Why does AT say the X1900's take the performance crown?

[deadseasquirrel]

Originally posted by: Ackmed

Originally posted by: deadseasquirrel
One could have bought 2 6800GTs and put them in SLI in January of 2005, for $650 total. 6 months later, the 7800GTX was released and merely, more-or-less, equaled performance of the SLI'd 6800GTs (better at 2048x1536, but less at 16x12). i.e. there wasn't much of a reason for the SLI 6800GT users to pay for an upgrade to get equal performance.

A 6800GT was much more than $325 last January. That negates your whole point.

Your response made me go look up my receipts. And you're right. They weren't $325 a piece. They were $375 a piece. Chumbo had a deal where they had 2 6800GTs for $750. They did make you buy a 550w PSU from them and that was $100 in order to get the deal. I needed one for the new build anyway (the old 350w Antec just wasn't gonna cut it). Obviously it wasn't a great deal to those who didn't need a PSU, but that was the best deal going at the time, and I jumped all over it. I'm sure the Hot Deal is still listed in the archives somewhere cuz that's where I saw it 12 months ago.

Regardless, whatever one had to pay for their SLI 6800GTs, there was no SINGLE card out for 12 months that significantly improved upon the performance at popular settings such as 1600x1200 4xAA 8xAF until the 1900XT this month (unless you count the GTX512 ghost edition).

But, Ackmed, you didn't even address flexy's post which stated that SLI is dumb... as that is what my entire post was countering. Does the fact that I was $50 off per GT really negate my whole post about the PERFORMANCE of SLI compared to single-card configurations over the last 12 months? As an SLI user yourself, do you feel it is "dumb" too?

edit for your edit:

Originally posted by: Ackmed
Not to mention the only SLI board then, was the Asus A8N-SLI, and it was about $230.

I bought the DFI SLI-D on Feb 4th for $189 at Monarch.

If your point is that SLI is more expensive to get into because of a beefier PSU and an SLI capable motherboard, I'm not denying that. I'm talking specifically about PERFORMANCE and how long it took for a SINGLE card to exceed that performance by a significant margin.

 

[Ackmed]

Originally posted by: deadseasquirrel
Your response made me go look up my receipts. And you're right. They weren't $325 a piece. They were $375 a piece. Chumbo had a deal where they had 2 6800GTs for $750. They did make you buy a 550w PSU from them and that was $100 in order to get the deal. I needed one for the new build anyway (the old 350w Antec just wasn't gonna cut it). Obviously it wasn't a great deal to those who didn't need a PSU, but that was the best deal going at the time, and I jumped all over it. I'm sure the Hot Deal is still listed in the archives somewhere cuz that's where I saw it 12 months ago.

Regardless, whatever one had to pay for their SLI 6800GTs, there was no SINGLE card out for 12 months that significantly improved upon the performance at popular settings such as 1600x1200 4xAA 8xAF until the 1900XT this month (unless you count the GTX512 ghost edition).

You got in on that very brief deal. Chumbo had the lowest prices, by far. Not for very long though, they were sold out. Everyone else had 6800GT's for much more, upwards of $500.

I know there wasnt any single card out there that could compete performance wise at the time. Like you mentioned however, 6 months later (little less), the GTX hit the streets. Which as you said, was pretty even with 2xGT's, and "just" $600. This review actually shows the GTX (430/1200 speeds too) faster in 6 out of 8 tests at 1600x1200 4xAA/8xAF. I was just pointing out that your $650 figure wasnt correct. You got a great deal at the time, most of the prices were more, and combine the extra cost of a SLI board, it was much more than $650. Closer to $800, most of the time higher.

Originally posted by: deadseasquirrel
But, Ackmed, you didn't even address flexy's post which stated that SLI is dumb... as that is what my entire post was countering. Does the fact that I was $50 off per GT really negate my whole post about the PERFORMANCE of SLI compared to single-card configurations over the last 12 months? As an SLI user yourself, do you feel it is "dumb" too?

No I dont think SLI overall or Crossfire for that matter, is dumb. Ive said many times before, I have issues with the "buy one now, and one later" mantra that a lot of people think. Many people at the time were saying, they would buy one GT then, which was commonly going for close to $500 for the first few months, then get another in a year or so. Who here would buy a second GT if they bought one a year ago? Not that many I would imagine. Selling it, and buying a GTX for $410, or a X1900XT for $500 would easily be the better thing to me to do. But thats just my opinion.

