More Xenon rumors

[overclocked]

DaveBaumann wrote

And it was significantly less complex (1 Vertex Shader to NV2A's two, half the FX Shader ALU's, lacks double stencil / Z).

That´s interesting. For "historical" reasons it would be nice to compare the GF3/4 in depth.

 

[Fox5]

aaaaa00 wrote

No. For xbox the specs were always for a 600 mhz processor AFAIK. That was raised to 733 mhz.

I think the 600Mhz cpu was a AMD Duron so that was probably responsable for the change. IMO the Duron would have been a better choise than the Celeron that now are in the Xbox.

I think it was already argued here that the xcpu is better than a celeron.

And how would the duron be a better choice? I don't think durons performed any better than p3 based celerons, and they were much hotter weren't they?

 

[Megadrive1988]

the original X-box specifications as widely reported online and in print, in 1999, called for an AMD Athlon 500 and Nvidia NV10-GeForce256 graphics.

http://groups.google.com/groups?selm=7qpgdp$go6$1@news.3dfx.com&output=gplain

 

[vliw]

I think you worry too much. By the time nextbox comes out, ATi will have been doing 256-bit buses for half a decade. They know how to implement this sort of stuff on the board level.
high clock and width bus is not so easy to develop and here we are talking about chips that are clocked very high

No, we're not. 25-ish GB/s is not very high, you can achieve more than that with plain DDR memory and nextbox will have DDR2 or GDDR3 mem.
.

The thing is, that you don't need large amount of bandwidth between main ram and edram, try to do some calculations by yourself, what kind of bw you need to transfer all the memory from main memory to edram?
If you upload 256 MB of data for one frame and multiply this number for 60(frames /sec) you see that the need for bandwidth is 15 GB/s...but for the full ram that is not likely to happen.
So with 25 Gb/s you have plenty of bw to use and you don't need a very expensive 256 Bit bus.
The thing can change if you think about bandwidth between frame buffer(EDRAM) and chip logic, for 3D graphics rendering the more, the better so if you can't use EDRAM(like pc card) you must do something else like GDDR3 and 256-512 bit because this is the only way you can gain much more bandwidth.

Besides, it's not as if they'd need to reinvent the wheel or anything. The memory chip traces can pretty much be copy/pasted from a graphics card if they want to.

Yuo can't develop complex chip with cut and paste, the process is more complex.

What makes you think they'd need to spend a lot?

Because high/large bus require to take care about pcb design for electromagnetic compatibility, for example, you have a cpu, memory and peripherals clocked very high(we are talking about a monster that works at 3,5 Ghz) so if you wish your monster works you must spend much more time to design and test and so you need more time/money.

eDRAM is typically not very big. It can't hold all geometry/texture data, and CPU needs plenty of bandwidth too - in addition to network, other I/O etc etc.

Xenon will be a stable platform for about 5 years so you can design how much ram you need to do everithing.
Engineers are not so stupid trust me.
Game cube has 3 Mb of EDRAM and is not a top notch machine, do you like metroid prime?


p.s. IMHO...

 

[jvd]

in a year and a half with ram that slow i wonder how tight they can get thier timmings .

I can get my ddr 400 up to 290mhz but it runs slower than if i had it at 271 with cas 2.5 and the other timings tighter .

As aposed to 290 with cas 3.0 and the rest looser

 

[overclocked]

Fox5 wrote

I think it was already argued here that the xcpu is better than a celeron.

And how would the duron be a better choice? I don't think durons performed any better than p3 based celerons, and they were much hotter weren't they?

As i remember it the Duron was faster than the Celeron in almost everything, my memory can be lacking of course. IF AMD had been in the Xbox i think it´s quite possible that it had been clocked higher, maybe 800Mhz who knows. And the K7 core has/had a very good and fast FPU compared to the P6 core. IMO it would been better but as i understand it
AMD wasn´t very happy with the bid they had while Intel basically "gave away" their cpu´s.

 

[Gubbi]

Actually the P6 core does surprisingly well. Look at the SpecINT scores (1528) for the newest 2GHz Dothan (which basically has a P6 OOO execution core).

