ATI & XBox

[Megadrive1988]

my probably incomplete flawed explaination is:

basicly, embedded memory is RAM memory put inside a processor, usually a graphics processor, but could also be a CPU processor, for the purpose of speeding up transmition times between memory and processor. if a processor has to go out across a bus to external memory, there is more delay for data to be processed. the bandwidth of embedded memory is typically much higher than that of external memory.

PlayStation2's graphics chip, GS and Gamecube's graphics processor, Flipper, both have embedded memory. 4 MB on GS, 3.12 MB on Flipper.

as mentioned just before my post here, PS2's 4MB embedded memory has an incredible 48 GB per second total bandwidth, compared to the external memory (RDRAM) which has only 3.2 GB/sec bandwidth.

Xbox's graphics processor, NV2A, has no embedded memory, not counting the small caches that are found in every processor.

there could be many reasons for the following, but I think that PS2 and Gamecube have more games that are 60fps, or at least smoother, because of the high-bandwidth embedded memory in their graphics chips.
with that said, with careful usage of Xbox's shared memory bandwidth, beautiful 60fps games can be achived, like PDO.

the downside to embedded memory is the massive amount of transistors it takes up. of Flipper's 51 million transistors, about half are from the 3.12 MB of embedded memory. of GS's 43+ transistors, about 2/3 to 3/4 of them are embedded memory.

 

[pc999]

Thanks Megadrive1988 and manux

 

[ERP]

Embedded memory has absolutley no bearing on 60fps.

The embedded memory can give you a big advantage when it comes to transparencies, and a lot of framebuffer blending can kill performance on Xbox. Having said that my experience with GC is that it runs into fill related performance problems before Xbox even with the embedded framebuffer.

Embedded memory has it's own set of tradeoffs, you are basically reducing the number of working transisitors on a chip to add the memory. Depending on what you percieve the bottlenecks in the final system to be it can either be a good thing or a bad thing.

 

[David_South#1]

Hey ERP,
Are you saying embedded has no effect in general or in relation to PS2?
As a matter 'fact' in some ways it's because of embedded memory that great PS2 games have to be 60FPS.
This has to do with the diminutive size of the Dual-port VDRAM and frame buffering.
PS2 uses its high bandwidth VDRAM and fillrate to pump interlaced lines out fresh every frame.
Then restarting each new frame.

What I've learned breaks down to about
1MB read, 1MB write, 1 MB Stored textures, 1MB Data.
The actually figure is that it produces about 12MB beyond this for each frame.
That comes from the aforementioned Fillrate and Bandwidth. Writing lines to a frame even as it is being drawn.
With a PC it had a larger Memory so whole frames can be done an stored with time to start the next.
PS2 requires some intense programming / system mojo to manage the visual magic.

Back to PS3, with a detail that’s a DOOZEY.
This is a V-DRAM figure that has had me holding my breath in anguish.
I can safely say to expect at least 128GB/sec. Truly not very shocking.
However my limited understanding sees the details differently.
Would you believe me if I told you the Cell GS VDRAM is rated at 512 GB/sec ???

I don't believe it myself. But one of the ways I digested the info say it could be.
The only reason I haven't tossed the idea right out the window...
Would you have believed the PS2's GS was going to do 48GB/sec two years before release?

Something to meditate and boggle on.

EDIT: I guess I lost track. Sorry about posting this in the ATI Xbox thread.
PM me a better place to put it and I will. Thanks.

 

[nAo]

As a matter 'fact' in some ways it's because of embedded memory that great PS2 games have to be 60FPS.

That's not the case.

Back to PS3, with a detail that’s a DOOZEY.
This is a V-DRAM figure that has had me holding my breath in anguish.
I can safely say to expect at least 128GB/sec. Truly not very shocking.
However my limited understanding sees the details differently.
Would you believe me if I told you the Cell GS VDRAM is rated at 512 GB/sec ???

The total (aggregated) Broadband Engine bandwith should be about 1 TByte/s, so your CELL GS EDRAM figure don't impress me much..I expect much more than that.

 

[ERP]

Hey ERP,
Are you saying embedded has no effect in general or in relation to PS2?
As a matter 'fact' in some ways it's because of embedded memory that great PS2 games have to be 60FPS.
This has to do with the diminutive size of the Dual-port VDRAM and frame buffering.
PS2 uses its high bandwidth VDRAM and fillrate to pump interlaced lines out fresh every frame.
Then restarting each new frame.
.

No not true, PS2 can and a lot of games do use full size frame buffers.
And for that matter Xbox can (although I no of no games that do) use field rendering.

