GeforceFX (vs. Radeon 9700 Pro) Benchmarks

Obiwan Kenobi writes "Looks like they guys at Maximum PC got lucky -- they scored the first ever GeforceFX benchmarks via an Alienware prototype machine. Two 'marks to notice: The Geforce FX scored 209 FPS in Quake 3 (1600x1200x32) and 41fps in 3dMark Game4 demo, while the Radeon 9700 Pro attained only 147fps in Quake 3 yet came back with 45fps in the 3dMark test. It seems that the GeforceFX is the clear leader in pure processing power, but in memory bandwidth the 9700 Pro is still king."

Comments

Finally, a decent frame rate.

[Anand_S]

ATI's 147 fps has always been a problem for me in Quake. I like to blink a lot.

Re:Finally, a decent frame rate.

[OldStash]

True gamers blink between frames.

Re:Finally, a decent frame rate.

[SimplyCosmic]

Conversely, simply having a higher average fps number doesn't guarentee that the highs and lows will be better than the lower average fps.

There's nothing in that single number to say that the higher average fps doesn't suffer from a number of wildly varying large peaks and valleys in performance, while the lower number could be much more steady, with relatively low variance in the number across its peaks and valleys.

It only makes you wish that these benchmarks, especially the "real world" Quake 3 tests, had a graph of fps throughout the test to see how performance was at any particular point.

Re:Finally, a decent frame rate.

[Masem]

I certainly have no problem with id and others trying to push the limits of real-time 3D rendering power, or the hardware makers making boards that can do that.

However, I do question the point where much of the work for games is put into the engine, and little of it put into the gameplay itself. I realize that we're almost at the point where one company is making the engine, and another company licenses that engine to make the game, so the responsibility of good gameplay is on the shoulders of the latter company. However, it seems that a lot more of the games that have been published of late focus more on the realism and the 3d-modeling that the playability of the game, and the continuous push to up the abilities of rendering does not seem to allow the developers of new games to step back and think about gameplay.

The other problem is that right now, with the specs we're getting on Doom3 and other games, it sounds like another focused hardware upgrade cycle if you want to play these games reasonably. Sure, you can drop the screen resolution, and there's probably hundreds of tweaks you can apply to the engine to cut back details, but older, but still viable cards, will have problems. I know people don't want to develop for outdated systems, but there is a point where you have to include a reasonable amount of backwards compatibility to allow non-power gamers to play new games as well. One of the reasons that Half-Life and CS sold so well was that the game was optimized for play on a previous generation of processor/vidcards compared to the average system that was sold. (HL/CS, also, IMO, excels at it's gameplay as mentioned above). I know a lot of PC game writers are of the opinion that the gaming market will only move forward when vidcard makers put out new features into cards, and then when PC game makers follow up by using those new features in predominate titles, but the PC gaming market is just not healthy right now, and to make games that require the latest-and-greatest hardware will limit sales further and may push this part of the market into a slump, while console gamers will continue to see more improved titles.

Again, I'm not against improvements in 3d rendering tech and pushing polys as fast as possible; it's the game makers themselves that need to realize what the average hardware of their target audience is going to be and not just to focus on how pretty the game looks.

Re:Finally, a decent frame rate.

[afidel]

The reason Carmack designs his engines for the top of the line card at the time of launch is that ID only makes ok money with their own games. Where they rake in the cash is selling the engine to other game companies, who then go and make another game around it. With game development lifecycles being fairly long with regards to hardware lifecycles this makes a fair amount of sense. The fanboys will go out the day doom3 ships and buy a new top of the line rig, for the rest of us that technology will get into our computer over the next year or two, which probably is in line with the amount of time it will take the companies that buy the ID engine to make their games. So Carmack puts everything into his engine because he knows that by the time most people use it their hardware will be up to snuff. Remember when he started working on the Doom3 engine the idea of programmable pixel shaders was just that an idea, now most people who play games have a card with a PPS. If Carmack did all of his engine design based on the hardware available when he starts the design it would be outdated before it ever got used outside ID.

Keep this in mind..

[OutRigged]

I'm sure the GeForce FX drivers they were using were early beta versions, and as such, not optimized to the standard of which release drivers are.

I'll wait for the final hardware and drivers before I decide which to buy.

Re:Keep this in mind..

[Anonymous Coward]

This is total bullshit. Why was this idiot modded up? Because his username is X-BOX LIVE DEV TEAM? Nvidia drivers from beta to final have had *dramatic* speed improvements in the past.

Re:Keep this in mind..

[prefect42]

Which is why in the article they say that the drivers are beta and as such the results should be viewed as beta too.

Come on, read the article.

The reluctance of NVidia to allow them to test the higher levels of AA is more telling if you ask me.

Re:Keep this in mind..

[nautical9]

OutRigged was correct - there can be MASSIVE differences between driver revisions, and there typically is, which is why "leaked" betas are so wildly popular in the O/Cing crowd. I grab a new one every month or so, and although I may not see much of an improvement 4 out of 5 months, my benchmarks will suddendly leap up 10-20% (even NVidia claims the recent 40.x reference drivers are 25% faster than the 30.x). I also find the newer your card, the bigger the performance gains you typically see by upgrading the driver.

