The Fermi Thread - Part 3

[96redformula]

if the leaks are right here is a graphic of the gtx480 i made bc i have too much free time. it has a 24% advantage overall so it could sell at $480vand offer same perf/dollar as 5870. i would have included 5970 but im having trouble finding reviews with these games.

idk why -1% ends up +1% on crysis.

Sigh..... that is off handpicked benches/benchmarks/settings from nvidia

36 hours couldn't come soon enough .

[Ramon Zarat]

Not really. Nvidia fanboys have taught us they'll buy anything even if it's overpriced. The GTX470 comes to mind. Same MSRP as the 5870, yet performs less and people are still willing to jump all over it.

You can't stop the green fanboys.

LMAO

[annihilat0r]

Not really. Nvidia fanboys have taught us they'll buy anything even if it's overpriced. The GTX470 comes to mind. Same MSRP as the 5870, yet performs less and people are still willing to jump all over it.

You can't stop the green fanboys.

lulz, it performs on par with the 5870 if these benchies are to be believed, and it looks like you can get it for $359 now.

Nvidia has several advantages like PhysX, CUDA and allegedly better tessellation performance. DX11 titles seem to favor Fermi more than other games so I'll take that as a good sign also. All in all, I think it's fair for it to be priced at HD5870 levels.

[Carfax]

I'm calling it now...

200.xx drivers will provide a significant boost in performance for Fermi, to the tune of 1000%

Thats Nvidia's style, and the 200.xx drivers will be a milestone so it makes sense..

[jaredpace]

Done

Sweet, thanks. Here are those Nvidia benches (in your sweet bar graph) next to the benches posted earlier...

[ElSel10]

Sigh..... that is off handpicked benches/benchmarks/settings from nvidia

36 hours couldn't come soon enough .

Those are all very popular games with reasonable settings of 1920x1200 4xAA?

[Chickenfeed]

I've been 30 inches for a while now and once you go there you just can't go back!

Lol epic english ftw. So sigged

If the 480 pulls ahead noticeably in the games im playing / interested in, it might get my attention. I've played a good bit of borderlands but it is a title which ATI hardware does poorly in. Been playing Just Cause 2 (really fun by the way if you like mindless explosions and gunplay in a giant sandbox environment which is very very nice to look at ) a lot today and although it does fairly well on a 5870 if a 480 can do at least 30-40% higher in it, that alone would be incentive enough for me to take a 2nd look. However the question is do I want to pay another $200 to do this. I might have another 5870 coming my way ( replacement for my 4870x2, fingers crossed ) so we shall see.

[Ramon Zarat]

lulz, it performs on par with the 5870 if these benchies are to be believed, and it looks like you can get it for $359 now.

Nvidia has several advantages like PhysX, CUDA and allegedly better tessellation performance. DX11 titles seem to favor Fermi more than other games so I'll take that as a good sign also. All in all, I think it's fair for it to be priced at HD5870 levels.

On par with 5870 performance and cheaper would make the 470 a great buy indeed. We will confirm that in less than 30 hours now...


There is 10-12 games that support PhysX, out of which 10 are very bad. OpenCL is what will take the center stage in Q2-Q3 2010 and will be in full swing in 2011. Closed and proprietary CUDA won't be here for much longer on the desktop. Everything GPGPU on the desktop will soon be driven through OpenCl, including physics acceleration, transcoding etc...

EDIT:

Sorry, There is many more games than 10-12 on PhysX. I meant to say only about 10 games are actually half good...

[zerazax]

Funny how some people don't like the rumored benched numbers of cards out in the wild, but will take PR number comparisons to heart!

Exactly. ATI has had what, 7 months or so to refine their drivers? Nvidia should easily add about 15 to 20% (possibly more) increase in performance from driver optimizations alone over the next few months..

I'm betting the 200.xx drivers will be the first to have real G100 optimizations..

I don't know about ATI, but this is fairly routine with Nvidia.

The G200's release drivers were the 170xx, but it wasn't until the 180xx drivers that the architecture received it's first significant boost....to the tune of 10 > 15%.. Of course the recent 190xx drivers added even more, so all in all I'd say my GTX 285 is about 20% faster on average right now than it was on release day.

And remember that the G200 isn't as radical a change in architecture as Fermi is..

