So is R580 supposed to be 48 pipes or 48 ALUs?

Well, I think this is a temporary situation. The process technologies are still slowing down their advancement.

 

And as a result, it's still listed at $749 or $100 over MSRP. It'll be interesting to see how prices of the X1900 cards are a month or two after release.

 

Missed the 7 posts talking about these benchmarks in this thread?

Anyway, regarding those benchmarks: Did anyone notice there was no mention of AF settings? I wonder how they tested.

For comparison with other video cards, Anandtech's benchmarks have the 7800GTX scoring only 1 fps slower in FEAR (with and without AA), so they look comparable. 7800 512MB gets 62fps without AA, 31fps with.

As for G71, I'm really curious if it is indeed 32 pipes or not. Some posts by Hubert seem to suggest otherwise, but I have no idea where he got his "info" from. The currently accepted rumours would make it a 24GTexel/s monster, and I don't think R580 can handle that.

 

Well, the 7800 GTX 512 was nVidia's first attempt at releasing a $650 MSRP product, and they vastly underestimated the demand. I doubt that ATI is going to do the same now.

Anyway, from what little I've seen, I think it's amazing just how well the GTX 512 does against the X1900, given that the X1900 is on 90nm with nearly 400 million transistors, whereas the GTX is closer to 300 million transistors on 11nm. One would naiively expect the X1900 to trounce the GTX 512, with nVidia having no hope of rebuttal until they produce their own 90nm GTX part. And yet, the GTX 512 seems to be doing okay, with what little benchmark information is currently available.

 

In that mini review he states:

Does this mean Nvidia's card were at Q and ATI's were at HQ during testing?

 

Perhaps it is CPU limited with a 1.8GHz processor.

 

Well this is whats interesting so far with the leaked 3dmark05, the pixel shader tests the r580 gets around 450 fps, where the gf 7800gtx 512 gets 400 fps. The r580 should put the hurt lock on with AA and AF, but without them its not going to be much faster IMO.

 

Nah, check out Anandtech. CPU limit is over 120fps on a FX-55.

 

But who's going to play games on $600 cards without AA and AF? Testing without these isn't anything like how people are going to use these cards.

 

That's the same old fallback line whenever ATi's hyped shader performance doesn't materialize. Make up your minds already!!

 

No, it's the same old fallback when someone claims that testing $600 cards without AA/AF is a vaild representation of how people play games.

I know Nvidia likes to cheer their raw frame performance, but really in this day and age, people are choosing to exchange those excess frames for IQ, and its valid to test for that. Heck, I've hardly played anything without AA since the days of Voodoo 5500. Get into the twenty-first century already!!

 

I understand your point and I agree. It's just that I see the IQ card played most recently whenever ATi's hyped shader performance falls flat. It happened with R520, and will happen with R580 if it doesn't deliver. Hopefully that won't happen and it will give G71 something to beat.

 

This is incorrect isn't it? Isn't the memory bus 512Bit? Oh wait .. the core is 512 the memory is 256. Nevermind.

US

 

I agree.
With 48 alus shouldn't it make a huge showing in fear without AA/AF and have the numbers become more sane with AA/AF?

 

Not saying anyone will be playing without it, just giving a possible result, its overall performance of its 48 ALU's will be right around the gf's 48 ALU's

 

My point exactly. Hopefully it won't come to that.

 

Conversly we've seen websites and individuals rave about G70's performance, while ignoring things like texture shimmering problems. I'd like to see a lot more emphasis on how good a card looks, not just how fast. We've already seen some 3DMark06 numbers with overclocked cards and cards not running at the same levels of IQ. It's not on anymore, and quite honestly I'm getting completely jaded by these misleading benchmark numbers, reviews, and testing apps like 3DMark06 that don't tell you the full story whilst pretending that they do.

 

Disappointed that it's only getting a few more FPS over the GTX 256. If it continues like this the GTX512 OC'd will thump it.

/me grumbles

FFS ATI .. if you gonna release a card to beat the green monster machine(GTX 512) then do it properely and not by 1-2 fps.

You(ATI) better show the R580 thumping the thing at high AA+AF(and i'm not talking HDR).

US

 

Looking at these results I fear(pun intended) that the GTX512 will already put hurting on the X1900XT.

http://www.anandtech.com/video/showdoc.aspx?i=2575&p=4

Shows 53fps with no AA+AF where as the dailytech shows only 45fps.

Actually .. the DailyTech Results don't make sense.

Code:

Game                 GTX256           X1900XT
BF2 1600x1200 4xAA   53 FPS           57 FPS

http://www.dailytech.com/article.aspx?newsid=310

while this shows the GTX256 and X1800XT and GTX512 gets higher scores at the same Res.
http://www.techreport.com/reviews/2005q4/geforce-7800gtx-512/bf2.gif

US

 

Yes, but you're avoiding the point I'm trying to make. Better pray that those dailytech scores for X1900XT are wrong because if they are, the memory controller will be the only saving grace for R580. That article shows 61fps on the X1900XT in FEAR at 16x12 - the GTX512 already averages higher than that.

I wouldnt want to be the one having to explain why a 48-shader part at 625Mhz is trailing a 24-shader part at 550Mhz in the most shader bound game on the market. There aren't any "shimmering" excuses out there to explain that! Having said that, those dailytech numbers just have to be complete bollocks.