nVidia shader patent (REYES/Raytracing/GI) destined for PS3?

[one]

not to mention the external memory requirements for essentially a stream processor, ala the PS3 & its subpar random memory access for this type of full application.)

IIRC XDR-DRAM is not bad in random access

 

[version]

not to mention the external memory requirements for essentially a stream processor, ala the PS3 & its subpar random memory access for this type of full application.)

IIRC XDR-DRAM is not bad in random access

yeah, but xdr access 32 bit(32wire) , gddr3 have 256wire =256 bit about same time

gddr3 faster 8 times than xdr

 

[one]

yeah, but xdr access 32 bit(32wire) , gddr3 have 256wire =256 bit about same time

gddr3 faster 8 times than xdr

No way. XDR supports up to 128bit interface width (16 * 8-channel). Also, XDR is 2.5x faster per-pin than GDDR3 in 2005.

http://www.rambus.com/news/newsroom/pressrelease.cfm?id=110

XDR memory's novel system topology allows point-to-point differential data interconnects to scale to multi-gigahertz speeds, while the bussed address and command signals allow a scalable range of memory system capacity supporting from one to 36 DRAM devices. XDR offers a roadmap to 6.4GHz and can scale to interface widths of up to 128-bits, enabling memory system bandwidths up to 100GB/s, 16 times more than today?s 6.4GB/s memory systems. XDR DRAMs will be available in multiple speed bins, device densities, and device widths. With densities ranging from 256Mb to 8Gb, and device widths ranging from x1 to x32, XDR DRAM satisfies the needs of both high-bandwidth and high-capacity systems.

Samsung, Elpida, and Toshiba are all DRAM licensees of Rambus's latest memory interface technology.

 

[version]

yeah, but xdr access 32 bit(32wire) , gddr3 have 256wire =256 bit about same time

gddr3 faster 8 times than xdr

No way. XDR supports up to 128bit interface width (16 * 8-channel). Also, XDR is 2.5x faster per-pin than GDDR3 in 2005.

http://www.rambus.com/news/newsroom/pressrelease.cfm?id=110

XDR memory's novel system topology allows point-to-point differential data interconnects to scale to multi-gigahertz speeds, while the bussed address and command signals allow a scalable range of memory system capacity supporting from one to 36 DRAM devices. XDR offers a roadmap to 6.4GHz and can scale to interface widths of up to 128-bits, enabling memory system bandwidths up to 100GB/s, 16 times more than today?s 6.4GB/s memory systems. XDR DRAMs will be available in multiple speed bins, device densities, and device widths. With densities ranging from 256Mb to 8Gb, and device widths ranging from x1 to x32, XDR DRAM satisfies the needs of both high-bandwidth and high-capacity systems.

Samsung, Elpida, and Toshiba are all DRAM licensees of Rambus's latest memory interface technology.

yes, xdr faster in burst mode, but with random access slow
ps3 have 32 or 64 bit(wire) for xdr

 

[one]

yes, xdr faster in burst mode, but with random access slow
ps3 have 32 or 64 bit(wire) for xdr

No, the expected bandwidth for the PS3 memory configuration is

256MB XDR DRAM (3.2Gtps) (256Mbit * 8 @ 128bit) = 51.2GB/s

or

128MB @ 64bit for Cell + 128MB @ 64bit for GPU = 25.6GB/s + 25.6GB/s

while Xenon is

512MB GDDR3 (1.6Gtps) (512Mbit * 8 @ 256bit) = 51.2GB/s

or

256MB GDDR3 (256Mbit * 8 @ 256bit) = 25.6GB/s

Anyway, read the picture above "Optimized XDR core timings yield higher effective bandwidth under random workloads"

 

[Phil]

Little work? so which is it that you want to argue for? on one side you argue they are providing tools and software and on the other you argue little work/effort. The two really don't go together. providing tools and software is a lot fo continuous work, once you can't describe as little "work/effort". Nvidia always designs a chip that will last for years to come.

Little work compared to their on-going cyclic business in their core-market. Little work in a sense that they'll be living off revenues of a system that's is bound to sell 100+ million? For at least 6 years after completion and not having to do a thing, while they can happily concentrate on their core-market again? Little work in that they won't even be responsible for the chips fabbing but just will be designing the thing and then forgetting about it.

As for "Nvidia always designs a chip that will last for years to come" - I'm still interested to see actual numbers of a chip that sold in same league the PlayStation brand does and is excected to continue in its third iteration. If there isn't - it just doesn't cut it. In fact, if their core-business was that lucrative, they wouldn't be doing a licencing deal with Sony, obviously. Having said that, Nvidia's actions speak far more than your arguments do...

