GeForce 9-series lineup revealed

[Quiksilver]

Originally posted by: jaredpace

Originally posted by: Quiksilver
Just noticed NH updated their spec's for the GTX (675/2200) rather than originally 750+/unknown, which now makes me disappointed unless they make a ton of performance tweaks to the card to make it worth it otherwise it might make more sense just to buy one of these: http://www.gainward.net/produc...il.php?products_id=155 sames its so very close to the 9800GTX spec's.

I would love to see a 9800gtx vs. that card.

I agree, I found a few sites that have it for ~$350 Euros which is quite expensive... about $500 USD.

 

[Demoth]

From an enthusiast point of view, this last several months has been boring. However, I like the trend of cheap near top end performance with lower power reqs. I also like seeing these companies focusing more on crossfire/sli.

Let's see some real performance out of the 9500GS levels and make CF/SLI a viable option for expandability for the budget gamer.

Right now, it is possible to mix and match ATI in a generation, so even a 3850, which could be selling for under $90 within the year, will give a noticeable improvement to the 3870X2. The future of this industry is to be able to swap in any new card in a multi-GPU solution while retaining the power from the previous purchased card.

Rumors on the late Q2 ATI HD 4000 series look promising. Fewer designs and more emphasis on crossfire means better drivers and faster driver fixes.

http://www.nordichardware.com/news,7356.html

All I have to say is don't worry too much about Nvidia. If they want to be lazy and rake in a few extra bucks by holding back R&D and design, they may just have to be spanked around a bit by the RV770 to be given a wake up call. And, wake up they will.

 

[Chaotic42]

Originally posted by: Demoth
From an enthusiast point of view, this last several months has been boring.

No kidding. I have a 7800GTX that I got in 2005 and I'm ready to replace it, but there's nothing exciting out there. I was hoping that the 9800 would be the real deal, but I'm skeptical now.

I'm buying the first single card that performs 100% better than an 8800Ultra for $500 or less. I'm thinking it will be 2009.

 

[Rusin]

If we would forget Crysis.. how big need for new much faster GPU's there would be? I mean 8800 GTX can still deal with almost all games at 1920x1200 4xAA 16xAF..pretty good for 1.5 year old card?

 

[Chaotic42]

Originally posted by: Rusin
If we would forget Crysis.. how big need for new much faster GPU's there would be? I mean 8800 GTX can still deal with almost all games at 1920x1200 4xAA 16xAF..pretty good for 1.5 year old card?

World in Conflict is pretty rough on a system. I'm also curious about how well Fallout 3 will perform.

 

[Demoth]

Originally posted by: Chaotic42

Originally posted by: Demoth
From an enthusiast point of view, this last several months has been boring.

No kidding. I have a 7800GTX that I got in 2005 and I'm ready to replace it, but there's nothing exciting out there. I was hoping that the 9800 would be the real deal, but I'm skeptical now.

I'm buying the first single card that performs 100% better than an 8800Ultra for $500 or less. I'm thinking it will be 2009.

Actually, the RV770 should be >50% faster and CrossfireX looks to be scaling near 100% for a single card, 60% for a third. You may be meeting all your reqs. by summer with the 4870X2.

 

[BFG10K]

So would that be yes or no for every parts of the games?

It would be "likely enough so that texturing in your context doesn't make a difference".

We already know the GTS 512 beats the GTX in many modern games when not using AA, many of which use FP HDR. So we know your texturing and memory bandwidth arguments are invalid in those situations.

We already knew AA can make a difference to bandwidth but again that has nothing to do with your texturing claims.

It's meaningless to games? It is tool to measure what the card is capable of. I wouldn't call it meaningless.

It's meaningless because games aren't bottlenecked in the manner the test operates in.

If I create a CPU test that reads and writes to a single register how relevant do you think it is to real-world programs?

What I meant is that graph you linked few months back showed that shader was used even in older games.

That's my point, a trend that is continuing upwards.

Exactly that graph showed that old games used shaders. Very little of it but still.

Very little? Explain "very little" when as of early 2006 some games had a shader/texture ratio of 7:1.

If anything the texturing was "very little" even back in 2006.

You reply to my message I'm replying back. If not don't reply to my posts. fair enough?

Replying isn?t the problem, it's your continual choice to ignore that graphs that were provided.

Do you see the advantage Radeon has over FX series?

Yep, with shaders, as was evident with 128 bit 96xx derivatives still competing with 256 bit 59xx derivatives in some situations.

Why are you talking about FX vs 9800 series again? This has no relevance to what we are talking about with current gaming situations and gpu.

Much like your multi-texturing graphs not being relevant to current games or GPUs.

Much like your "raw texture fillrate will kill everything" comments not being relevant to today?s games or GPUs.

I think you?re stuck back in 1999/2000 when the GTS didn't have enough bandwidth to take advantage of it?s 4x2 configuration until the Ultra came along. Somehow you think that?s relevant for today when it isn?t.

I never said all you need to do is increase the texture fillrate.

Yes you did, repeatedly.

I believe long as a card is not being hold back by shader or bandwidth fillrate is king which includes pixel and texture fillrate.