Also the XCPU has a 133MHz bus interface, double that of celeron. And even though the level 2 cache is the same size, 128KB, it has twice the associativity.

Cheers
Gubbi

 

[Fox5]

I think you worry too much. By the time nextbox comes out, ATi will have been doing 256-bit buses for half a decade. They know how to implement this sort of stuff on the board level.
high clock and width bus is not so easy to develop and here we are talking about chips that are clocked very high

No, we're not. 25-ish GB/s is not very high, you can achieve more than that with plain DDR memory and nextbox will have DDR2 or GDDR3 mem.
.

The thing is, that you don't need large amount of bandwidth between main ram and edram, try to do some calculations by yourself, what kind of bw you need to transfer all the memory from main memory to edram?
If you upload 256 MB of data for one frame and multiply this number for 60(frames /sec) you see that the need for bandwidth is 15 GB/s...but for the full ram that is not likely to happen.
So with 25 Gb/s you have plenty of bw to use and you don't need a very expensive 256 Bit bus.
The thing can change if you think about bandwidth between frame buffer(EDRAM) and chip logic, for 3D graphics rendering the more, the better so if you can't use EDRAM(like pc card) you must do something else like GDDR3 and 256-512 bit because this is the only way you can gain much more bandwidth.

Besides, it's not as if they'd need to reinvent the wheel or anything. The memory chip traces can pretty much be copy/pasted from a graphics card if they want to.

Yuo can't develop complex chip with cut and paste, the process is more complex.

What makes you think they'd need to spend a lot?

Because high/large bus require to take care about pcb design for electromagnetic compatibility, for example, you have a cpu, memory and peripherals clocked very high(we are talking about a monster that works at 3,5 Ghz) so if you wish your monster works you must spend much more time to design and test and so you need more time/money.

eDRAM is typically not very big. It can't hold all geometry/texture data, and CPU needs plenty of bandwidth too - in addition to network, other I/O etc etc.

Xenon will be a stable platform for about 5 years so you can design how much ram you need to do everithing.
Engineers are not so stupid trust me.
Game cube has 3 Mb of EDRAM and is not a top notch machine, do you like metroid prime?


p.s. IMHO...

I think gamecube was severely hurt by only having 3MB of video ram, and then having less than 2GB of bandwidth(after the cpu uses some) to main memory. I think gamecube would have been helped a lot more by having the same memory system as the xbox.

 

[vliw]

[
I think gamecube was severely hurt by only having 3MB of video ram, and then having less than 2GB of bandwidth(after the cpu uses some) to main memory. I think gamecube would have been helped a lot more by having the same memory system as the xbox.

I don't think x-box has been a well designed console, and i think that Xenon will be far more advanced from a concept view(i'm not talking about performances)..imho....

 

[Megadrive1988]

I don't think x-box has been a well designed console, and i think that Xenon will be far more advanced from a concept view(i'm not talking about performances)..imho....

agreed.

 

[Fox5]

[
I think gamecube was severely hurt by only having 3MB of video ram, and then having less than 2GB of bandwidth(after the cpu uses some) to main memory. I think gamecube would have been helped a lot more by having the same memory system as the xbox.

I don't think x-box has been a well designed console, and i think that Xenon will be far more advanced from a concept view(i'm not talking about performances)..imho....

It may not have been all that well designed, but I think its strengths outweigh its disadvantages.(and you have people who argue that the less than 6.4GB of bandwidth to its graphics chip are more than enough)
Gamecube's graphics chip is less powerful than xbox's, so less than 6.4GB of bandwidth should be more than enough. I think gamecube would benefit far more from having a substantial amount of video ram and half the video bandwidth than what it currently has.

 

[archie4oz]

Also the XCPU has a 133MHz bus interface, double that of celeron. And even though the level 2 cache is the same size, 128KB, it has twice the associativity.

Actually XCPU *is* a mobile celeron...

 

[AlNom]

I don't suppose there's a way to put in something beefier than a celeron, eh? or faster RAM or faster HD....

