ATI X1900X: Ahead of the pack

[KeepItRed]

Not sure if this is a repost, but this looks impressive:

X1900

Now we just have to wait on the specs. for the number of pipelines and clock/memory speeds.

 

[Nirach]

Psht.

Look at the screenshot. It's just an edited 1800.

I bet it's even crossfire, to get that score.

 

[KeepItRed]

Originally posted by: Nirach
Psht.

Look at the screenshot. It's just an edited 1800.

I bet it's even crossfire, to get that score.

The 3DMark score probably is FUD, but I meant the first screenshot showing the statistics compared to the X1800.

 

[Nirach]

Originally posted by: KeepItRed

Originally posted by: Nirach
Psht.

Look at the screenshot. It's just an edited 1800.

I bet it's even crossfire, to get that score.

The 3DMark score probably is FUD, but I meant the first screenshot showing the statistics compared to the X1800.

The first one is slightly more believable, but I'm still going to doubt it somewhat until the thing hits reviewers, to be honest.

 

[Keysplayr]

What happened to the fancy schmancy ring bus?

 

[nts]

Originally posted by: keysplayr2003
What happened to the fancy schmancy ring bus?

Its still there...

 

[Drayvn]

One thing im wierded out about is the quad pipelines.

The thing that throws this into the FUD packet for me is that the Ultra Threaded processor can only speak to 4 of the quads and then the quads have to speak and pass info through each other to output it. Isnt that way kinda like really slow?

 

[Munky]

Originally posted by: Drayvn
One thing im wierded out about is the quad pipelines.

The thing that throws this into the FUD packet for me is that the Ultra Threaded processor can only speak to 4 of the quads and then the quads have to speak and pass info through each other to output it. Isnt that way kinda like really slow?

The way I understand it is that the main scheduler only controls the quads, not the individual shaders within each quad. In the case of the r580, each quad will have 12 pixel shaders, and from there each "quad" of shaders within the quad will have access to a quad of texture units. It may be turn-based or thread-based scheduling within each "12-group" of shaders, but I'm fairly certain that one quad does not have to pass it's data through the 2 other quads to access the external units.

 

[Wreckage]

They must have been using this Crossfire setup. Notice the dongles for increased 3dmark score.

 

[Shamrock]

Originally posted by: Nirach
Psht.

Look at the screenshot. It's just an edited 1800.

I bet it's even crossfire, to get that score.

Yep, look at the first 3dMOCK scores. I agree with you. Where it says what vid card you have. It's edited out, and they didnt bother to correct the brush size, because part of the X and 1 is blotted out. Also, if they wanted to tout the ATI card, why would they leave out AA !?!? I know it would hinder the score, but ATI is knwon for it's AA

 

[dunno99]

If that first screenshot is to be believed, then we can compute the number of ALUs per shader right now given known information (like 625MHz) and settle that question once and for all. Even if known information deviates by tens of MHz from the real thing, we can still draw correct conclusion with some reasonable rounding.

 

[dunno99]

From the numbers, it seems like the core will be 600MHz (650MHz * 166 / (3 * 60)) and will have 1 ALU per shader ((553.8 * (650 / 600) / 48) / (272.5 / (16 * 1.5)) approx= 1, or it means that the core will run at about 650MHz ). It seems to me that the single ALU in the R580 will be doing two instructions in one pass on all three components or something, otherwise I don't know how it's getting the 166 shader ops number while only maintaining a 2x flop count on a 3x number of shader gain (after normalizing for gpu clock). Obviously, these are all "theoretical" performance and real-life scenarios will be less.