So in other words you?re saying something is the most important as long as other things more important than it aren?t lacking.

:roll:

Shader performance is lacking and that is the point.

Also the way you interchange pixel fillrate and texel fillrate as if it?s somehow the same thing is fallacious.

There's no such thing as magic drivers and uber tweaks.

Have you forgotten numerous examples where ATi, nVidia and even 3dfx provided large performance gains with drivers?

Then explain the 9600gt phenomenon.

Again it could be anything from the core to drivers. In any case the 9600 GT has much lower texel fillrate than a 8800 GT anyway.

You say shader creates "rich" pixel but current games aren't shader bound with modern GPU.

Of course they are, and that's one of the reasons why ATi don't do so well. At best they have 320 shaders but at worst they have much less given code is seldom 100% optimal for their configuration. That and nVidia's shader clocks are much higher than ATi?s.

According to BFG however shader dictates game performance and create "rich" pixels. Memory bandwidth and fillrate doesn't matter.

Nope, by all means, raise the specs everywhere if possible. But if you can?t then focus on shaders first as that?s where you?ll see the biggest gains, even with 256 bit memory.

 

[AzN]

You make me laugh. Your SP theory has been debunked in the other thread. All you do is argue. Good luck...

 

[apoppin]

Originally posted by: Azn
You make me laugh. Your SP theory has been debunked in the other thread. All you do is argue. Good luck...

You *all* make me "laugh"

no theory has been debunked or proved
-all you guys are doing is pointless arguing over speculations about an unreleased product

 

[AzN]

We are not even arguing about future products.

His theory about shader "rich" pixels that dictate performance with current gpu has been debunked sorry to say. The first time was when he thought G92GTS was going to beat the GTX and the ultra because it had faster shader clocks.

Second time was when Sickbeat and Keysplyr2003 tested with 64 shader with their 8800gt and GTS and had very little impact in the other thread.

He just argues and nitpicks at my sentence structure when theres clear evidence.

 

[TheSlamma]

looks like 2008 is going to be year of the milk for video cards.

 

[apoppin]

Originally posted by: Azn
We are not even arguing about future products.

His theory about shader "rich" pixels that dictate performance with current gpu has been debunked sorry to say. The first time was when he thought G92GTS was going to beat the GTX and the ultra because it had faster shader clocks.

Second time was when Sickbeat and Keysplyr2003 tested with 64 shader with their 8800gt and GTS and had very little impact in the other thread.

He just argues and nitpicks at my sentence structure when theres clear evidence.

You are still projecting performance for a future product ... we know about the 9600 [it does have faster clocks but are still unsure of its impact] ... we don't know about 9800 yet

wait for the whole picture

 

[BFG10K]

You make me laugh. Your SP theory has been debunked in the other thread. All you do is argue. Good luck...

I posted figures in that thread the demonstrated I was right.

The first time was when he thought G92GTS was going to beat the GTX and the ultra because it had faster shader clocks.

And it did.

Second time was when Sickbeat and Keysplyr2003 tested with 64 shader with their 8800gt and GTS and had very little impact in the other thread.

Like I said in the other thread their tests were largely flawed.

Keys used a low resolution in a relatively primitive game engine while Sickbeast told us "it doesn't feel any different to me".

So you can believe them or you can look at the real figures I provided from a range of games that prove I'm correct.

 

[Tempered81]

@bfg what are you proving you're correct about?

 

[AzN]

nothing. He likes to argue about what you didn't say.

 

[BFG10K]

@bfg what are you proving you're correct about?

In the other thread a few of them ran some tests in CoD 4 and concluded there is no difference between 64 SPs and 128 SPs and then Azn was running around saying ?I told you so?.

I then produced benchmarks to the contrary demonstrating a massive performance gain with 128 SPs over 64 SPs with my 8800 Ultra, in a range of games.

 

[AzN]

He tests in some uber high resolutions with mass amounts of AA where it need more shader effects. Picks on some particular games where it would be shader bound.

Sickbeat and Keys tested with 64 shader in a modern game and saw no noticeable differences.

BFG is upset again. I never told him I told you so other than his theory has been proven wrong the first time he said G92GTS and second with Keys and Sickbeat.

Again G92GTS does not beat the GTX. It sometimes beat GTX in lower resolutions where it isn't constraint by memory bandwidth and pixel fillrate so that G92GTS can flex its texturing muscles. Cookie Monster described it point on. G92 is unbalanced where it needs more bandwidth.

He is the one who goes around telling me I told you so and replying to my posts to tell me I was proven wrong in the first place. The nerve of this guy is simply amazing.

Originally posted by: BFG10K

Shader is the future and currently games aren't shader bound where SP dictates massive performance difference than fillrate and bandwidth.

We've been over this multiple times before and you were proven wrong. Remember the graphs I posted that demonstrated how prolific shaders are in modern games?

What it needs is more ROP and memory bandwidth.

ROPs and memory bandwidth aren't a factor when shaders are running calculations to generate ?rich? pixels.

