The Real Deal: 2900XT vs. 8800GTS

[Keysplayr]

I did a little review of my own. You probably won't get more accurate numbers than these.

http://endeavorquest.net:8880/2900XTvs8800GTS/index.htm

[SirBaron]

Not bad!

I still believe that the ATI card was 7 months too late, this should of come out when the Nvidia launched the Geforce8 series last year.

[infowar]

ati has always sucked from a driver standpoint

[Darth Shaidar]

lol, you're a funny guy, infowar

[Keysplayr]

Thanks guys. I still have to write up a conclusion. But in short, the 8800GTS 640 is kind of an obvious choice over the 2900XT for most reasons. Performance is close between the two until AA is maxxed at higher resolutions, then the 2900XT tanks a bit, but many other reasons remain to keep the GTS on top choice here.

Cost
Heat
Power consumption
Noise
Performance

And until CUDA makes a stand, GPGPU still belongs to ATI.

So, it's a no brainer. There was antoher guy I was doing the benches with. He also had a GTS640 and a 2900XT. He is keeping his 2900XT for the sole reason that he has the Intel P-35 Crossfire mobo, and wants to get a 2nd 2900XT. Other than that, I think he would have kept the GTS he had for the benches.

So in short, the 2900XT is a very good performer expecially with the new CAT 7.6 drivers which remedied a lot of gaming issues like it had in STALKER with the CAT 7.5's and prior. The 2900XT does fall short of the GTS640 however.

[Chuki]

Just now reading that, I must say excellent review! I would rep ya if I could. I'm really enjoying my 2900XT, but for $50 I can't complain can I?

[infowar]

direct x 10 benchmarks are out. still looks bad for ati

[SirBaron]

ATI has lost this round that's all i can say,

Had they released it last November, i would think well, you win and you lose some. However this took what 7 months to come out to still be slower?

Hopefully ATI can pick it up and released a competetive product on the next set of cards otherwise it's going to hurt them really bad.

I may prefer Nvidia cards at the moment, but no competition is bad for everyone

[Darth Shaidar]

For the record the 2900 pwns in Bioshock DX9. Outperforming the 8800 Ultra at 1680x1050/1600x1200.

[infowar]

do you have a link for that? and is it with any AA or AF?
a while ago you admited it currently sucked, but to wait for dx10 benchmarks.
everything i read about 2900 at release said it was never faster than an 8800 ultra in dx9 and was often beat by the x1950.

also, check out that great bioshock performace on the x850. remember when all of rage3d said sm3 would never matter?

[Darth Shaidar]

do you have a link for that? and is it with any AA or AF?

Yes. http://www.bit-tech.net/gaming/2007/..._performance/8
No AA, but any game that performs better on R600 w/AA than on 8800 Ultra would have to be bugged. As I said previously, outperforming 8800 Ultra @ 1680x1050 and 1600x1200.

a while ago you admited it currently sucked, but to wait for dx10 benchmarks.
everything i read about 2900 at release said it was never faster than an 8800 ultra in dx9 and was often beat by the x1950.

Times have changed.

also, check out that great bioshock performace on the x850. remember when all of rage3d said sm3 would never matter?

Wow. 3+ years after launch and finally a game worth playing that doesn't run on X8x0 hardware out of the box? There is a 3rd party effort underway to make Bioshock run on SM2 hardware, progressing quite nicely from what I hear. http://bioshocksm2.blogspot.com/

[infowar]

very interesting. thanks for the link.
have the drivers for the 2900 gotten better in dx9 all around or is this specific to bioshock?
as for sm3 support, im glad i have a 6800. i think the x850 would have been much more at that time. i predict we will see many more quick and dirty console ports passed off as PC games that require sm3. there still are people using cards from that era and they run alot of current games.

[Darth Shaidar]

very interesting. thanks for the link.
have the drivers for the 2900 gotten better in dx9 all around or is this specific to bioshock?
as for sm3 support, im glad i have a 6800. i think the x850 would have been much more at that time. i predict we will see many more quick and dirty console ports passed off as PC games that require sm3. there still are people using cards from that era and they run alot of current games.

I understand there are a lot of gamers still running SM2 hardware. It does suck having to shell out for a new video card every year or so to play the latest greatest games at high quality settings, but that's the cost of PC gaming. I think 3+ years is pretty good life for a specific family of products, especially in the fast-paced gaming market.

