Radeon X800PRO will beat NV40 Ultra

[digitalwanderer]

Hmmm... So even "the big sites" don't have the R420 yet? If that is true, will we see *any* (p)reviews on the 26th then? Or does *no* hardware review site have a R420 board yet?

I'm pretty sure "the big sites" don't have one, ("the big3" SUCK for leaking info IMHO), but I'm pretty sure one site has one and maybe one other. (I'm guessing that B3D has one already, they're pretty well trusted.)

I don't KNOW that, I'm guessing on this one.

 

[madshi]

I'm pretty sure "the big sites" don't have one, ("the big3" SUCK for leaking info IMHO), but I'm pretty sure one site has one and maybe one other. (I'm guessing that B3D has one already, they're pretty well trusted.)

I don't KNOW that, I'm guessing on this one.

Okidoki, thank you.

 

[jolle]

um how about its just a faster architecture? 8)

fair enough, that is a reason, but where do you take that from, what have
you seen or heard that leads you to claim that the chip will be alot faster
then the NV40..

Ok, lets just try this..

NV40
400Mhz or 475Mhz, we can use the 400Mhz as its the official one at this time.
16 pipes (same old tech for Z passes which is 32x0 in doom3 zpass, might not count that tho, since its not clear to me if anything else uses that cept D3 engine yet)
500-600 mem estimated, somewhere around 1100Mhz..
2 PS units per pipe, 32 in total,
1 TMUs per pipe..
6 VS units..
Fillrate: 6400 Mtexels/sec.. (if core is 400, at 475 = 7600)

Now, that is just numbers and theoretical values, put R420 next to this
and do some real explaining on what if the foundation of your certainty
if you please..
Im not trying to prove you wrong, Im just hella curious, and you seem
like you know something unless your just assuming things..

 

[beancounter]

Why the X800 XT

There are two reasons that the X800 XT might be released early, say two weeks after the Pro. The first is that X800 XT is absolutely necessary to compete against the 6800 Ultra, but then why not wait until after the 6800 Ultra releases (that way, ATI would know what clocking was needed to beat the Nvidia product)? The second reason might be that originally the X800 XT was planned for later release because it was assumed that yields for the faster clocked and flawless part was going to be poorer than it turned out to be, meaning there are more 16P parts (and fewer 12P parts) than the original marketing/production plans called for.

If the first suggestion is true, ATI are being pretty dumb about when to release what part. If the second is true, the good news might be that X800 XTs could be cheaper than was originally planned, but the bad news might be that X800 Pros could become more expensive, to try and shift demand to the higher-end part. We'll see.

By the way, for those wondering why make the XT at all if the Pro turns out to be a similar speed or faster than the 6800 Ultra, we've all noted that the R420 die was a 16P part all along. There would be no reason to have used the extra area to allow for 16 pipes unless ATI always planned to release a 16 pipe part. Even if they were the only ones in the graphics market, they would release using the full number of pipes; maybe the chip would be declocked severely, pushing the yield up, but you'd get 16 operating pipes.

Beancounter

 

[Frank]

Does it makes sense to develop an in-between technology (VS/PS 3.0) when you already have an outstanding one based on the current and near-future technology and another one that is almost ready that is based on the future standard (and are in bed with the ones that make the specs for that one)?

Why spend time developing that in-between technology in the first place?

Better AA, now that is something everyone can appreciate, it even improves old games. And it is something that is hinted at a lot for the R420.

ATi might even take the lead in AF, if they improved upon it, as nVidia has copied the ATi way of doing that...

 

[PaulS]

Re: Why the X800 XT

By the way, for those wondering why make the XT at all if the Pro turns out to be a similar speed or faster than the 6800 Ultra, we've all noted that the R420 die was a 16P part all along. There would be no reason to have used the extra area to allow for 16 pipes unless ATI always planned to release a 16 pipe part. Even if they were the only ones in the graphics market, they would release using the full number of pipes; maybe the chip would be declocked severely, pushing the yield up, but you'd get 16 operating pipes.