And again, Nvidia has had at least 4 months, if not more, working on drivers. Specs might not have been finalized, but the drivers and architecture were well known in advance. Yes, it's a newer architecture, but the driver team has definitely had a lot longer to work on it than GT200 did. So in relation to GT200, we're probably at the 180.xx range more so than the 170.xx range.

Either way, I'm not betting on the drivers to unlock 10-20% when the card is already 6 months late. We're talking about the fact that in 6 months, new competition/refreshes will be out and they'd have to be compared to those cards.

if the leaks are right here is a graphic of the gtx480 i made bc i have too much free time. it has a 24% advantage overall so it could sell at $480vand offer same perf/dollar as 5870. i would have included 5970 but im having trouble finding reviews with these games.

idk why -1% ends up +1% on crysis.

I've made the calculations:

GTX 480 beats the HD5870 %24 average on 1920 + 4xAA, and 23% average on 2560 + 4xAA.

GTX 470 performs nearly completely equal to HD5870 on both settings.

if those benches are correct.

You guys are calculating +24% overall when it's based on a limited selection of games, chosen by Nvidia, with settings unknown. I mean, look at Batman:AA... everyone knows Nvidia does better on that game by a good amount, but that's being factored into an already small sample size, meaning the %'s will look even better for Nvidia.

Let's wait for the real numbers before concluding +%'s shall we? Especially from PR numbers!

[Trigger5521]

I got three letters for you, S-L-I... 3008WFP here and I had 3007WFP before this one. I've been 30 inches for a while now and once you go there you just can't go back! Look at the increases with SLI, IMO the scaling is better than crossfire solution which is mostly why I'm still sticking with Nvidia.

I'm with ya...just a little disappointed that I won't be able to use the display port on the 3008WFP.

I'd like to see 470 #'s in SLI. If they are halfway decent it's a more attractive option for me as I can use my current power supply. I REALLY don't want to rewire my system, I spent retarded hours on wire management.

[Ramon Zarat]

I double checked on Nvidia website about Physx hardware accelerated game and it seems I was right the first time... *Only* 17 games total, most of which are total crap.

Batman: Arkham Asylum
Crazy Machines II
Cryostasis
Dark Void
Darkest of Days
Tom Clancy's Ghost Recon Advanced Warfighter 2 (GRAW2)
Metal Knight Zero Online (MKZ)
Mirror's Edge
Nurien
PT Boats: Knights of the Sea
Sacred 2
Star Tales
Unreal Tournament 3
Unreal Tournament 3: Extreme Physics Mod
U-WARS
Warmonger: Operation Downtown

Sources: http://www.nzone.com/object/nzone_physxgames_home.html

Anyone still think Physx is a valid argument to mention when talking about Nvidia?

[trinibwoy]

Heh, these 'PR' numbers sure are rattling nerves. You would almost think folks were hoping it's slower lol.

[zerazax]

Heh, these 'PR' numbers sure are rattling nerves. You would almost think folks were hoping it's slower lol.

Rattling nerves? More like watching people downplay independent reviewers / leaks out there, then taking the PR slides as gospel

Anywho, question... I thought AA was disabled for ATI cards. How do those slides show Batman:AA with 4x? I know you can brute force it, but there's an example of "selective" benching

[Ramon Zarat]

26 hours left...

[annihilat0r]

You guys are calculating +24% overall when it's based on a limited selection of games, chosen by Nvidia, with settings unknown. I mean, look at Batman:AA... everyone knows Nvidia does better on that game by a good amount, but that's being factored into an already small sample size, meaning the %'s will look even better for Nvidia.

Let's wait for the real numbers before concluding +%'s shall we? Especially from PR numbers!

How many games are needed to make an accurate comparison, do you think? Most review sites don't even bench this many games. I obv. agree with you on that these are Nvidia numbers and we shouldn't base conclusions on these, however.

About the other benchmarks you're talking about, which ones? AFAIK there's a circulating Crysis benchmark which seems to be in line with these scores, a couple of Dirt + FC2 benchs which probably were ripped from this chart. Crysis is the game Fermi seems to be the worst on.

[zerazax]

How many games are needed to make an accurate comparison, do you think? Most review sites don't even bench this many games. I obv. agree with you on that these are Nvidia numbers and we shouldn't base conclusions on these, however.

The question isn't how many games... it's what games.

In a relatively small sample size, if you let too many outliers in, you're going to skew data.

IIRC, back when G80/R600 were the cards on the market, people would bench Call of Juarez and say "look, R600 is not as slow!" etc.

