nvidia GPU at 1 Ghz in ps3?

[jvd]

Just because intel and tmsc are going to offer it tells us nothing about the yields of that process .

Intel can use it for thier p4 ee line and ship 5k processers and consider it shipping , Tmsc can do it too .

Its my understanding that the ps3 will launch using 90nm in early 06 and move to 65nm by that winter or 2007

 

[3dcgi]

Notice the part about low power.

True, but where are you going with that?

Just trying to give more information. The article mentions three versions. General purpose, optimized for low power, and optimized for high performance. The low power version will be ready first.

 

[one]

Goto is just talking about general WGF 2.0 GPU in a long term, not specifically about ATI or NVIDIA. His impression is that some vendors are making their GPU more general purpose by hardware unified shaders, so CPU and GPU are getting closer. Hardware unified shaders can be grouped as a separate unit in a typical implementation. The general rule is, while it's difficult to raise the clock speed of complicated fixed-function units, it's easy to raise that of simple general-purpose units. But a general-purpose unit is still inefficient compared with a fixed-function unit, so it's a trade-off. Anyway, like in a PSP System-on-Chip, it's possible different parts have different clock speeds in a future GPU.

In the near future, GPU vendors may start to clock mid-range GPU products higher instead of adding more shader units. He just mentions the Cell GPU project in the past tense and the subjunctive mood (means the actual PS3 doesn't have it), just to add to the general trend.

Tessellator and Geometry Shader were to be included in WGF2.0, but Tessellator was removed. Goto speculates it's dropped since it's still argued whether it should be implemented with programmable shaders or with a fixed function unit, while it's assumed in WGF 2.0 Geometry Shader is implemented with shader units in a unified shader array.

 

[blakjedi]

Are either ATI or Nvidia's console parts expected to be WGF 2.0 compliant? I didnt think so.

The quote on TSMC having 65nm parts already out by December is intriguing.... makes me think x360

 

[j^aws]

To all the semiconductor insiders, what's the latest on 65nm timeline for STI?

 

[Dave Baumann]

Hardware unified shaders can be grouped as a separate unit in a typical implementation. The general rule is, while it's difficult to raise the clock speed of complicated fixed-function units, it's easy to raise that of simple general-purpose units.

As 3dcgi pointed out, this isn't really any different from what we have at the moment - the "shader cores" within each pipeline are hardly burdened with "fixed function units" as they are already quite distinct. Its been pointed out recently that the fixed function ROP's on current chips are already not running at the core clock rate of the rest of the chip, but are in fact in step with the memory timings (so, for instance, the shader cores on NV40 are running at 400MHz whilst the ROP's are running at 550MHz).

In the near future, GPU vendors may start to clock mid-range GPU products higher instead of adding more shader units. He just mentions the Cell GPU project in the past tense and the subjunctive mood (means the actual PS3 doesn't have it), just to add to the general trend.

This sounds like what ATI were looking towards with Fast14.

Are either ATI or Nvidia's console parts expected to be WGF 2.0 compliant? I didnt think so.

Considering the timescales for WGF2.0 now (Longhorn is now end of 2006 and will initially ship with only WGF1.0 - DX9) probably not. ATI's Xenon part is sounding closer to this unified approach - it won't have full WGF2.0 capabilities, but it is expected to be the platform ATI develop their eventual WFG2.0 architecture from.

The quote on TSMC having 65nm parts already out by December is intriguing.... makes me think x360

IMO, highly unlikely. 90nm was "available" at TSMC for over a year now and we are still waiting for 90nm graphics (R520, probably Xenon graphics) - this lag is fiarly usual for a customer process beng available and for complex ASIC's to appear.

 

[Brimstone]

A little off topic, but does anyone know if Microsoft is going to be fabbing their CPU's at TSMC, or are they going to be purchasing them from IBM? I ask because I recently read that TSMC is going to be aiming for 65nm production by year's end as well, and wondering how and when that might start having a positive spillover effect for Microsoft.