If they were the only ones in the market, they wouldn't have designed it as a 16 pipe chip to start with.

 

[Hellbinder]

I dont suppose you have concidered the idea that

1. ATi always planned on releasing the XT at this time

2. Has an R450 in preperation right now for release later this year.

3. Is releasing the X800XT simply to establish a *Dominant* leadership position to aid their Goal of gaining 60% market share this year.

 

[madshi]

2. Has an R450 in preperation right now for release later this year.

I thought it was code named R480?

 

[Unknown Soldier]

Well Jolle lets have a look(take note if this info is real).

Speculation 1

ard|OCP]reports that various versions of the R420 will be branded as RADEON X800 PRO, RADEON X800 XT, RADEON X880 XT and RADEON X800 SE citing a “classified ATI documentâ€￾. The solutions will enable different number of rendering pipelines: 12, 16, 16, 8 and will function at clock-speeds of 500MHz, 600MHz, 600MHz and 450MHz respectively. Higher-end graphics cards will have 256 or 512MB of GDDR3 memory functioning at 1000 and 1200MHz for “PROâ€￾ and “XTâ€￾ parts. The performance-mainstream offering RADEON X800 SE will boast with 800MHz DDR memory with 128-bit bus.

So lets see.

X800XT
256bit bus
600Mhz
16 pipes <-- Extreme Pipes??
600Mhz GDDR3(1200Mhz)
1 TMUs per pipe
Fillrate: 9600 Mtexels/sec <-- If Clock speed is true of course.

X800Pro
256bit bus
500Mhz
12 pipes <-- Extreme Pipes??
500Mhz GDDR3(1000Mhz)
1 TMUs per pipe
Fillrate: 6000 Mtexels/sec <-- If Clock speed is true of course.

X800SE
128Bit bus
450Mhz
8 pipes <-- Extreme Pipes??
400Mhz GDDR3(800Mhz)
1 TMUs per pipe
Fillrate: 1800 Mtexels/sec <-- If Clock speed is true of course.

----------------------------------
Speculation 2

As for the specs, the R420PRO or should we call it X800PRO will be clocked at 475 MHz, 25 MHZ higher then we suggested since yields are better, while the memory will be clocked at 900MHz as we suggested before.

Everything is packed into 180 millions of transistors and the card has 12 pipelines as we suggested. The R420XT, Radeon X800XT will be higher clocked card with 16 pipelines which came as a huge surprise to us.

http://www.theinquirer.net/?article=15377

X800XT
256bit bus
500Mhz ?? Just says the XT will be clocked higher than the Pro
16 pipes <-- Extreme Pipes??
550Mhz GDDR3(1100Mhz) ?? Doesn't say so lets say 550 GDDR3
1 TMUs per pipe
Fillrate: 8000 Mtexels/sec <-- If Clock speed is true of course.

X800Pro
256bit bus
475Mhz
12 pipes <-- Extreme Pipes??
450Mhz GDDR3(900Mhz) ??
1 TMUs per pipe
Fillrate: 5700 Mtexels/sec <-- If Clock speed is true of course.

------------------------------

Speculation 3

So just like the 6800 Ultra, the X800 Pro will be fabbed at 130nm. It will support GDDR 3 memory across a 256-bit bus clocked to 1GHz. It to is priced at around $399 and will initially appear in AGP 8x boards. The X800 Pro's core is clocked to 500MHz.

Unlike the Ultra, the X800 Pro contains 12 rather than 16 pixel pipelines. However, its big brother, the Radeon X800 XT, due to chip on 31 May, according to the report, will feature 16 pixel pipelines, along with a core clocked at 600MHz and memory running at 1.2GHz.

In the same timeframe, ATI will release the R420-based X800 SE, a lower end part clocked at 450MHz and with only eight pixel pipelines enabled. The memory bandwidth is halved too, with the bus set at 128 bits wide. It will use regular DDR SDRAM, and only 128MB of it

http://www.theregister.co.uk/2004/04/15/ati_r420/

X800XT
256bit bus
600Mhz
16 pipes <-- Extreme Pipes??
600Mhz GDDR3(1200Mhz)
1 TMUs per pipe
Fillrate: 9600 Mtexels/sec <-- If Clock speed is true of course.