The reasons for a speed increase aren't always related to the graphics card itself, but can be due to the motherboard chipset, type of RAM, BIOS, or even a specific game or app itself. These tweaks will change how the card communicates with these in specific circumstances, which can be vary greatly between different consumers' machines.

Since the graphics card industry is hugely competitive right now, it's in their best interest to spend a lot of time tweaking their drivers to the max.

The reason consoles don't worry too much about it is because they have a standard set of hardware (read: one graphics card - no competing card that customers can benchmark against) that ALL game developers must work with. This also simplifies game development because they know the exact config and driver set that EVERY user will be using.

Even though I'm sure they COULD tweak the drivers (forgetting the expense of distributing a firmware patch), they'd prefer to leave the tweaking in the game code. Besides, you can't easily benchmark the various consoles against each other, whereas the graphics card folks for PCs know that every performance site and magazine is going to use the exact same hardware config and same game to test their card against all others.

competition

[vistic]

Competition like this kicks ass. The big players taking turns taking the lead. I only wish Matrox were making a larger effort than the Parhelia.

What I am surprised about though is that prices are so high for graphics cards still even with relatively good competition in the marketplace. I mean even the Parhelia debuted at like $400 didn't it?

It always seemed to be that the benefit of having AMD competing with Intel, was that I could get a really good CPU pretty cheap. (Though now it seems AMD is taking it easy for awhile, so that benefit may have been short-lived.) Yet I don't see the competition driving video card prices down.

There's some evil conspiracy afoot here, I know it!

Re:competition

[netwiz]

What I am surprised about though is that prices are so high for graphics cards still even with relatively good competition in the marketplace. I mean even the Parhelia debuted at like $400 didn't it?

It's mostly due to high-end video chipsets costing so much, plus the added expense of the uber-fast memory that these cards require, but mostly, it's driven by ultimate demand for these products. Everyone needs a P4/Athlon XP, but only a few people need the absolute fastest display adapter out there. As a result, fewer units get produced, as fewer units will actually sell. Combine that with the already higher cost of producing core logic that's 1.5-2x the transistor count of high-end CPUs and RAM that's 2-3x faster than desktop stuff, and you've got a recipe for pricey hardware.

Also, don't forget that most products these days are priced at what the market will bear. People will pay $400 for the fastest thing on the block, so that's what they sell for. My general rule of thumb is to wait a month or two after the new, fast, whiz-bang product, then buy whatever card has the fewest problems and costs $300.

3dMark 2001 is not the end all be all

[Aggrazel]

3dMark 2001 is a guestimate on how fast things will work, its meant to torture your card and Game 4 (nature) is just that, the most punishing thing they could come up with.

But it is actual game performance that is important with most people, so while you may get better 3dmark scores, most people aren't running that a whole bunch to see those nifty graphics, they'd rather be running games.

Also, don't forget to mention that all these tests were run with 2xFSAA on.

Q3 Framerates are WITH 2xAA turned on

[uncleFester]

For those who don't rtfa, the quake3 framerates for both cards had 2x antialiasing turned on. When thrown in the mix, this becomes a bit more impressive than the simple 1xx fps rate shown, as a number of current cards can achieve mid- or high-100s speeds but with no AA.

It's not simply the frame rate, but what's actually being generated in that frame.

=r

From the Article....

[RebelTycoon]

And if you really twisted our arms, we'd bet money that it will be running on a 0.13-micron core and using 256-bit DDR II memory.

And if we grab your nuts and twist, you'll confirm this? And if we threaten to cut them off... I think you'll scream just about anything...

So let me guess, they know what's coming from ATI... But like they said, its not about bandwidth, its about GPU processing power, so how will a even bigger pipe that ATi isn't filling going to help....

Re:Don't wait....

[AmigaAvenger]

ATI has had some notoriously bad drivers that should not have shipped though... (ask some early owners of the various AIW cards...)

Who cares

[nuggz]

Geforce FX scored 209 FPS in Quake 3 (1600x1200x32)
while the Radeon 9700 Pro attained only 147fps


So what they are saying is that even at a ridiculous resolution, either card is capable of a higher framerate than your monitor, and your eyes.

Re:Who cares

[Lethyos]

So what they are saying is that even at a ridiculous resolution, either card is capable of a higher framerate than your monitor, and your eyes.

With frame-rate to spare, you can achieve really silky-smooth images by syncing with the monitor's refresh. This prevents those ugly redraw lines that can occur from the next frame being drawn right as the display refreshes. So, having a rate higher than your refresh can be useful.

nVidia and driver performance

[netwiz]

I'm willing to bet that there's another 20-30% in the FX due to driver tuning. nVidia typically releases a new product, then, after about two to three months, releases a driver that actually makes the card fast.

Plus, if this is the first of the GigaPixel cores, then there should definitely be more in it, and the fact that it's down on memory B/W shouldn't make much of a difference.

