Next gen consoles: hints?

[Panajev2001a]

This, running at 3-4 GHz ( possible IMHO... the only parts that would run that high would be the APUs, the busses would only run at 1 GHz [they are 1,024 bits busses after-all, this would already mean 128 GB/s at 1 GHz] and the e-DRAM could run at 1 GHz SDR or 500 MHz but use DDR signalling for data transfers ) would provide a nice amount of power.

Features:

24 APUs in total

2x Pixel Engine + Image Cache + CRTC controller ( each Pixel Engine fills an independent triangle as the pixel pipelines are not tied ).

Shared e-DRAM block.

4 PEs: 2 of them are of the "Visualizer" kind.

APUs' clock: 3-4 GHz

PUs' clock: 1.5-2 GHz ( preferrably it stays 1/2 the APU speed ) or less ( APUs can be pretty independent )

FP/FX Performance: 24 APUs * 8 ops/clock [FP or FX] * 3-4 GHz = 576-768 GFLOPS/GOPS

Bus Bandwidth local: ~128 GB/s ( 128 bytes per cycle or 1,024 bits per cycle )

Local Storage to Register File Bandiwdth ( in each APU ): 256 bits per cycle ( two 128 bits vectors worth of data ).

A Redwood based bus could connect all 4 PEs together: in the diagrams it specifies it to be naturallya 1,024 bits bus, but Redwood's high Data Signalling Rate could allow for a smaller bus or maybe Redwood would be used to connect this big MPU and the I/O ASIC which would contain the Memory Controller for the External Memory.



With this configuration all the concerns of Mr. Dave Baumann should disappear

 

[marconelly!]

ATi is centuries a head of sony in terms of Video Processing Technology. even MS knows this and is also using ATi. PS2 is weak compared to GC and X-box graphics..........

Well, no wonder it's weaker when it's more than 1.5 years older. At the time PS2 appeared, there was little on the PC equipped with ATI cards that compared with the games we're seeing on the PS2.

AS for GC not playing DVDs yeah thats ok as I would much rather put the disk in my sony player and let it play automatically without having to use a game controller........

PS2 and Xbox (when you have the add-on) play DVDs exactly the same way. You pop them in, they start playing.

 

[Fox5]

ATi is centuries a head of sony in terms of Video Processing Technology. even MS knows this and is also using ATi. PS2 is weak compared to GC and X-box graphics..........

Well, no wonder it's weaker when it's more than 1.5 years older. At the time PS2 appeared, there was little on the PC equipped with ATI cards that compared with the games we're seeing on the PS2.

When ps2 was out, I think the geforce 2 ultra was out. Outside of the lack of programmable pixel shaders, I believe that has everything the geforce 3 has, and early ps2 games were really crappy looking and ports of PC games sucked. Now while ps2's graphics have improved, so have the graphics you could see on a geforce 2 ultra, and eventually it will be running doom 3, though how well I can't say.

 

[Dave Baumann]

I’d still like to hear more about the Visualisers sampling and fragment processing abilities though, I’d like to know how it operates with texture data as an input and how efficient that will be.

I've already told you, it doesn't exist. Many of the things you want to know the patent hasn't went over in terms of the VS.

Paul, I’m aware of this, I’m saying that I’d like to know more because to a certain extent everyone appears to be taking a leap of faith as to its abilities and intended operation – there are certain part of that pipeline that appear kinda backwards for fragment operation.

Toshiba and Rambus demonstrated the worlds fastest parallel interface in October running at 6.4Ghz, this is Redwood which has been liscensed to be used in the Broadband Engine.

I see the specification is from 400MHz to 6.4GHz – has there been any indication on where in that range the PS3 specification settles for the external interconnect?

Fast as it is I agree that this won't top PS3 e-DRAM at hundreds of GB's per second.

You could be right on this, and any interconnect that is lower bandwidth than the internal bandwidth, even slightly, is probably going to be a bottleneck if both are used as a single “pool†of resources which is why I would still guess that each chip would have a fairly distinct usage in many cases.

The other more fundamental question is do you want a your ?GPU? to act functionally equivalent to the ?CPU?, or spend its transistor budget being focused on tasks that are commonly required for a ?GPU??

That depends on what kind of tasks we're talking about. There is certain parts of fixed functionality I would like to see them include, but beyond that I like my freedom.

Question: Where in the DX approach feel that it hinders your freedom unreasonably (for you)?

 

[Dave Baumann]

With this configuration all the concerns of Mr. Dave Baumann should disappear

Well you've just cut your processing power in half on a clock for clock basis and we still don't know how the fragment processing is supposed to operate.

 

[cthellis42]

When ps2 was out, I think the geforce 2 ultra was out.

Close, but the GF2 Ultra still had a few quarters extra. The GTS was out shortly after PS2's launch, though. (Not that there was much difference between them excepting clock rate.)

 

[Paul]

I see the specification is from 400MHz to 6.4GHz – has there been any indication on where in that range the PS3 specification settles for the external interconnect?

Well this was probably a test for the Broadband Engine as Toshiba and Sony did liscense it for use in BE.

This new evaluation chip is implemented on Toshiba's 90-nanometer ASIC process and is capable of running at speeds up to 6.4GHz, which is six times faster than processor busses available today. The chip is being used as a test vehicle for future customer platforms. The Redwood interface has been designed for high volume, cost-sensitive applications.

Toshiba will apply the result of this joint effort, based on the most advanced process technologies including 65-nanometer, to leading system LSIs."


I really have no doubt that if PS3 needs the fast connection between BE and it's GPU that Toshiba will input a 6.4Ghz bus, if that type of speed isn't needed something like 3.2 could be used.