The PS2 is an interesting point here, because they traded off a lot of features for that embedded RAM. Personally I think they traded off too much, but I know a lot of other developers who'd probably disagree.

 

[Megadrive1988]

GSCube(16) had 50.3 GB / sec bandwidth for the main RDRAM memory
(128 MB x 16)

and VRAM Bandwidth of 755 GB / sec (47.2GB/s x 16) from the combined bandwidth of 512 MB eDRAM (32 MB x 16)

so yeah I would hope that PS3's Broadband Engine and Visualizer have at least 1 TB of total bandwidth, each.

 

[David_South#1]

ERP,
Please name a 60FPS PS2 game that uses a full frame buffering.
Most PS2 games like MGS2 and GT3 use full even / odd frames.
But the frame does not contain both.

You have to undertsand that there is a read write buffer in effect.
I've never heard of a game that used a whole 2MB area to write a frame.

nOa and MegaDrive,
The both of you refer to "Engine Implementations"
My PS3 expectations consist of one PE and one VS elements.
I would like to note that if the VS is half the speed of the PE then 256GB/sec is my expectation.
Which would match Mega's 1 TeraByte/sec potential.

I am very pleased to see that my idea is not shocking.

 

[Paul]

The both of you refer to "Engine Implementations"

Emotion Engine, Handheld Engine, Wega Engine, Broadband Engine.

1.11 Broadband Engine. "Broadband Engine" means the Microprocessor designed
pursuant to the Playstation 2 for use in Playstation 3 Product.

I digress, it must be just an "Engine Implementation"

My PS3 expectations consist of one PE and one VS elements.

SCE has different plans, this I assure you.

 

[David_South#1]

Paul,
Please share a link to the controct you quoted.


It has been my personal feelings that the BE is part of the server side supporting a Cell network. But would not reside within a console.

It's also been my general feeling that PE's are packaged seperately.
That a BE would be some sort of Multi-Chip Module.

 

[ERP]

ERP,
Please name a 60FPS PS2 game that uses a full frame buffering.
Most PS2 games like MGS2 and GT3 use full even / odd frames.
But the frame does not contain both.

You have to undertsand that there is a read write buffer in effect.
I've never heard of a game that used a whole 2MB area to write a frame.

No offense but you have absolutely no idea what your talking about.

Basically your argument seems to boil down to Sony made it difficult to use a full size frame buffer, so developers wrote more 60fps games. Wow why didn't MS and Nintendo just remove the bit of the API that lets developers use a fullsize frame?

There might actually be an element of truth to that, but it hardly links 60fps to embedded RAM, it just links it to a lack of VRAM embedded or otherwise.

Your also laboring under the misunderstanding of what progressive output costs you in a PS2 game. For the most part, the only real cost is a reduction in maximum texture size. If your using reasonably sized vertex packets, the DMA transfer of a texture will probably be completed before you finish transforming the first packet, so all you really need in VRAM along with the frame buffer is a double buffered texture.

And in answer to your question, although I have to admit I've stopped looking at PS2 games in any real detail, I believe that Baldurs Gate is 60fps and uses a much larger than 640x480 framebuffer.

 

[Ug Lee]

And for that matter Xbox can (although I no of no games that do) use field rendering.

I think Kung-Fu Chaos uses field rendering.

Please name a 60FPS PS2 game that uses a full frame buffering.

Burnout 2, Jak 2, ESPN NBA, ESPN NFL, Ratchet and Clank 2, Soul Calibur 2, Tekken 4, SSX3.

Those are just off the top of my head.

 

[Megadrive1988]

It's also been my general feeling that PE's are packaged seperately.
That a BE would be some sort of Multi-Chip Module.

lets say that BE is going to be one chip. 4 PEs on one chip form a BE.

what would be really cool is, then we have 4 BEs in some sort of MCM, and thus giving us multi-TFLOP performance. probably won't happen...

...and I'm not saying that your feeling about PEs packaged seperately and therefore a BE would be some kind of MCM is wrong, though.

 

[marconelly!]

Most PS2 games like MGS2 and GT3 use full even / odd frames.
But the frame does not contain both.

MGS2 uses full frame buffer, and you can play it in progressive scan mode if you force it using the blaze VGA adapter. Early PS2 games often used half frame buffer like you suspected, but nowadays it's rarely being used anymore, no matter if the game is 30 or 60FPS.

 

[bbot]

No. R500 is based on the old R400

That is the popular thought, thanks to the new report that came out when ATI got the XB2 deal, however I now have clear information that points this assumption to being incorrect.

Will you hear more? Yes, hopefully, but not just yet.

Yes. I'd like to know what happened to Nintendo's contract with ATI. Has it been cancelled? Or is the same vpu be used for both Nintendo and Microsoft next-generation consoles?

