Hi, folks! Sorry for bringing this up again. I will soon get this card as a gift as an upgrade from my 7600GT. How does this card compete with the HD4670?

Here we are in April 2007.

Predictions were correct - the 1900 canes the 1800 in pretty much any modern game, and is holding out significantly better than the 7800 (and 7900) architecture route than nVidia took.

Not that nvidia are crying though, they're winning all the benchmarks with their new G80.

great review, I just can't read properly.

Oh, I guess that was supposed to be funny because Michael Moore is fat? I get it, fat people are funny.

No wait, that wasn't funny at all, ass.

I barely spend more than $600 on a whole computer. These high end graphics solutions are for the birds. No interest whatsoever. Wake me when something in the $200 segment outruns a 6800GS by a good margin.

Well from the benchmarks it looks like 2 7800GT's are the way to go.

X1900XT came out as I predicted. It fare very well at games that depend greatly on pixel shading performance, otherwise it perform virtually identical to X1800 in older texture pumping depended games.

Nvidia and ATI are taking two routes at this point. Nvidia is still focusing on texture pumping performance, while ATI focus on pixel shaders as the heaviest factor in future gaming tiles.

Anad graph proudness are better than TR lately...what is that ?

;o) Just to cover various comments already made.

1. X1900 is an update of X1800 so large gains should not have been expected (although nVidia fared better with 6800 -> 7800). Perhaps the "48 pixel leakage gave people false hopes"?

2. The gaping mid-range hole ($170-$550); it's a pity both ATi & nVidia wait so long before releasing the mid-range offerings based off a new card. I wonder how many sales ATi are missing out on because 6800GS & 7800GT are the compelling choices in their relevant price brackets.

3. I agree with those who pine for the 9700Pro days where gains over the previous generation were very big and ATi stood a good chance of levelling the playing field for us consumers. It's a pity they missed such a superb opportunity. nVidia's GeForce 6 launch was right up there though; partly because the GeForce 5 were so disappointing and partly because the 6800GT was such a sweet card and announced so quickly.

4. IMHO Radeon 9500Pro was the best card ever released as opposed to 9700Pro, although 9700Pro came first and set a new precedent the 9500Pro quickly followed and was awesomely powerful for the time and price, esp if you o/c'ed as the core often gave more than an extra 30%!

Anyone have a link to where this is for sale at?

I checked Newegg and ZZF and could not find it.

It is limited by pipelines. Tradditional pipelines consist of Pixel Shader Processors, TMUs and ROPs, and transfer data in that order. In the R520, ATI introduced a crossbar to make the Pixel Shader Processors independent and make better use of the TMUs (which are now the front end of the pipeline) by mutithreading their Pixel Shader Processors, which pretty much saturated the TMUs (or pixel pipelines as the TMUs are now the front end). In the R580, tripled the number of Pixel Shader Processors but doing so was already at the point of diminishing returns as the mutithreading had already doubled utilization of the TMUs (which I am basing off the fact that performance doubled in the R520 with only 16 pipelines), so there was not much more overhead for ATI to exploit by increasing the number of Pixel Shader Processors. The only potential performance gain ATI will get from this is if the Pixel Shaders used in games (they are software written by developers contrary to popular belief that the Pixel Shader Processors are actually Pixel Shaders) become more complex such that they take longer to execute. In those situations, Nvidia's G70's fps will go down while ATI's R580's fps will degrade much more gracefully, or not at all.

The solution to get more tangible gains in performance is to either add more pipelines, or do away with the concept of a pixel pipeline and add more TMUs utilizing a crossbar to better utilize the existing ROPs to save power and reduce the transistor budget like Nvidia does now.

Congratulations to ATI on the successful hard launch of a great new video card.
Any card that can compete with and/or beat the mighty GTX512MB and can be bought for approx.$550 makes the open class graphics processor market more competitive and gives us a roadmap to the kind of performance that the mid range cards of next year will possess.
Lets hope the ATI and Nvidia fans of this site can now put to rest the trashing of each companies products and begin to refocus the discussions bearing in mind this website's motto.."Personal computing explored" rather than the recent "Personal attacks encouraged" atmosphere.
* A fan = someone who enthusiastically supports a company or product.
* A fanboy = someone who denigrates and abuses the choices and likes of others regardless of whether they own or are likely to buy the product in question.
Perhaps I have put that a little clumsily,however I hope you appreciate the spirit in which it is written.