[Rollo]

For the record the 2900 pwns in Bioshock DX9. Outperforming the 8800 Ultra at 1680x1050/1600x1200.

It's a little misleading to state the 2900 "pwns" at Bioshock. While it does offer great performance WITHOUT AA, the performance in DX10 is very poor, and the performance with AA in DX9 is very poor.

http://www.firingsquad.com/hardware/...ance/page5.asp
At your 16X12 0X16X setting the 1GB 2900 is at 42fps average, and the GTS is 53fps average. A GTX is all the way up to 71fps average at this setting! No "pwning" there.

http://www.firingsquad.com/hardware/...a_performance/

We asked AMD if they plan on adding AA support to BioShock in a future Catalyst driver revision but couldn’t get a direct answer. The only way to enable AA under BioShock with Radeon cards is to rename the BioShock executable from “bioshock.exe” to “Oblivion.exe”. Keep in mind by doing this though, AMD’s driver-level optimizations for the game are automatically disabled, and as a result, performance suffers – we recorded a frame rate of just 26.7 fps for the 2900 XT with the executable renamed to Oblivion.exe running our manual walkthrough sequence under 4xAA at 1600x1200. (Surprisingly enough, the Radeon X1950 Pro delivered a frame rate of 20.9 fps under the exact same scenario) Because of this, AMD’s Dave Baumann told us flat out not to run the game with the renamed executable.

DEFINITELY no "pwning" there, and times haven't changed. The 2900XT is pretty much tied with the 320MB GTS as the last card you'd buy to run Bioshock at modern gaming settings. Who buys $400 DX10 cards to run them without DX10 on games it's clearly possible, or without AA when it's clearly possible?

I don't like everything to look like it's jagged and broken....

[infowar]

traditionally AA performance has been tied to memory bandwidth as i understand it. the aa on a card with 256bit memory is almost like "free aa" compared to a 128bit.
so how can a card with a 512bit bus do worse in AA than one with a 320bit bus or whatever nvidia has?
if AA performance is still horrible this long after release, id say part of the r600 chip is beyond the repair of drivers.

[Rollo]

traditionally AA performance has been tied to memory bandwidth as i understand it. the aa on a card with 256bit memory is almost like "free aa" compared to a 128bit.
so how can a card with a 512bit bus do worse in AA than one with a 320bit bus or whatever nvidia has?
if AA performance is still horrible this long after release, id say part of the r600 chip is beyond the repair of drivers.

In this game, the horrible performance is partly because to get AA to work, you have to rename the executable, which disables any driver specific optomizations.

R600 AA does put a bigger hit on framerates because it's done with shaders and the 2900XTs shader architecture is VLIW, and probably can't achieve peak efficiency due to that.

http://www.custompc.co.uk/news/11277...rformance.html

The idea I'm getting is the 2900XT will never really approach it theoretical potential unless games are written specifically for it.

[m3talc0re]

That's a pretty big mistake on their part. This isn't the console market :/

[Darth Shaidar]

FS's Bioshock numbers for the 2900 were stated to be in error by Dave Baumann @ B3D. http://forum.beyond3d.com/showpost.p...&postcount=380 Too bad Brandon never updated his numbers... Better luck next time rollo ;)

[Rollo]

FS's Bioshock numbers for the 2900 were stated to be in error by Dave Baumann @ B3D. http://forum.beyond3d.com/showpost.p...&postcount=380 Too bad Brandon never updated his numbers... Better luck next time rollo ;)

What what what?!?! ATi PR person Wavey Dave disagrees with some benches that show his company's products in a less than favorable light??! No way!
;)

In any case, FS likely doesn't use ATi PR benches for their reviews, they use their own. I'll believe Dave's when I see them replicated on an independent review site like AT,H, or FS.

[Darth Shaidar]

What what what?!?! ATi PR person Wavey Dave disagrees with some benches that show his company's products in a less than favorable light??! No way!
;)

Let's not go there, Mr. NV User Group Member ;)

In any case, FS likely doesn't use ATi PR benches for their reviews, they use their own. I'll believe Dave's when I see them replicated on an independent review site like AT,H, or FS.

Do you think Dave has any reason to lie about this? Anyway, Bioshock performance on HD 2900 XT in DX9 mode is excellent, as seen @ bit-tech: http://www.bit-tech.net/gaming/2007/..._performance/8. Looks like HD2900 loves UE3 games. R6 Vegas, Bioshock, MOH:A (http://www.legionhardware.com/document.php?id=683&p=2) all run great on 2900, as does CoJ, Prey, and a whole host of others. Clearly the 2900 isn't the bottom dweller some would like to make it out to be.