Now imagine if a review had 10 games, 3 of which were Call of Juarez esque... instead of the realistic G80 being 30% faster than R600, the average might have dropped to 20%, etc.

If you're going to make a definitive overall comparison, with just 10 games, you'd want all 10 to be as close to neutral as possible.

I mean look at the original big chart... just taking a glance, you've got Batman:AA and FarCry 2, well known to favor Nvidia. Aside from the obvious question of how AA was enabled for ATI's cards (again, issues with settings), you have one game accounting for a good chunk of the performance increase. And do you factor in a review like Crysis Warhead 2560x1600 8xAA/16xAF where 17.2/4.7 is going to be "OMGWTF 300% faster!!!" etc.

PR slides at its best. They'll no doubt champion "25% faster than 5870" but whats the methodology?

(that's also why I tend to ignore reviews that don't clue us in to how things were tested, what was tested, and having a conclusion based off a small set of samples. junk science at its best)

[annihilat0r]

The question isn't how many games... it's what games.

In a relatively small sample size, if you let too many outliers in, you're going to skew data.

IIRC, back when G80/R600 were the cards on the market, people would bench Call of Juarez and say "look, R600 is not as slow!" etc.

Now imagine if a review had 10 games, 3 of which were Call of Juarez esque... instead of the realistic G80 being 30% faster than R600, the average might have dropped to 20%, etc.

If you're going to take an overall comparison, with just 10 games, you'd want all 10 to be as close to neutral as possible.

I mean look at the original big chart... you've got Batman:AA and FarCry 2, well known to favor Nvidia. Aside from the obvious question of how AA was enabled for ATI's cards (again, issues with settings), you have one game accounting for a good chunk of the performance increase. And do you factor in a review like Crysis Warhead 2560x1600 8xAA/16xAF where 17.2/4.7 is going to be "OMGWTF 300% faster!!!" etc.

PR slides at its best

You are making absolutely no sense to me. There are about 15 games in that list and all are popular games which are constantly being used in reviews. Do you think some crucial games were left out in favor of Nvidia-centric titles? I can't see any (except Call of Pripyat). But I do see some titles ATI has been promoting heavily, like Battleforge, Dirt 2 and AvP.

And how is it that we decide which title is "neutral". I don't care if a game is neutral or not, and so shouldn't you. I just look at the importance of the game for me. If you are going to review a card, the games you must use should be popular titles that you know a lot of people would care about. A consumer doesn't and shouldn't care about how the card he is going to buy performs in a "neutral" game, he cares about games he is actually going to play.

I see nothing related to PR in these benchmarks, there are a lot of games including ATI titles, and a lot of settings some of which clearly don't favor Nvidia. If someone was looking for 1680x1050 numbers he shouldn't be buying a GTX 470 or 480 in the first place.

[Ramon Zarat]

The question isn't how many games... it's what games.

In a relatively small sample size, if you let too many outliers in, you're going to skew data.

IIRC, back when G80/R600 were the cards on the market, people would bench Call of Juarez and say "look, R600 is not as slow!" etc.

Now imagine if a review had 10 games, 3 of which were Call of Juarez esque... instead of the realistic G80 being 30% faster than R600, the average might have dropped to 20%, etc.

If you're going to make a definitive overall comparison, with just 10 games, you'd want all 10 to be as close to neutral as possible.

I mean look at the original big chart... just taking a glance, you've got Batman:AA and FarCry 2, well known to favor Nvidia. Aside from the obvious question of how AA was enabled for ATI's cards (again, issues with settings), you have one game accounting for a good chunk of the performance increase. And do you factor in a review like Crysis Warhead 2560x1600 8xAA/16xAF where 17.2/4.7 is going to be "OMGWTF 300% faster!!!" etc.

PR slides at its best. They'll no doubt champion "25% faster than 5870" but whats the methodology?

(that's also why I tend to ignore reviews that don't clue us in to how things were tested, what was tested, and having a conclusion based off a small set of samples. junk science at its best)

That's why we will need to be *EXTRA* vigilant as to who we trust when it comes to official review on the 26. All manufacturers are known to twist reviewer's arms a bit, but IMO, Nvidia -definitively- have a worse track record in that department.