I very much doubt TSMC is going to fab the CPU. It' has been hinted at that the Power cores utilize SOI and TSMC already announced that don't have a single customer despite offering a SOI process.

SAN JOSE, Calif. — Lack of demand for silicon-on-insulator (SOI) wafers has prompted Taiwan Semiconductor Manufacturing Co. Ltd. (TSMC) to push out the technology until 2007, the company said on Tuesday (April 26).
Shang-yi Chiang, senior vice president of R&D at TSMC, said that the company does not have a single customer for SOI wafers despite offering the technology in various forms over the years.

"We are looking for a customer," Chiang said during a keynote address at the company's Technology Symposium here. "We don't have a customer."

Due to lackluster — if not poor — demand in the marketplace, TSMC will not offer SOI until the 2007 time frame, he said.

This is not to say SOI is dead in the water; on the contrary, AMD, IBM and others have adopted the technology. Intel Corp., however, has dismissed SOI technology, it was noted.

http://eetimes.com/news/semi/showArticle.jhtml?articleID=161600650



Team Xbox reported this which sites the Chief Operating Officer from Soitech clearly pointing out the CPU's from both Sony and Microsoft will have SOI.

According to Pascal Mauberger, Chief Operating Officer of Soitec, IBM’s silicon-on-insulator (SOI) semiconductor technology will be used in both the Microsoft Xbox successor and Sony’s PlayStation 3.

"SOI is in Xbox through IBM. It is in the Sony PlayStation. And we will soon have three 300-mm wafer fabs producing SOI," Mauberger told Silicon Strategies. "IBM East Fishkill is running 90-nm SOI. Sony's Nagasaki plant is due online by the end of 2004," he added.

Soitec is a manufacturer and provider of SOI wafers. The silicon-on-insulator is an advanced semiconductor manufacturing technology that produces higher performing, lower power processors than traditional silicon techniques.


This SOI technique speeds the flow of electrons through transistors to increase performance and provides an insulating layer in the silicon that isolates transistors, decreasing this way the power consumption. The technology allows manufacturers to fabricate higher clocked processors while lowering the power demands of the components.

A real example of SOI is IBM’s Power Architecture microprocessors, found in Apple’s PowerMac G5 computers and in Xbox 2 alpha SDKs. While processors from Intel or AMD consume as much as 100W (you might have seen that big heatsink inside your PC), IBM's current PowerPC 970FX chip consumes about 50 watts in its 2.5GHz version. The previous PowerPC 970 processors, manufactured on a 130-nm process, required 66 watts at a lower clock speed.

They Are Everywhere…

There is one thing that Sony, Nintendo and now Microsoft have in common: IBM. The Big Blue, Sony, and Toshiba partnered in 2001 to develop the "Cell" technology that will power the PlayStation 3 among other consumer electronics. IBM also provides the Gecko processor used in the Nintendo GameCube and industry insiders have revealed that “Revolution†will also use an IBM processor.

Who is the last one to join the party? As TeamXbox revealed in a world exclusive, IBM is the technology partner Microsoft chose to build a state-of-the-art processor for the Xbox successor that is supposed to be a multi-core CPU, capable of processing six threads simultaneously. This processor might also be the first PowerPC built on a 65-nm process, as we informed in this previous story.

Since IBM is the pioneer in silicon-on-insulator technology, it is a no-brainer that all these processors will be built using SOI as well as other IBM’s technologies.

Stay tuned, we’ll have more on Xbox 2 soon.

http://news.teamxbox.com/xbox/6201/SiliconOnInsulator-to-Power-Xbox-2-Processor


It looks to me downright impossible for TSMC to be manufacturing the XB-360 CPU's. This begs the question, what is TSMC manufacturing at the .90 nm node for MS?

 

[AzBat]

The GPU and possibly north bridge from SiS. I had always expected IBM to manufacture the main CPU. Doesn't make sense for TSMC to manufacture it.

Tommy McClain

 

[Carl B]

@Brimstone: Thanks a lot for those links - that was exactly what I wanted to know.

