will be quite different with Nvidia's desktop products.
aaronspink
Veteran
Vince said:
Go read up on arguements and debate and come back when you have an understanding.
Ahh, classic vince. Losing the arguement! Change the subject!
aaronspink
Veteran
passerby said:
No problem, I'll just start ignoring vince's posts.
Aaron spink
speaking for myself inc.
Martin Eddy
Regular
I just wish one of you would ignore the other one. We've had almost 18 pages of "Yes it is dumbass" , "No it's not moron"
It's wearing a bit thin.
Please somebody lock this thread! God save us all!
It's wearing a bit thin.
Please somebody lock this thread! God save us all!
aaronspink is Teh Wiener!!!
He was first to agree on ignoring!
aaronsping 1- Vince 0
He was first to agree on ignoring!
aaronsping 1- Vince 0
Sorry.
I want to add, before we start calling bitterness, jadedness or jealousy, aaron's industry experience is probably greater than all us armchair critics combine.
I find aaron is just telling everyone to wait before jumping the gun, something which i observe, the rest of those with industry dealings, similarly putting across.
Not that they wil be more right to predict the future, but years of watching much hypes and promises go past them, their extra caution is probably justified and hard earned.
It may also seem some bickering is forming, but i think aaron just peeved with some members' confrontational as-a-matter-of-fact-rightly-so-my-attitude style replies lately.
Aaron probably posted more in the last few days than since joining B3D last year!
I want to add, before we start calling bitterness, jadedness or jealousy, aaron's industry experience is probably greater than all us armchair critics combine.
I find aaron is just telling everyone to wait before jumping the gun, something which i observe, the rest of those with industry dealings, similarly putting across.
Not that they wil be more right to predict the future, but years of watching much hypes and promises go past them, their extra caution is probably justified and hard earned.
It may also seem some bickering is forming, but i think aaron just peeved with some members' confrontational as-a-matter-of-fact-rightly-so-my-attitude style replies lately.
Aaron probably posted more in the last few days than since joining B3D last year!
Bingo (on both points).
Its nothing new.
- Status
Similar threads