 

[Guden Oden]

It's also been my general feeling that PE's are packaged seperately.
That a BE would be some sort of Multi-Chip Module.

You have any idea how EXPENSIVE a MCM gets with a large amount of pins involved? Pretty darn, that's what!

We've been over this territory before. Multichip setups just complicates things unneccessarily both from a programming standpoint and an engineeering one. It's not ever going to be a transparent jump if you chop up a 4PE broadband engine into four single PE chips. In the BB, you have ONE LARGE on-chip memory, in the 4x single PE version there's four much smaller, separate pools where data has to be copied back and forth (hogs bandwidth) and/or duplicated (hogs space). It's impractical.

Then there's the matter of external memory. BB only needs one external memory interface, say 4 XDR modules in parallel, four single PEs would either have four modules connected to ONE PE chip (which means the interconnecting links between chips would become saturated when other chips request access to main memory, thus starving APUs of data when they request access to other chips' on-chip memory, and the "further-most" PE would suffer considerable latency when accessing RAM through two intermediary chips, assuming there's a ring-shaped network between the separate PE chips). Bad idea. OR, you'd have one XDR module per chip, with only a quarter bandwidth per chip compared to a full BB implementation AND the same data copying and duplication issues as with on-chip memory.

I don't think this is the path Sony will take, for many obvious reasons. Can we now just forget this idea, please?

Edit: ooh, forgot one thing! The same issues regarding interfacing with the GPU and a possible I/O + sound chip would exist in the multichip approach as with access to main memory, except Sony couldn't very well quadruple interfaces on other chips so all PEs can have a dedicated link... That doesn't work. They'd have to go through other PEs, maybe have one interfacing with the GPU and another with the IOP/sound chip and thus cause traffic jams on the internal network. Again, this is a bad idea.

 

[nAo]

I don't think STI will go with MCM.
IMHO they will just add a redundant APU per PE, like a patent by IBM (the inventor is CELL chief architect) suggests.

 

[Fafalada]

Most PS2 games like MGS2 and GT3 use full even / odd frames.

If by "most" you mean 5%, then yeah, "most" PS2 games use half-height buffers.

 

[David_South#1]

Thank you Falfalada, marconelly, Ug Lee, etc
Dang, in regards to PS2 VDRAM I’ve been living in my ignorance for quite a while as to how good games for the PS2 must operate.

I had "The Document of MGS2" and got this initial impression from it. Then when talking with friends in another forum that are somewhat PC bias they explained that they doubt 60FPS PS2 games use full frame buffers. Though I haven’t repeated this erroneous understanding elsewhere, you’ve made it clear that I must be mistaken in this view.

ERP,
Basically you boiled it down about right. (Harsh, but fair.) I was also implying that initially it was very difficult for them to do 60FPS on the PS2. (By including the extra crappy video use explanation.) And that with time things have gotten better, kind of like what Marconelly said when he was correcting me.

But I had no idea that SSX3 & Balder’s Gate used full frame buffers. This in and of itself tells me that I am clearly wrong. So, I accept that I must be in error as to the constraints of the VDRAM.

My apologies, from David South.

 

[David_South#1]

Guden Oden,
Yes I know MCM’s to be very expensive. That’s why I figured just one element of each, PE and VS. Two chips. Because as of yet in regards to the BE, I am unable to perceive 32 pair of VU0&VU1 on a single chip with 4 PU (Whether it be PPC440 or MIPS64 20Kc), and 4 DMACs.

At no time have I or am I going to suggest an MCM in a console.
It is because I believe Broadband Engine to be an MCM that I stand against suggestion of its use in a console. The only Cell MCMs I can imagine are to be in Servers for Video on Demand and Cell Network hosting hubs, not in consoles.

The GScube is representary of a Visualizer Engine (4x VS).
To support SCE’s Final Fantasy Movie Demos, they used an SGI Origin 3000.

Of that I have this excerpt,

For SCEI, the SGI Origin 3000 series represents "the kind of power needed to fuel the next generation of broadband entertainment," said Ken Kutaragi, president and chief executive officer, Sony Computer Entertainment Inc. "We have a vision of the future of entertainment and SGI is an integral partner in demonstrating that vision."

Do ya’ll really think something like that will be used in a console?

I would very much like to continue discussing you second paragraph on Memory.
But I’m going to take a break to better absorb it better.
The first I completely follow. The second is a little fuzzy to me.

Also,

Administrators.
Please if it is possible seperate this Hijacking into a new thread.
I feel that my activity takes away from the topics value.

Title it as you best judge. May I suggest "Cell: South's Perspective".