When I think SLI makes the most sense, is buying to cards at the same time. Or very close to it.

Originally posted by: deadseasquirrel
edit for your edit:

Originally posted by: Ackmed
Not to mention the only SLI board then, was the Asus A8N-SLI, and it was about $230.

I bought the DFI SLI-D on Feb 4th for $189 at Monarch.

If your point is that SLI is more expensive to get into because of a beefier PSU and an SLI capable motherboard, I'm not denying that. I'm talking specifically about PERFORMANCE and how long it took for a SINGLE card to exceed that performance by a significant margin.

I got my A8N-SLI at the end of Jan, mine was $250. They were going for up to $300, but I got a "deal". It was the first SLI board out, and costed the most for a while.

I know what you are saying though. Who knows whats coming out down the road though, and the chance is has a more advanced GPU? The X1900XTX beats 2xGTX's in some tests, sometimes by twice the frames. GTX's came out just about 7 months ago. Buying two when they first came out for $600, only to be bested (sometimes) by a XTX that also costs $600 the day it came out, and is more advanced. Its a tricky trade trying to guess whats coming down the pike. If you can afford it, more power to you. Im pretty happy with my SLI'd GTX's. But I am also a little disheartened too. I got my second GTX just two months ago, and my favorite game (BF2) is faster on a single XTX than my GTX's in SLI. Would I have gotten the second GTX if I knew now, what I knew then? No, but you cant look behind you with regrets all the time eithe. Im pretty happy with them, and they are faster than the XTX in almost every other game. Who knows, the G71 may be close to 2x512MB GTX's speed wise. That would certainly make me feel a little sad if I just paid $1500+ for two of them.

I dont know why I rambled on like that. In short, there are pro's and con's to SLI/XFire. I dont consider either dumb, and am grateful we at least have the option.

 

[deadseasquirrel]

Originally posted by: Ackmed
I dont know why I rambled on like that. In short, there are pro's and con's to SLI/XFire. I dont consider either dumb, and am grateful we at least have the option.

I don't mind the ramble... been accused of that myself more than once. I think my keyboard automatically types 4 words for every one word I actually type.

You make good arguments about newer cards with advanced features, etc. Which is why it's good discussing things with you, as opposed to flexy's "SLI is dumb" argument. I understand that he is just expressing his opinion, but to make a blanket statement, he either a) ignores the cases where it gives the best bang for buck (like the one I gave) or b) feels that his opinion for his buying and playing habits are the standard that everyone else bases theirs around. Both of which are wrong.

 

[Pete]

If anyone is interested in AA sample patterns (Matthias and others), Tech Report answers your question with roughly a thousand pictures (so they're worth roughly a hojillion words): SLI & XFire AA comparison chart. Red dot means geometry sample, green dot means texture sample. More than one green dot means super sampling (2 dots = 2x SSAA, 4 dots = 4xSSAA, etc.). Super sampling implies a bigger performance hit (because you're sampling X times the textures in addition to X times the geometry) but potentially nicer IQ than multisampling (that extra texture sampling touches the whole screen, not just geometry/polygon edges).

Sketchy details follow, for those interested. When you talk about AA smoothing lines, you're looking for a higher edge equivalent resolution (EER). Basically, you want as many non-overlapping samples as possible, or one red dot per X/Y row/column. ATI offers more sample positions, so they can go to 6x MSAA. Apparently NV's sample grid can only support 4x MSAA, so they need to use SSAA above that. Adding SSAA on top of MSAA can help edges, but you don't necessarily double the EER, as you can see when going from NV's "4x" MSAA to their "8xS" MS+SS AA. You double the samples in the Y axis, but X axis samples remain the same due to overlap. NV's "8x" SLI AA mode should ofer improved EER b/c there's less sample overlap. Similarly, ATI's "14x" SuperAA mode looks to have a sparser and less regular sample pattern than NV's "16x" SLI AA mode, which may mean better EER (if you can even tell the difference at that level of EER); OTOH, NV's 16x uses 4xSSAA, so their texture quality should be nicer across the whole screen (potentially less shimmering on the ground and less popping on alpha textures).

