ATI Radeon X800 XT Platinum Edition / PRO Review

ATi is ATI and ... is ...

Mmmmm

Performance, die size and frequency bring some old stories to my memories... But with somewhat changed... I wander what will happen this time ? People will stick to the brand or to their former conclusions ?

Interesting...

But the R420 looks like a bench-killer : Build to bench good on old benchs. Normal mapping is just a cheap displacement mapping, and nothing new but better performance... It's an ATI like card : Very good, but inovation left to the others. A bit disapointing for a new generation chip. I don't really expect great improvement with new drivers, and the core clock max seems to be close to the PE clock. So...
What's next ?

On the other hand, The NV40 has room for OC, and it looks like the arch is a 512bits-32pipes-youcanaddVSifyouwant capable. Yes... what a die size... this would fit well in a 2 slot solution... lol. So stop kidding now.

Alway he same scheme : ATI secure & NVIDIA nutts. May this war last a long time because I'm an happy consumer.

Kriptomik

 

DB,
Great review. Is there any way you could give us a comparison between 3DC and the DX format (DXT5?)? Thanks

*edit* At what time will we have a look @ the other Dave interview?

 

Please be more careful folks, it is not Temporal AA. It is "Temporal FSAA": essentially it means that they change the Spatial AA sample pattern from one frame to another. Sure it is a well-thought, useful hack; but it is still just a form of (Sparse Grid Multi Sampling) Spatial AA.

Temporal AA is different: you actually blend the consecutive frames or parts of frames together (in an accumulation buffer) before displaying, perhaps to do soft shadows or motion blur, or to mimick 60fps on a slow 30fps-capable LCD. "Temporal FSAA" has nothing to do with Temporal AA. R420 doesn't "have" Temporal AA. (But can you do the same thing through multiple render targets and framebuffer blends?)

Just thought I'd clarify this one for the newcomers. Shame on you ATI, you have been so refreshingly clear and informative so far; don't you start mudding up our established terminology...

 

Re: R420 can not output 16 pixels / clock

Fallacious? I think not.

If you re-read what I said, I understand about the die size and the difficulties they carry.

But all things considered, the cards are close enough in performance that both vendors are playing with mhz to win. Which is why you have:

Ultra's
XT's
Pro's
GT's
Whatever...

 

x800 XT and Next-Gen Unreal Engine Capability

What I'm really interested in knowing is... how capable is the x800 with the next-gen graphic engines (forget Half-Life 2 or Doom 3; we know they run fine even on 9800 hardware).

Everyone knows or has seen the Unreal Demo footage, which was supposed to run "decently" on the Geforce 6800, does the lack of PS 3.0 (?) etc have any effect on this for the x800?

For those who are gonna argue that this engine is still far down the line, bear in mind most people want a card which can last at least 3 - 4 years (I think).

Thanks.

 

It was running at the same time on a X800 in another room you know

If you make a quick search in the technology forum you should find that.

 

Great review Dave 8)


The avarage consumer looks at price/speed this was always the case in past generations and will be so for generations to come......
ofcourse brandname counts for a lot too but the whole ps 3.0 thing the avararge joe smoo couldn't care less......

 

I was wondering: is the Z-buffer the only part that uses virtual memory?

This is quite probably a spin-off of the XboX2 and SM 4.0 design, the first needs virtual memory for the EDRAM, while it is part of the specs for the second.

Is it a general virtual memory implementation, or does it work only for the Z-buffer? (Not that there are that many other buffers that could benefit, but interesting anyway.)

 

What's the verdict on this? Is ATI's new method just misnamed spatial AA or should it be counted as temporal AA (although there is no blending of temporally separate results)?

 

Call it what you want. Your brain does the blending. I am interested to see what modes people can get working and at what is usable on a X800.

 

Whats described there is not virtual memory. The chip has a buffer that is used to store the lower levels of the hierarchical Z-Buffer (the full size Z-buffer is stored in main memory). If the target being rendered to is too big then only part of the target will have a hierarchical Z-Buffer, the rest of the target will only have a standard z-buffer.

 

Ah, yes, I see that I didn't read it well enough. Thanks for pointing it out.

 

That's a way to make things clear

In that case it isn't Temporal AA, not by the established definition. (Which I think refers to a specific, implemented method, not just what happens to happen.)

If ATI's method only amounts to varying the spatial sample pattern from frame to frame, then it is just a Spatial AA method. I think ATI's cheating here

 

Established by who? got a link?

Well duh.

I hate to break it to you but motion pictures are an illusion.

When someone creates a true display for pc's, you let me know.

 

Surely I'm not qualified to determine what it is, and I'm hesitant to call things whatever I want -- making up my own words tends to make discussions more difficult

Okay, seriously now. As to the method, Simon says here that it's not temporal AA and his judgement is good enough for me. Apparently the "brain does the blending" doesn't qualify for TAA; you have to have an accumulation mechanism. In other words, a TAA implementation has to blend the temporal samples in a controlled way before display, and "controlled way before display" is the heart of the matter...

In short, R420's "Temporal FSAA" is just a marketing name for a specific Spatial AA method, it isn't Temporal AA.

But is R420 otherwise capable of TAA -- does it have accumulation buffer support? It would be nice to have for older, slower flat panels (which are around for cheap). 120fps rendering downsampled into 30fps display could be very nice and feel smooth even though the display is slow. I guess TAA would remove some jaggies in the process, too? Or maybe a combo of TAA and SAA (say 3X + 4X at 1024x768), now that would be nice

 

[The above post is mine. For some reason the forum logged me out in the middle of browsing -- PEBCAK, probably...]

 

I apologize for my flippancy.

While I agree it's not the same as Temporal AA, I don't think ATi is deliberately trying to fool anyone to think that Temporal FSAA is Temporal AA. As for it being a marketing term, I think if they really wanted to sell the feature they could come up with something much more catchy than Temporal FSAA. Maybe B3D could run a contest and we could think of something.

 

No need!

Hey, that's an excellent idea! And for once easy to participate in a competition here

 

Re: ATi is ATI and ... is ...

Of course, that room is to house the power generator.

 

Your review was the most complete of any I've read.

I thank you for your thoroughness. I appreciate the depths of detail you go to. I usually just browse the articles and go on to the next site, but this review compelled me to register to say thank you. Your review details the similarities and the differences between the R3XX and R4XX in a most readable manner explaning the technology behind the benchmarks clearly and thoroughly.
I'm clinging to my AIW9700pro but the x800xt PE looks like it will cause another clash between my PC and my wife. Upgrading for me has always been a decision based on features, image quality and speed. This new card appears to have risen to the level that "justifies" an outlay of cash.
When I made the jump from the GF2GTS to the 9700pro, I saw a very tangible benefit in the change. This card appears to make the same jump in performance right when the games industry is starting to produce titles that will utilyze the performance of the new cards. Eye candy, high resolution, image quality and frame rates make for a happy gamer.