 

[Guden Oden]

The thing is, that you don't need large amount of bandwidth between main ram and edram, try to do some calculations by yourself

You're still stuck in your rut. Main memory is not ONLY used for just GPU purposes, man! That's why it is called main memory!

Also, while the system features memory bandwidth-saving features like GPU reading data from CPU L2, this can't be used at all times. If the GPU is drawing buffers larger than will fit in 10MB of eDRAM (will happen pretty much immediately with floating-point framebuffers) the GPU will have to partition up the framebuffer and finish one at a time. However, that means it can't read polys straight from L2, because the CPU might not be working on polys located in the current section of the framebuffer.

So with 25 Gb/s you have plenty of bw to use and you don't need a very expensive 256 Bit bus.

Stuck in your rut again I see. What on EARTH makes you think a 256-bit bus is "very expensive"? It's just traces on the mobo, and mobos aren't particulary expensive.

On the contrary, if you DON'T go 256-bit, THEN the memory interface gets very expensive since it will require esotheric 1500MHz+ memory chips which barely exists yet vs rather pedestrian 720-something MHz mem at 256-bit. 720 is a cake-walk, 1500+ is not. Remember the 12-layer board Nvidia had to use for its dustbuster GFFX with 1GHz DDR2 when ATI did 256-bit at 8 layers or maybe it was even less and got way more total bandwidth?

Yuo can't develop complex chip with cut and paste, the process is more complex.

I wasn't talking about the CHIP, read my post before you reply...

Because high/large bus require to take care about pcb design for electromagnetic compatibility, for example, you have a cpu, memory and peripherals clocked very high(we are talking about a monster that works at 3,5 Ghz) so if you wish your monster works you must spend much more time to design and test and so you need more time/money.

Dude, you're not making any sense here. CPU's (assumed) speed is 3.5GHz yes, but that's only INTERNALLY, so has nothing to do with main mem bus width. Also, a wider bus can be run slower, so you actually get LESS noise produced. Not more.

Xenon will be a stable platform for about 5 years

Oh yeah, like the current box will be a 5-year platform?

so you can design how much ram you need to do everithing.
Engineers are not so stupid trust me.

Uhm....yeah. Sure dude, whatever you say!

I guess no nextbox game will use levels with art assets that exceed a sum total of 10MB then, including screen buffer...

Game cube has 3 Mb of EDRAM and is not a top notch machine, do you like metroid prime?

Remind me, where does GC factor into this discussion?

 

[archie4oz]

I don't suppose there's a way to put in something beefier than a celeron, eh? or faster RAM or faster HD....

My friend has a 1.4GHz Tualatin in his and a 120GB 7200rpm HDD...

 

[Fox5]

I don't suppose there's a way to put in something beefier than a celeron, eh? or faster RAM or faster HD....

You can put in all those things. Fastest and still compatible cpu available is a normal pentium 3.(there's also a 1.4ghz celeron, p4 based maybe? but it breaks compatibility with many games) You can't put in faster ram(where would you buy it, and how would the motherboard recognize it?) you can double the ram as well, and you can replace the harddrive completely, as well as add in an 80 pin cable(instead of 40) to speed up load times.

 

[Quaz51]

on the Xenon diagram it's 10Mo of EDRAM (enhanced DRAM) not eDRAM (embedded DRAM)

 

[Megadrive1988]

does that mean the 10 MB is not embedded into Xenon's graphics processor?

 

[Quaz51]

does that mean the 10 MB is not embedded into Xenon's graphics processor?

Not in this diagram

it's 10MB of EDRAM (external) Not eDRAM

 

[Fox5]

does that mean the 10 MB is not embedded into Xenon's graphics processor?

Not in this diagram

it's 10MB of EDRAM (external) Not eDRAM

How misleading.

I still find it strange how 3dfx decided 4MB was the bare minimum a 3d accelerator needed to have to be useful, but nintendo got by on compression and transferring to main memory, and now xbox 2 is only going to have 10MB of dedicated ram? Come on, you can't even do 32bit color in that!