X800Pro
256bit bus
500Mhz
12 pipes <-- Extreme Pipes??
500Mhz GDDR3(1000Mhz)
1 TMUs per pipe
Fillrate: 6000 Mtexels/sec <-- If Clock speed is true of course.

X800SE
128bit bus
450Mhz
8 pipes <-- Extreme Pipes??
400Mhz GDDR3(800Mhz)
1 TMUs per pipe
Fillrate: 1800 Mtexels/sec <-- If Clock speed is true of course.

The Register and HOCP's specifications look the same.. so might have the same source. The Inq's source specs are lower than the other two.

US <-- Corrections made

 

[jolle]

Tnx alot man, finally some decent info...

What does the Extreme pipeline mean?
the XT looks really nasty hehe..

no word on shaderunits yet?

 

[Dave Baumann]

WRT to texture to pipes. Think about what is sensible - with DX8 the texture to math ratio was going in the direction of 1:1, so even then two TMU's is beginning to make less sense, the more we get into DX9 and beyond, with longer shaders, the more the ratio is biased to the Math side.

 

[martrox]

For the most part, that's all BS........ Don't believe any of it!

edit: Not what Dave said....but what [H], The Register & the Inquirer said.

 

[anaqer]

Why spend time developing that in-between technology in the first place?

Hm let's see... maybe your main rival having that tech and marketing it as the best thing since sliced bread has something to do with it...?
Nah, who would care about that.

 

[Evildeus]

X800=15k 3dm2k3 default.

Any other questions?

In your dreams?

 

[Evildeus]

26th premiere, 5th-o-May available in stores.

nVidia will be available in stores in July sometime.

Let me doubt both parts

 

[Tim]

X800=15k 3dm2k3 default.

Any other questions?

In your dreams?

If (big if) the XT is 16 @ 600Mhz, 15K does not seem out of the question.

 

[Unknown Soldier]

For the most part, that's all BS........ Don't believe any of it!

That may be Martrox.. but it's the only info out there atm. Until Dave and the others release the final details next week26th April(?? or the week after i.e 5th may) then we'll know for sure.

Jolle: Just in case you didn't catch what extreme pipes is -

WRT to texture to pipes. Think about what is sensible - with DX8 the texture to math ratio was going in the direction of 1:1, so even then two TMU's is beginning to make less sense, the more we get into DX9 and beyond, with longer shaders, the more the ratio is biased to the Math side.

US <-- Dave if those numbers of the HOCP and TheReg are right would the fillrate I added for the X800SE be right?? I take it's Clock Speed x Pipe Count divided by 2(for 128bit bus) = Fillrate??. For 256bit bus I see it's Clock Speed x Pipe Count = Fillrate

 

[Evildeus]

On topic, if the X800 pro is faster than the 6800U that would be stunning. But then why release the XT?

ED, maybe because ATI is going for the jugular? I've noticed all this supposition about how the 16 pipe is being released because of the capibilities of the NV40.... yet ATI hasn't ever said much about anything. Few here have mentioned the possibility that the 16 pipe R420 was inteneded all along.......

Of course, but till then that hasn't been Ati strategy (which could change of course), and according to HB (great Ati's evangelist) the XT part is made without any profit (at loss) so let me be puzzled (even if i think this INQ's rumor will be wrong )

 

[jolle]

Eh.. so its sorta like NV40 with 2 PS units and 1 TMU on each pipe?

 

[Frank]

Why spend time developing that in-between technology in the first place?

Hm let's see... maybe your main rival having that tech and marketing it as the best thing since sliced bread has something to do with it...?
Nah, who would care about that.

Unless your recent parts are so much faster it doesn't matter and your future parts with the next tech are almost ready to roll.