 

[Tempered81]

does make sense though... in a way both u guys are right. the 9 series needs mem bandwidth 384 or 512, and memory 768/1gb/2gb. theres no doubt about it. Also its correct in saying that the low SP G92/94 9600's and midrange cards will suffer in high res's like BFG's test proved. Yet midrange 9600's are going to excell in easy tests like medium res tests in cod4 done by keys and sickbeast. If you want performance and AA on 1024 or 1280 go with a midrange 9600 type, if you want high res uber performance, (it's sad to say this) but get the 384bit/768mb cards. (we all know which cards these are)

edit: if the 9800gtx actually ships with 256bit/512mb we may not see the same high res performance on this card that a 384bit/768mb 8800gtx/ultra gives.

 

[apoppin]

Originally posted by: jaredpace
does make sense though... in a way both u guys are right. the 9 series needs mem bandwidth 384 or 512, and memory 768/1gb/2gb. theres no doubt about it. Also its correct in saying that the low SP G92/94 9600's and midrange cards will suffer in high res's like BFG's test proved. Yet midrange 9600's are going to excell in easy tests like medium res tests in cod4 done by keys and sickbeast. If you want performance and AA on 1024 or 1280 go with a midrange 9600 type, if you want high res uber performance, (it's sad to say this) but get the 384bit/768mb cards. (we all know which cards these are)

edit: if the 9800gtx actually ships with 256bit/512mb we may not see the same high res performance on this card that a 384bit/768mb 8800gtx/ultra gives.

now we're getting somewhere .. they both are right ... as far as they can see of a "piece" of the entire picture

i'd like to point out that i think we are not looking at nvidia's top card
-it is a lesser card

 

[bryanW1995]

Originally posted by: jaredpace
does make sense though... in a way both u guys are right. the 9 series needs mem bandwidth 384 or 512, and memory 768/1gb/2gb. theres no doubt about it. Also its correct in saying that the low SP G92/94 9600's and midrange cards will suffer in high res's like BFG's test proved. Yet midrange 9600's are going to excell in easy tests like medium res tests in cod4 done by keys and sickbeast. If you want performance and AA on 1024 or 1280 go with a midrange 9600 type, if you want high res uber performance, (it's sad to say this) but get the 384bit/768mb cards. (we all know which cards these are)

edit: if the 9800gtx actually ships with 256bit/512mb we may not see the same high res performance on this card that a 384bit/768mb 8800gtx/ultra gives.

I think that nvidia will probably be able to get enough performance with better memory management and higher clockspeeds to beat the 8800, but it's still pretty ballsy of them to call this "9800gtx" instead of "8801gtx"

 

[Tempered81]

8800gts512 / 9800gtx512 / 8800gtx768 gpu-z side-by-side comparison:

http://www.xtremesystems.org/f...howthread.php?t=178457

post #25


Looks like this is the real thing.... However, I do hope its all a carefully plotted ploy to distract from the real BOMB monster gpu coming out march 11th!

 

[bryanW1995]

unless they pull some sort of magical rabbit/driver out of their hat with g92-based optimizations that don't help g80, this 9800gtx is going to suck.

 

[AzN]

Originally posted by: apoppin

Originally posted by: jaredpace
does make sense though... in a way both u guys are right. the 9 series needs mem bandwidth 384 or 512, and memory 768/1gb/2gb. theres no doubt about it. Also its correct in saying that the low SP G92/94 9600's and midrange cards will suffer in high res's like BFG's test proved. Yet midrange 9600's are going to excell in easy tests like medium res tests in cod4 done by keys and sickbeast. If you want performance and AA on 1024 or 1280 go with a midrange 9600 type, if you want high res uber performance, (it's sad to say this) but get the 384bit/768mb cards. (we all know which cards these are)

edit: if the 9800gtx actually ships with 256bit/512mb we may not see the same high res performance on this card that a 384bit/768mb 8800gtx/ultra gives.

now we're getting somewhere .. they both are right ... as far as they can see of a "piece" of the entire picture

i'd like to point out that i think we are not looking at nvidia's top card
-it is a lesser card

It seems that when BFG tested his ultra with 64 SP disabled he didn't realize that 1/2 texture fillrate would be cut into half which would hamper his performance not because of 1/2 of SP is disabled.

http://forums.anandtech.com/me...d=2158058&STARTPAGE=11

 

[Dadofamunky]

Originally posted by: bryanW1995
unless they pull some sort of magical rabbit/driver out of their hat with g92-based optimizations that don't help g80, this 9800gtx is going to suck.

I know, I think I'll wait until some poor saps start selling off their 8800GTX or Ultras for $250 on Craig's List

 

[BFG10K]

It seems that when BFG tested his ultra with 64 SP disabled he didn't realize that 1/2 texture fillrate would be cut into half which would hamper his performance not because of 1/2 of SP is disabled.

But again you need to demonstrate that the reduction of texturing hurts performance more than the reduction of shading ability.

You would also need to demonstrate how your multi-texturing results are relevant in today?s games.

So far you have failed to do either while I have provided ample evidence to demonstrate shaders becoming more and more widespread compared to texturing.

We also have commentary from ATi in the article I linked to that confirms what I'm saying.