@AzBat: Well, I always thought IBM would be fabbing them, but I have no doubt that that is going to cost Microsoft more per chip than if they were having it done at TSMC and I just wanted some sort of confirmation. And plus does that mean also that the chip itself is being bought as a finished product from IBM, rather than seperate IP ownership/fabbing agreements like with the chipset and VPU? That is to say, does Microsoft own/have a license for the IP to the tri-core? I was just curious to get a mental image in my mind of what the possible cost breakdown per component will be for the 360.

 

[Farid]

The quote on TSMC having 65nm parts already out by December is intriguing.... makes me think x360

TSMC consider a process "available", when they're able to do, what they call, some Risk Productions.

MS, also, would need to start production of a few millions Xbox360 by, at least, early Q4 2005.
Thoses, TSMC and MS, timelines and quantities don't seem to match.

 

[one]

Though there's been no PR about MS so far, I think Chartered can manufacture Xbox 360 CPU.

 

[blakjedi]

This processor might also be the first PowerPC built on a 65-nm process, as we informed in this previous story.

Stay tuned, we’ll have more on Xbox 2 soon.

*shrug*

 

[karlotta]

This processor might also be the first PowerPC built on a 65-nm process, as we informed in this previous story.

Stay tuned, we’ll have more on Xbox 2 soon.

*shrug*

I think the R500 is a TSMC part, not the "IBM PC" part. and the .65 wil be out maybe in a year....

 

[Brimstone]

This article sheds some light on whats going on between IBM and Chartered right now.

AMD is going to start production at their new FAB 36 which will utilize .65 nm in the first quater of 2006. Chartered's .90 nm is going to be used as a safty net incase problems arise.

Is Chartered's involvement is a positive for customers like Nintendo and Microsoft? "That's a good assumption," said IBM's Lang. "If you look at what's happening with the common platform and our technologies, and the feedback from major customers, it's all positive, because look at what we're doing: We're bringing IBM technology in a collaboration mode to the marketplace with a huge capex sitting behind us now, because of our partners.


"In addition to that, we bring in our applications, we bring in our Watson research people, we bring in our ETS [Engineering and Technology Services] folks and their design and consultative capabilities," Lang added. "We're constructing an ecosystem that nobody else can match. I've gotten it back from customers about what a terrific value proposition this is. And one of the key aspects is that the manufacturing capability inside this ecosystem is fantastic."

...

Jelinek hinted that there is a "big guy" in the graphics processor space that has been working with Chartered and IBM that has not yet been announced as a customer.

http://www.semireporter.com/public/9187.cfm


I'm confused by the XB-360 "leaked" specs. How they will hit 3.0 Ghz + speeds with a tri-core cpu on a .90 nm process. IBM is supposed to have . 65 nm ready for productio this year at Fishkill and I'm more inclined to believe thats the process that will be used for the MS CPU.

 

[Carl B]

I'm confused by the XB-360 "leaked" specs. How they will hit 3.0 Ghz + speeds with a tri-core cpu on a .90 nm process. IBM is supposed to have . 65 nm ready for productio this year at Fishkill and I'm more inclined to believe thats the process that will be used for the MS CPU.

The Fishkill line is being fleshed out in part with heavy investment from Sony though, and that investment included a slice of the production capacity there, so I wouldn't be surprised if initial 65nm production is in other areas, including Cell.

As for 3 GHz at 90nm, that's not too hard to believe if the cores used in the tri-core are indeed similar to the PPE in Cell, which we already know can hit 5.2 GHz. In that context, 3 GHz might even be relatively 'cool' running.

 

[Urian]

I am going to say a stupid thing, I am sure of it.

But why not a dual core with two different cores?

The first core prerenders the scene, without shaders effects, only geometry+texturing+alpha blending, nothing more and puts the image in the eDRAM divided in a number of blocks.

The second core reads the blocks (one block per ALU or one block for a determined number of ALU) and applies all the effects.

I Believe if they divide the image in pixels the results could be very good.

But probably I am wrong.

 

[trinibwoy]

I'm no expert either but I would think that the texturing and blending stages may be dependent on shader output.