 

[Munky]

Originally posted by: Pete
If anyone is interested in AA sample patterns (Matthias and others), Tech Report answers your question with roughly a thousand pictures (so they're worth roughly a hojillion words): SLI & XFire AA comparison chart. Red dot means geometry sample, green dot means texture sample. More than one green dot means super sampling (2 dots = 2x SSAA, 4 dots = 4xSSAA, etc.). Super sampling implies a bigger performance hit (because you're sampling X times the textures in addition to X times the geometry) but potentially nicer IQ than multisampling (that extra texture sampling touches the whole screen, not just geometry/polygon edges).

Sketchy details follow, for those interested. When you talk about AA smoothing lines, you're looking for a higher edge equivalent resolution (EER). Basically, you want as many non-overlapping samples as possible, or one red dot per X/Y row/column. ATI offers more sample positions, so they can go to 6x MSAA. Apparently NV's sample grid can only support 4x MSAA, so they need to use SSAA above that. Adding SSAA on top of MSAA can help edges, but you don't necessarily double the EER, as you can see when going from NV's "4x" MSAA to their "8xS" MS+SS AA. You double the samples in the Y axis, but X axis samples remain the same due to overlap. NV's "8x" SLI AA mode should ofer improved EER b/c there's less sample overlap. Similarly, ATI's "14x" SuperAA mode looks to have a sparser and less regular sample pattern than NV's "16x" SLI AA mode, which may mean better EER (if you can even tell the difference at that level of EER); OTOH, NV's 16x uses 4xSSAA, so their texture quality should be nicer across the whole screen (potentially less shimmering on the ground and less popping on alpha textures).

Thanks for that post. I was actually gonna comment about the whole SLIAA vs SuperAA performance difference, since a number of people think the reason SLIAA is so much slower is because it uses SSAA, and that SuperAA it only MSAA, which looks worse but runs faster.

I posted earlier in another thread that it does appear SuperAA 8x is only using MSAA, while SLI does 2xSSAA in the 8x mode. Furthermore, it also looks like SuperAA 14x is only doing 2x SSAA, while SLI in 16x mode is doing 4xSSAA, which would again make it slower but looking better than SuperAA.

After closer inspection of the TR article, and the crossfire reviews at b3d, it actually turns out that SuperAA is in fact always doing 2xSSAA in addition to varying levels of MSAA. The confusion comes from the fact that in 8x and 12x modes SuperAA is taking 2 texture samples at the exact same position, which would make it look like pure MSAA, but the performance advantage can not be the result of pure MSAA. This can further be confirmed by looking at the performance chart on the same TR page you linked, which shows that 8xAA performs identically to 10xAA, and 12xAA runs at the same speed as 14xAA. So, this means 2 things:

1. The numbering scheme in AA levels is not interchangeable between Nvidia and Ati, and to be accurate, Nv's 8x mode should really be compared to Ati's 10x mode; and

2. The performance advantage of SuperAA compared to Nvidia's 8x mode cannot be blamed on faster pure MSAA with lower image quality, because Ati does use 2xSSAA in all modes of SuperAA.

 

[SiTeS]

Alright guys,

I know that you guys have argued this back and forth but I want to make sure I make the right decision in buying a video card. I am usually a Nvidia guy and have always had good luck with Nvidia but people are telling me that this card is better. I was planning on buying a ASUS EN7800GTX/2DHTV/512M Geforce 7800GTX 512MB 256-bit GDDR3 PCI Express x16 Video Card Heres the link but now I am reconsidering getting the ASUS Extreme AX1900XTX/2DHTV Radeon X1900XTX 512MB 256-bit GDDR3 VIVO PCI Express x16 CrossFire Heres the link
Does it matter that the Memory is 150mhz slower than the 7800GTX 512mb. I plan on building this within the next week. I already have the motherboard and the processor, DVD burner and hard drives on the way.
I would really appreciate if you guys could give me an idea of what to go with.
Thanks a lot for any help you can offer.
I am putting it with:
ASUS A8N32-SLI Deluxe
AMD64 4400+ Toledo X2
2x1gig sticks of OCZ Titanium
X-FI Sound
LIAN LI/KOOLANCE (watercooled) case
2- 74gig 10,000RPM Western Digital Raptor
DVD Burner
Hiper Type-R 580W Modular Power Supply (SLI certified)I have been told that this is a good Power Supply ( can ya guys give me any insight on whether this is good or not.