 

[Dave Baumann]

We seem to be saying it was licensed for use "in" the BE - does this 6.4GHz figure relate to internal speeds? What are the external bus rates?

 

[Paul]

Well the 6.4Ghz figure is for the connect between the BE and the GPU.

WHEREAS, the parties desire to cooperate with each other to enable SCE
and Toshiba (and [*] as SCE's sublicensee) to implement Rambus' Redwood Rambus
Interface Technology and Yellowstone Rambus Interface Technology as the bus
interfaces in and with the Broadband Engine

You can see one of Goto's estimation pictures if you want.

http://pc.watch.impress.co.jp/docs/2003/0529/kaigai05l.gif

 

[Teasy]

PS2 and Xbox (when you have the add-on) play DVDs exactly the same way. You pop them in, they start playing.

The PS2 doesn't even play SVCD's though, its a pretty aweful DVD player compared to even the cheapest (£30) DVD players out these days.

 

[Dio]

With this configuration all the concerns of Mr. Dave Baumann should disappear

Where are miss latencies compensated?

 

[Dave Baumann]

Well the 6.4Ghz figure is for the connect between the BE and the GPU.

WHEREAS, the parties desire to cooperate with each other to enable SCE
and Toshiba (and [*] as SCE's sublicensee) to implement Rambus' Redwood Rambus
Interface Technology and Yellowstone Rambus Interface Technology as the bus
interfaces in and with the Broadband Engine

You can see one of Goto's estimation pictures if you want.

http://pc.watch.impress.co.jp/docs/2003/0529/kaigai05l.gif

That doesn't show what the external speeds are though.

So far we appear to have tets of an interface running at 6.4GHz, but I'm not clear on anything actually pinpointing the external bud speeds running at this rate for PS3 - do you have anything that actually states that?

IIRC external busses at these rates are problematic due to the electirc interferance they can end up causing.

 

[Paul]

Having bought the Rambus, Toshiba, Sony contract I can say that it mentions the test chip that Toshiba has just made.

This ASIC according to the contract was to be made to test the Redwood configuration the Broadband Engine is to have.

A Yellowstone(now XDR) test chip should arrive eventually too.

But I cannot post anything from the contract as that would be illegal, you can view a sample of it in my sig though.

 

[Dave Baumann]

I still don't appear to have an answer to the question I asked.

I see they have a Rewood text "chip" - has there been a test chip that has passed data to another chip and if so do we know the external transfer rates?

 

[Paul]

Oh it's the numbers you want in GB/s. I do not know, the Rambus website only says a bus speed of 6.4Ghz.

And the Toshiba test chip news only says 6.4Ghz too, no numbers in GB's per second.

No clue on the bandwidth numbers as nothing says what they are.

 

[Grall]

Dave, your question is just way too early to answer. Even speculating (a la Panajev ) is a dubious practice IMO at this stage because nobody can REALLY tell what clock rate buses, memories, APUs etc will run at. I'm not sure Sony itself knows yet what speed is realistically attainable. We saw M$ degrade GPU speed twice after finding yields were fairly bad initially, I guess it will be quite a while into next year at least until the final decision is made.


*G*

 

[Qroach]

Actually Grall, MS downgraded the GPU speed once from the time they had actual silicon. From 250-233. That 300mhz promise came from gigapixel.

Nintendo also downgraded the speed of flipper before it was released. As a few gamecube developers originally recieved faster version of dev hardware. In both cases it was done to improve yields.

 

[Panajev2001a]

With this configuration all the concerns of Mr. Dave Baumann should disappear

Well you've just cut your processing power in half on a clock for clock basis and we still don't know how the fragment processing is supposed to operate.

You are an expensive guy to please, aren't you ?

At 3-4 GHz for the APUs we are talking about 512-768 GFLOPS, would you think that is not very useful for Vertex and Pixel Shading ?

 

[Panajev2001a]

Well the 6.4Ghz figure is for the connect between the BE and the GPU.

WHEREAS, the parties desire to cooperate with each other to enable SCE
and Toshiba (and [*] as SCE's sublicensee) to implement Rambus' Redwood Rambus
Interface Technology and Yellowstone Rambus Interface Technology as the bus
interfaces in and with the Broadband Engine

You can see one of Goto's estimation pictures if you want.

http://pc.watch.impress.co.jp/docs/2003/0529/kaigai05l.gif

That doesn't show what the external speeds are though.

So far we appear to have tets of an interface running at 6.4GHz, but I'm not clear on anything actually pinpointing the external bud speeds running at this rate for PS3 - do you have anything that actually states that?

IIRC external busses at these rates are problematic due to the electirc interferance they can end up causing.

Dave,

You know how XDR works like, right ?

3.2 GHz is only signalling rate ( it also uses differential signalling ), but the frequency running on the external bus is only 400 MHz and 800 MHz for 6.4 GHz signalling rate: I expect ODR ( the technology used in XDR Memory Technology ) to be used in Redwood.

 

[V3]

I still don't appear to have an answer to the question I asked.

I see they have a Rewood text "chip" - has there been a test chip that has passed data to another chip and if so do we know the external transfer rates?

Redwood for PC as can be seen here

http://www.rambus.co.jp/events/Tech5-3_Rambus_Jean-Marc.pdf

gives 64bit @ 400 MHz for 3.2 GB/s

And Redwood meant to scale from 400 MHz to 6.4 GHz as they claim here

http://www.rambus.com/products/redwood/redwood.pdf

I assume the upper end 6.4 GHz Redwood, would gives 64bit @ 6.4 Ghz for 51.2 GB/s

This is for PC obviously.