[Rollo]

Let's not go there, Mr. NV User Group Member ;)

Do you think Dave has any reason to lie about this? Anyway, Bioshock performance on HD 2900 XT in DX9 mode is excellent, as seen @ bit-tech: http://www.bit-tech.net/gaming/2007/..._performance/8. Looks like HD2900 loves UE3 games. R6 Vegas, Bioshock, MOH:A (http://www.legionhardware.com/document.php?id=683&p=2) all run great on 2900, as does CoJ, Prey, and a whole host of others. Clearly the 2900 isn't the bottom dweller some would like to make it out to be.

There's a BIG difference in user group membership and working for the firm in question. For example, I don't think the 2900 is a "bottom dweller" as you say- I think it's either the 3rd or 4th best card to have, depending on the game. I also think it competes well for the price range it's in.

As far as Bit Tech goes, you mean those AA-less for DX9? Or those DX10 benches where ATi's flagship part is 50% below NVIDIA's flagship part?

Legion Hardware has issues- there is no DX10 for MOH Airborne, yet there they are with a DX10 review. Go figure.

[Darth Shaidar]

There's a BIG difference in user group membership and working for the firm in question. For example, I don't think the 2900 is a "bottom dweller" as you say- I think it's either the 3rd or 4th best card to have, depending on the game. I also think it competes well for the price range it's in.

As far as Bit Tech goes, you mean those AA-less for DX9? Or those DX10 benches where ATi's flagship part is 50% below NVIDIA's flagship part?

Legion Hardware has issues- there is no DX10 for MOH Airborne, yet there they are with a DX10 review. Go figure.

Ah, spin at its finest. When your side loses, attack the validity of the other side's win. A win is a win is a win (as long as you're not cheating - coughGFFXcough).

[Rollo]

Ah, spin at its finest. When your side loses, attack the validity of the other side's win. A win is a win is a win (as long as you're not cheating - coughGFFXcough).

Ah, spin at its finest. When your side loses, attack the validity of the other side's win. A win is a win is a win (as long as you're not cheating - coughGFFXcough).

1. It's your position we should trust ATi's PR department more than independent review sites? We'll have to agree to disagree on that one.
2. It's your position Dave and I have the same constraints on what we post? We'll have to disagree there too- it's his JOB to post positives about ATi, if he posts negatives, he gets fired. Know what happens to me if I post something negative about NVIDIA? Like when I posted on nZone a few weeks back that I was getting some TDRs with a driver revision and brought it to their attention? They thanked me for bringing it to their attention and for posting my experience.
Just at basic level, you've got to realize getting some free hardware is less of a influence on me than getting a paycheck and feeding the family is on Dave.
3. If you think people buy DX10 cards to play in DX9, or $400 video cards to play without AA, we'll have to agree to disagree. Even the guys at Rage3d agree with me on the AA issue:
http://www.rage3d.com/board/showthread.php?t=33902626
ATi screwed the pooch with the R600 line Darth, performance is way too spotty to merit your defense.
4. A review of DX10 performance of a game that doesn't have DX10 doesn't exactly fill me with confidence, but if you think that benching games in Vista automatically makes them DX10 like Legion Hardware does, heh, more power to you. I'll run you some UT2004 DX10 benches tonight! ;)

[Rollo]

FS's Bioshock numbers for the 2900 were stated to be in error by Dave Baumann @ B3D. http://forum.beyond3d.com/showpost.p...&postcount=380 Too bad Brandon never updated his numbers... Better luck next time rollo ;)

Brandon updated his numbers, used the Cat 7.9s and Bioshock still stinks on ATi:

http://www.firingsquad.com/hardware/...date/page9.asp

BioShock is the one title that AMD really needs to work on. As we noted in our BioShock performance articles, DX10 performance is significantly slower than DX9, and the game doesn’t scale at the moment with CrossFire. The Radeon HD cards also lack support for AA in BioShock. You have to rename the game’s executable in order to force AA, and as we discussed in our BioShock AA article, this disables all of AMD’s driver optimizations for BioShock. While we didn’t run specific performance comparisons, it seems like NVIDIA’s performance in BioShock has improved tremendously with this latest ForceWare driver. We’d guess performance has improved by over 10%.