 

[RobertR1]

Originally posted by: munky

Originally posted by: Pete
If anyone is interested in AA sample patterns (Matthias and others), Tech Report answers your question with roughly a thousand pictures (so they're worth roughly a hojillion words): SLI & XFire AA comparison chart. Red dot means geometry sample, green dot means texture sample. More than one green dot means super sampling (2 dots = 2x SSAA, 4 dots = 4xSSAA, etc.). Super sampling implies a bigger performance hit (because you're sampling X times the textures in addition to X times the geometry) but potentially nicer IQ than multisampling (that extra texture sampling touches the whole screen, not just geometry/polygon edges).

Sketchy details follow, for those interested. When you talk about AA smoothing lines, you're looking for a higher edge equivalent resolution (EER). Basically, you want as many non-overlapping samples as possible, or one red dot per X/Y row/column. ATI offers more sample positions, so they can go to 6x MSAA. Apparently NV's sample grid can only support 4x MSAA, so they need to use SSAA above that. Adding SSAA on top of MSAA can help edges, but you don't necessarily double the EER, as you can see when going from NV's "4x" MSAA to their "8xS" MS+SS AA. You double the samples in the Y axis, but X axis samples remain the same due to overlap. NV's "8x" SLI AA mode should ofer improved EER b/c there's less sample overlap. Similarly, ATI's "14x" SuperAA mode looks to have a sparser and less regular sample pattern than NV's "16x" SLI AA mode, which may mean better EER (if you can even tell the difference at that level of EER); OTOH, NV's 16x uses 4xSSAA, so their texture quality should be nicer across the whole screen (potentially less shimmering on the ground and less popping on alpha textures).

Thanks for that post. I was actually gonna comment about the whole SLIAA vs SuperAA performance difference, since a number of people think the reason SLIAA is so much slower is because it uses SSAA, and that SuperAA it only MSAA, which looks worse but runs faster.

I posted earlier in another thread that it does appear SuperAA 8x is only using MSAA, while SLI does 2xSSAA in the 8x mode. Furthermore, it also looks like SuperAA 14x is only doing 2x SSAA, while SLI in 16x mode is doing 4xSSAA, which would again make it slower but looking better than SuperAA.

After closer inspection of the TR article, and the crossfire reviews at b3d, it actually turns out that SuperAA is in fact always doing 2xSSAA in addition to varying levels of MSAA. The confusion comes from the fact that in 8x and 12x modes SuperAA is taking 2 texture samples at the exact same position, which would make it look like pure MSAA, but the performance advantage can not be the result of pure MSAA. This can further be confirmed by looking at the performance chart on the same TR page you linked, which shows that 8xAA performs identically to 10xAA, and 12xAA runs at the same speed as 14xAA. So, this means 2 things:

1. The numbering scheme in AA levels is not interchangeable between Nvidia and Ati, and to be accurate, Nv's 8x mode should really be compared to Ati's 10x mode; and

2. The performance advantage of SuperAA compared to Nvidia's 8x mode cannot be blamed on faster pure MSAA with lower image quality, because Ati does use 2xSSAA in all modes of SuperAA.

In short, the difference between SuperAA and SLIAA is that SliAA uses 4xSSAA on top of (x)xMSAA while SuperAA uses 2xSSAA on top of (x)xMSAA???

 

[Pete]

I'd save $200 and get a 1900XT--or get it at NewEgg for ~$520, if you'd rather not open another account. Pay more for a retail version if you want a longer warranty. The XTX doesn't seem worth the extra $100+, unless you want to overclock it (in which case, forget the warranty).

A $760 GTX-512 isn't a serious option at the moment, in my opinion, considering nVidia is set to launch a 7900 that's rumored to be much faster and that'll probably sell for closer to X1900 prices in at least one model. Let me stress that this is just my opinion (lest I be flamed to heck and back), and that the GTX-512 does have its good points, like a much quieter cooler and SLI and NV's typically better OpenGL performance and Linux drivers. But I wouldn't pay $760 for one this close to the 7900, and especially when a 1900XT competes very well with it for as much as $250 less.