[zerazax]

You are making absolutely no sense to me. There are about 15 games in that list and all are popular games which are constantly being used in reviews. Do you think some crucial games were left out in favor of Nvidia-centric titles? I can't see any (except Call of Pripyat). But I do see some titles ATI has been promoting heavily, like Battleforge, Dirt 2 and AvP.

And how is it that we decide which title is "neutral". I don't care if a game is neutral or not, and so shouldn't you. I just look at the importance of the game for me. If you are going to review a card, the games you must use should be popular titles that you know a lot of people would care about. A consumer doesn't and shouldn't care about how the card he is going to buy performs in a "neutral" game, he cares about games he is actually going to play.

Okay, and what does your statement have anything to do with comparing across the board? You just said you don't care if a title isn't neutral...

And I agree to get it for the games you care about, but again, if people say 24% across the board, but the games you care about are on the 10% end and not the 50% end, who's selling what?

I see nothing related to PR in these benchmarks, there are a lot of games including ATI titles, and a lot of settings some of which clearly don't favor Nvidia. If someone was looking for 1680x1050 numbers he shouldn't be buying a GTX 470 or 480 in the first place.

A lot of settings which clearly don't favor Nvidia? Where's the list of these settings?? I don't see where you're drawing that from at all

[GoldenTiger]

Heh, these 'PR' numbers sure are rattling nerves. You would almost think folks were hoping it's slower lol.

Yep, sure seems to be... ROFL.

[zerazax]

Rattling as much as people were pissed off about performance numbers leaked? Of course, pr slides get released and those must be true!!

That's why we will need to be *EXTRA* vigilant as to who we trust when it comes to official review on the 26. All manufacturers are known to twist reviewer's arms a bit, but IMO, Nvidia -definitively- have a worse track record in that department.

That's exactly what I mean by watching out for testing methodology, settings, etc.

If the reviewers pump out reviews that are vague on specifics, I usually end up second guessing how they got to those numbers. After all, you wouldn't submit a science report for your lab if you can't give specific values and instructions so others can replicate it within the margin of error. Otherwise you'd be a politician

[hurleybird]

What drivers do you think the Nvidia PR numbers are using. 10.3? Doubtful. 10.2 is more likely, but they might even be older knowing how PR works.

[96redformula]

After all, you wouldn't submit a science report for your lab if you can't give specific values and instructions so others can replicate it within the margin of error. Otherwise you'd be a politician

Best quote ever right there

I think I might Sig that....

[Chumbucket843]

You guys are calculating +24% overall when it's based on a limited selection of games, chosen by Nvidia, with settings unknown. I mean, look at Batman:AA... everyone knows Nvidia does better on that game by a good amount, but that's being factored into an already small sample size, meaning the %'s will look even better for Nvidia.

Let's wait for the real numbers before concluding +%'s shall we? Especially from PR numbers!

i cant really make up data that they dont give me. i just assume that people realize that it could be fake or manipulated. the criteria for games were dx11 and newer gpu heavy games. out of the 8 games i chose to use 4 could not be maxed out by the gtx480.

[annihilat0r]

Okay, and what does your statement have anything to do with comparing across the board? You just said you don't care if a title isn't neutral...

And I agree to get it for the games you care about, but again, if people say 24% across the board, but the games you care about are on the 10% end and not the 50% end, who's selling what?

Sure I agree that "across the board" performance summaries don't mean much when deciding to buy a video card. But I am talking about them (and have calculated those scores) because there had been a lot of talk regarding how faster than 5870 it would be generally. Charlie said 5%, some other guy 25%, someone 15%, someone 40% etc.

But you're right in that if Charlie wanted to, he could cherry pick settings, games and scenarios to make himself correct. So could pretty much everyone (remember Nvidia says a 8600gt can be faster than HD5870 lol), but these good selection of games I think would allow for a generalized "this card is better than that card generally % something, at this resolution and settings" statement.

A lot of settings which clearly don't favor Nvidia? Where's the list of these settings?? I don't see where you're drawing that from at all

What I meant was that the resolutions weren't cherry picked from game to game. There are numbers for 1920+4xAA, 1920+8xAA, 2560+4xAA and 2560+8xAA (the only settings a Fermi buyer should care about) for pretty much every game in the benchmark, and in some of those resolutions HD5870 performs a lot better compared to other resolutions, but these weren't omitted and a lot of data was presented whether it favored Nvidia or not.

Anyway, if these benchmarks are fake I'll have a good laugh at this discussion