From 9/21/07 (yesterday) using the latest ATi drivers. So we can either believe the independent review site with nothing to gain or lose, or we can listen to the guy selling us ATi cards. (Dave)

Tough choice there.

[Darth Shaidar]

1. It's your position we should trust ATi's PR department more than independent review sites? We'll have to agree to disagree on that one.

Not at all, but when other sites like Bit-tech support Dave's statements I'm more inclined to believe that something is wrong with FS's test configuration, as Dave implied.

2. It's your position Dave and I have the same constraints on what we post? We'll have to disagree there too- it's his JOB to post positives about ATi, if he posts negatives, he gets fired. Know what happens to me if I post something negative about NVIDIA? Like when I posted on nZone a few weeks back that I was getting some TDRs with a driver revision and brought it to their attention? They thanked me for bringing it to their attention and for posting my experience.
Just at basic level, you've got to realize getting some free hardware is less of a influence on me than getting a paycheck and feeding the family is on Dave.

I really don't want to get into this with you, especially not on an NV fan site ;)

3. If you think people buy DX10 cards to play in DX9, or $400 video cards to play without AA, we'll have to agree to disagree. Even the guys at Rage3d agree with me on the AA issue:
http://www.rage3d.com/board/showthread.php?t=33902626
ATi screwed the pooch with the R600 line Darth, performance is way too spotty to merit your defense.

Bah! HD2900 Pro is all I have to say to that one. $250 + an easily-attainable o/c for HD2900 XT performance is delicious! Stock performance is still > than 8800 GTS 320 for less money and with a larger frame buffer.

4. A review of DX10 performance of a game that doesn't have DX10 doesn't exactly fill me with confidence, but if you think that benching games in Vista automatically makes them DX10 like Legion Hardware does, heh, more power to you. I'll run you some UT2004 DX10 benches tonight! ;)

I never said that, Legion did. A simple mis-statement on their part doesn't invalidate their results, as much as you'd love it to ;)

[Darth Shaidar]

Brandon updated his numbers, used the Cat 7.9s and Bioshock still stinks on ATi:

http://www.firingsquad.com/hardware/...date/page9.asp

From 9/21/07 (yesterday) using the latest ATi drivers. So we can either believe the independent review site with nothing to gain or lose, or we can listen to the guy selling us ATi cards. (Dave)

Tough choice there.

See previous post. There's more than one review out there showing Bioshock running fantastically on HD2900, as is expected given the fact that it's a UE3 game and *every* other UE3 game on the market performs well on HD2900. As for not wanting to play "DX9 games" on DX10 hardware, what a crock! There isn't a true DX10 game out there to play on said DX10 hardware in the first place. The "DX10 games" we have now show very little image quality difference between DX9 and DX10 render paths. Once a real from-the-ground-up DX10 title comes out it would be worthwhile to revisit this topic. Until then it's just useless posturing.

[SirBaron]

By the time that happens i'll be on a Geforce10900

[Uber_Tiny]

By the time that happens i'll be on a Geforce10900

[Rollo]

See previous post. There's more than one review out there showing Bioshock running fantastically on HD2900, as is expected given the fact that it's a UE3 game and *every* other UE3 game on the market performs well on HD2900. As for not wanting to play "DX9 games" on DX10 hardware, what a crock! There isn't a true DX10 game out there to play on said DX10 hardware in the first place. The "DX10 games" we have now show very little image quality difference between DX9 and DX10 render paths. Once a real from-the-ground-up DX10 title comes out it would be worthwhile to revisit this topic. Until then it's just useless posturing.

I don't think anyone doubts 2900XTs perform great without AA Darth. It's just that no one really cares, because AA is pretty much considered a requirement for cards in it's pricerange.

This is like if Chevy produced a sports car without power steering. Hey great, it goes fast, but it lacks a feature everyone takes for granted that enhances the experience using the product.

I've already admitted it could well win AA-less DX9. I just consider DX9+AA and DX10 the important victories these days.

BTW- what happened to you caring about image quality Darth? Now you don't seem to care if ATi's products are offering the lowest image quality, your logic seems to be "ATi makes it, it must be better".

It's pretty much a given that using AA is better IQ than not.

In the games with DX10, there are some differences, usually not big, but still some differences. (e.g. more realistic fur in Lost Planet on their coats)

ATi AF is angle dependent, while NVIDIA's is angle invariant.

NVIDIA offers some better AA modes, ATi gives us the laughable blur inducing tent modes.

I'd think you would be agreeing with me, not pointing at AA-less benchmarks.