IMO, at this moment, a $500 1900XT seems like the best buy at the high-end. See why here. Don't take my word for it, tho--read as many reviews as you can, too. My thinking is that it performs very close to the XTX and the GTX-512, but costs a lot less. You can use the (big) money you saved to upgrade to the next generation even sooner--which is handy, considering both ATI and NV are supposed to be getting their next-gen cards ready for the end of this year, alongside Windows Vista. Yeah, they're getting pretty crazy with the release schedules.

And you could always "SLI" the X1900XT, with CrossFire. If you consider that, remember that the X1900 "Master Card" runs only at XT speed, so if you buy the XTX and a Master Card I think both will be clocked at just XT speeds--in other words, a waste of money on the XTX ... unless you plan on OCing, of course.

The Radeons have slower RAM, yeah, but they seem to make up for that by being more efficient with it.

No clue about the PSU, but I tend to distrust the flashier ones. You're going to want to be very picky with the PSU if you plan on running two cards like the X1900XT or GTX-512 (or 7900 GTX-512, I'm guessing) in one system. Make sure wherever you buy it from has a good return or exchange policy, or get them to specify that the PSU you're buying is good for SLIing two top-end cards.

Looks like a wicked rig. :thumbsup:

 

[BenSkywalker]

The point is you can't expect x1900CF to have a huge lead over gtx sli when the cpu is the limiting factor.

I would say the real point is that Dave needs to do whatever it takes to move beyond 1998 and get a real monitor

Towards the general conversation- it is extremely amusing to watch some of the idiots here trying to defend the magical price point of whatever their IHV of choice is at for their highest end part. 99% of the public will ignore Crossfire/SLI? 99.999% of the public is going to ignore the 1900xt/xtx too- what the he!l does that prove?

I think you all need a reality check- we are rather fanatical geeks. We will spend $500 or so and up a year on graphics cards. A huge portion of the population spends less then that on computer hardware every five years. Trying to argue that the $500 graphcs card is perfectly reasonable while the $800 is absurd is extremely amusing to say the least. We choose to spend our free time debating and discussing graphics card technology with people we have never met and do so at great lengths at times- the rest of the fvcking world thinks we are nuts no matter how much we dropped on our GPU/GPUs of choice

 

[Munky]

Originally posted by: RobertR1
In short, the difference between SuperAA and SLIAA is that SliAA uses 4xSSAA on top of (x)xMSAA while SuperAA uses 2xSSAA on top of (x)xMSAA???

SLIAA uses 4xSSAA in 16x mode, and 2xSSAA in 8x mode. SuperAA uses 2xSSAA in all modes.

 

[Pete]

Originally posted by: BenSkywalker
I would say the real point is that Dave needs to do whatever it takes to move beyond 1998 and get a real monitor

Heh, actually Dave just got one of those new Dell 30" LCDs. Can't tell if it'll help or hurt with AA comparisons, but at least he gets to see whether NV's AA is truly better with LCDs and ATI's, with CRTs.

 

[SiTeS]

Thanks for the advice Pete. I play Americas Army which is a OpenGL based graphics.. I believe(sorry for my stupidity when it comes to this stuff). Should that effect my decision at all?

 

[Pete]

No problem. I just hope my advice is right. To be on the safe side, maybe start your own thread on the subject and see what others recommend.

AA is pretty old by now, so performance should fly with any high-end card. It's not OGL according to the AA website. In fact, I think it's built on the old Unreal Engine, so it should default to D3D.

Seriously, tho, I'd suggest reading as many X1900 reviews and forum threads as possible. Read the reviews b/c most shoud compare a X1900XT with a 7800 GTX 512, so you'll see which is faster in which games and with what IQ settings. Read the forum threads by actual owners so you'll maybe get a heads up on any "real world" issues (like compgeek's thread on his X1800XT and how apparently ATI doesn't like ALT-TABing out of games, which may concern you--tho this may depend on the game, as I've had varying luck ALT-TABing with my 9800).