Peace.

Regards
Fp.

Anyone heard if ATI will try to implement Physics processing via Pixel Shaders? That could explain their abundance...

Rather than split the single/SLI/Crossfire graphs, why not just put the SLI/Crossfire bars in a different color from the single-card bars? Best of both worlds...

9700Pro was an amazing card; I used one for almost three years. A legend in its own time, indeed.

Here's to hoping I can do the same with my 7800GT.

The X1900XTX may be the fastest single card known to man, but is it really the fastest solution for the money? In pretty much all your benchmarks the SLI 7800GTs outperform the X1900XTX, and if you shop around you can get them for $275 a piece after rebate ( http://www.newegg.com/Product/Product.asp?Item=N82E16814130247 ). That means two of them add up to $550, which is $50 cheaper than the XTX, and of course in stock right now. Considering SLI boards like the A8N-SLI can be had for around $120, the dual 7800GTs seem like a nicer deal.

Even retail stores like Compusa have this card in stock already. Much better launch day than last time that's for sure.

http://www.compusa.com/products/product_info.asp?product_code=33746...

Perhaps all those pixel shaders are in there not to increase performance in graphics rendering, but to allow for physics rendering on the GPU.

http://www.xbitlabs.com/news/video/display/20051005202331.html

That's a huge performance and price hole between the 1600 and 1900 lines. What happened to the ever popular 200 dollar segment?

Mmmmmmmmm....Cheese Bread...

3 pages of real game benchmarks and 4 pages of just 3D Mark 06, and another 1 page for Shadermark.

Good job guys...

Snore... Wake me when Ati releases a competitive midrange product.

Nice card, but with 3 times the PS units, thats kinda dissapointing, granted they didn't up the numbers of the other units though. It will get better as more demanding games come out apprently, but by then nvidia will have its own answer to the 1900s.

I think they should have gone with 32 PS, and then 24 texture units, render back-ends, and Z campares with 12 to 16 vertex shaders.

That would have been more well rounded, but who knows how much extra time that would have taken. In the end, it doesn't warrant for me to go out and build a new system.

I feel for everyone who splashed out on an x1800 xt, mighty cynical of those ATI chaps to release a superior product so quickly. Altho I guess people who buy hardware for e-penis reasons deserve to be burnt a little :)

Excellent review as always, but the card is just ok in my book.

Not quite the same leap as when they went from a 8500 to 9700Pro.

Or even Nvidia's 6800U to 7800GTX.

I know we can't expect that everytime, but on can hope.

On the other hand, thanks for the power info.

My 7800GT seems to sip power at idle and load, and still plays all current games great. :)

Excellent review as always. I hope a x1700 derived from this core is in the works. That would be a pretty sweet card even with many of the shaders disabled.

Question though. Is it harder to create a defect free texture unit than it is a pixel shader? I don't see why that would be the case but I am curious why ati didnt go with say... 20-24 texture units with 40 pixel or even a 32x32 config.

I know they wanted to pack on the pixel shaders but we are far from being overrun with pixel heavy games. It seems like some texture love would have helped.

Will Nvidia bother to hit back? If/when they move over to 90nm I'm sure they could squeeze s'more outta their designs right?

I remember the day of 9800Pro to X800XT. The increase was more than 2x on every games and 3D Mark. Now the leap is very small compare to last generation card. Not worth to shell out the money! Need to wait for two generations to see the huge performance difference.

Wouldnt mind if SLI and crossfire graphs were separated and included in a separate section. Your graphs could be easier to read. Good job otherwise, I liked the shader tests to test the new fragment processors.

I wanna see how these babies OC...

So far they are not near those 13000 3D marks everyone was hyping about...