 

[version]

yes, xdr faster in burst mode, but with random access slow
ps3 have 32 or 64 bit(wire) for xdr

No, the expected bandwidth for the PS3 memory configuration is

256MB XDR DRAM (3.2Gtps) (256Mbit * 8 @ 128bit) = 51.2GB/s

or

128MB @ 64bit for Cell + 128MB @ 64bit for GPU = 25.6GB/s + 25.6GB/s

while Xenon is

512MB GDDR3 (1.6Gtps) (512Mbit * 8 @ 256bit) = 51.2GB/s

or

256MB GDDR3 (256Mbit * 8 @ 256bit) = 25.6GB/s

Anyway, read the picture above "Optimized XDR core timings yield higher effective bandwidth under random workloads"

PSP has 32MB ram+8MBflash, PS3 128MB?? HUMOUR!

1. 256MB with 25GB/s
2. 512MB with 50GB/s
3. 1024MB with 100GB/s

 

[one]

PSP has 32MB ram+8MBflash, PS3 128MB?? HUMOUR!

PSP = $185 including LCD
PS3 = $? including Cell, media processor, Blu-ray drive, whatsoever

1. 256MB with 25GB/s
2. 512MB with 50GB/s
3. 1024MB with 100GB/s

What are those numbers? If you assume 512MB XDR DRAM, the total bandwidth are 100GB/s.

 

[version]

PSP has 32MB ram+8MBflash, PS3 128MB?? HUMOUR!

PSP = $185 including LCD
PS3 = $? including Cell, media processor, Blu-ray drive, whatsoever

1. 256MB with 25GB/s
2. 512MB with 50GB/s
3. 1024MB with 100GB/s

What are those numbers? If you assume 512MB XDR DRAM, the total bandwidth are 100GB/s.

CELL dont write too much to memory in game only read that , gpu too

25 GB/s / 60 frame = 416MB/frame , this is enough for 256 MB ram

 

[Brimstone]

Arriving at final conclusions about the X-Box 2 memory architechture and it's performance might be a little pre-mature at this point. I for one won't be surprised to see something special because of ATI and their history of spearheading the technology assualt on memory issues. The xddr rumor involving VIA and ATI in paticular comes to mind. It will be intresting to see if something sophisticated ends up in the next X-Box.

 

[London Geezer]

1. 256MB with 25GB/s
2. 512MB with 50GB/s
3. 1024MB with 100GB/s

Why do you keep scaling memory size proportionally with Bandwidth?

 

[nAo]

Why do you keep scaling memory size proportionally with Bandwidth?

bandwith x memory module.
I believe One is right ...Version is off mark by a factor 2

 

[version]

1. 256MB with 25GB/s
2. 512MB with 50GB/s
3. 1024MB with 100GB/s

Why do you keep scaling memory size proportionally with Bandwidth?

cpu+gpu do read about full memory in every frame at once(geometry,texture)
if memory faster this is wastage

 

[version]

ps2 32MB - 3.2GB/s
xbox 64MB - 6.4GB/s
128- 12
256-24 .......


this is consol's balance of LAW

 

[Fafalada]

Why do some people have a problem that XeCPU is a clean design, designed and built from scrarch for this one job.

Because it's so much easier to delude oneself that it's a 3.5Ghz Power5 with 3 cores and thus supercomputer equivalent?

 

[version]

Why do some people have a problem that XeCPU is a clean design, designed and built from scrarch for this one job.

Because it's so much easier to delude oneself that it's a 3.5Ghz Power5 with 3 cores and thus supercomputer equivalent?

xbox equ with PC
xenon with MAC

 

[ERP]

Why do some people have a problem that XeCPU is a clean design, designed and built from scrarch for this one job.

Because it's so much easier to delude oneself that it's a 3.5Ghz Power5 with 3 cores and thus supercomputer equivalent?

That and MS hasn't done enough declaring it to be a huge leap in computing architecture.

People believe what they want to believe even when there are real facts to dispute their beliefs, so in this case where there is no solid info it's hardly surprising.

 

[Fox5]

Why do some people have a problem that XeCPU is a clean design, designed and built from scrarch for this one job.

Because it's so much easier to delude oneself that it's a 3.5Ghz Power5 with 3 cores and thus supercomputer equivalent?

That and MS hasn't done enough declaring it to be a huge leap in computing architecture.

People believe what they want to believe even when there are real facts to dispute their beliefs, so in this case where there is no solid info it's hardly surprising.

What would you consider the gecko in gamecube?

 

[nAo]

That and MS hasn't done enough declaring it to be a huge leap in computing architecture.

MS should declare specs and features not how 'a huge leap is their CPU'.
Is it public how many moneys MS invested in Xenon CPU development?
Tell me how much it costs..and I tell you how fast it is..

ciao,
Marco

 

[Xenus]

Don't forget how long its been under development it is another gauge on just how custom it is.