Re:What's the big deal?

[Anonymous Coward]

Yet another misguided fool. The eye can see considerably more than that, up to about 70-90 FPS. You will notice the difference between a game that runs at 30 stable FPS and one that runs at 60 stable FPS. The 24-30 FPS on TV and cinema is only perceived as fluid because of motion blur.

And even in a game, having FPS over say, 70 is useful because the frame rate will vary. When there's suddenly much action on the screen the frame rate will drop...

What I miss in all these benchmarks

[archeopterix]

They never test the number of text lines per second in text mode. Or Nethack FPS. My card does 7.5 FPS in Nethack, if I click the keys really fast.

Re:What I miss in all these benchmarks

[Zathrus]

While humorous, once upon a time it did matter how fast you could scroll text, and cards would be benchmarked based on how fast they could do it in a window (doing it in a FS text session was a non-issue).

I won a Number9 Imagine128 card at Comdex back in the early 90s... I distinctly remember being amazed since for the first time ever it was faster to scroll text in a window than it was full-screen.

Nowadays it's a total non-issue of course.

Oh, and I get far better FPS in Nethack. You're just a slow typist ;)

Re:What's the big deal?

[martyn s]

Aside from all the other things being mentioned here, another thing to take into account is motion blur. When you watch a movie at 24 fps, or television at 30 fps, motion blur makes sure you don't miss anything between frames. If the action is moving fast enough, it will appear as a blur in a single frame, so you're not missing anything "between" frames. Not so in a videogame. In a videogame, if the action is moving too fast there will be "gaps". So even though, technically the eye can't see anything above a certain frame rate, you can't really directly compare television or movies to video games as far as framerate is concerned.

Not very fair

[GauteL]

Comparing future products against real shipping products is not very fair without at least keeping this in mind. This article barely mentions it.

ATI might very well ship an improved version around the time GeForce FX ships.

Re:nVidia vs Everyone else

[Ninja Master Gara]

The driver optimizations have similar effects on game performances. It's why nVidia's drivers are so highly praised. The benchmark optimization is retarded, but everyone does it. But nVidia gets it with the games too.

these figures look more realistic than the figures Nvidia told when the card was announced!

This is true. Marketing is evil. Evil evil evil.

this card will probably not be much more than 20-25% faster than an radeon 9700

When DirectX 9 is out the door, it will not only be faster, but look better. Much better. I suppose I'll catch flack for buying into the hype, but I've been blown away every time.

Besides - who cares about those $400+ gfx cards?

You're right about that last part. For me, it's just about keeping track of who's leading the industry and what technology my next card will have, when I buy it in a year for $120 :)

4xAA

[cca93014]

They need to run the demos with 4xAA. The 9700 and NV30 are so fast as to make FPS irrelevant and eye-candy relevant.

If the game is running at 100 fps people are going to up the eye-candy, right?

Assuming this is the case, I seem to remember the 9700 getting very similar scores whether the card was set to no AA, 2xAA or 4xAA, i.e. the AA processing was almost (but not quite) 'free'.

I know the benchmarks are very very early and it really needs to get the full treatment from a hardware site, but the important figures IMHO are ones where the card is set to run everything maxed out...I have a feeling the NV30 is not going to be in such a prominent position in that instance...

AMD have NOT lost the CPU war

[W2k]

I see this alot nowadays - people saying that AMD have "lost their edge", or "been taking it easy for a while" ... that is simply not true. An AMD Athlon XP 2800+ _will_ beat an Intel Pentium IV at 2.8 GHz in most benchmarks (and the 3.06 GHz P4 in quite a few - see the latest ones at THG or AT if you don't trust me), just as it is supposed to. And you can still practically get two Athlons (not 2800+'s mind you) for the same price as one high-end Pentium IV. Surely no-one here thinks that a single P4, HT or no HT, stands a chance against a true SMP system (given apps that take advantage of both CPU's)?

Furthermore, there's no app or game available on this earth, and there probably won't be for at least two years to come, where the speed difference between an AXP/2800+ and a P4/3GHz is big enough to really mean anything to anyone other than the fanatical overclocking crowd, who will spend any amount of money just to have the fastest stuff on the market, only to use it for stuff like playing Counter-Strike, which uses perhaps 20% of the total CPU and graphics card capacity. Well, if you're into that sort of stuff, sure. Get a P4 and enjoy having the fastest CPU there is .. until the next model P4/AXP is out, that is.

For the rest of us, who base our computer purchases on common sense, for speed, stability and price, the obvious choice is still the Athlon XP.

Besides, the Pentium IV still has a pretty fucked up design. See this page if you don't know what I'm talking about. I always laugh at people who whine that Windows is poorly designed, only to praise Intel CPU's in the next breath.

Anyone care to disagree? Remember, modding me down is so much easier than posting an intelligent reply.

Re:Hm...

[Zathrus]

Hey, why don't we test it with Quake1? Bet we can come close to 1000 fps.

And, yet, even with Q1 running at 1000 fps UT2k3 only runs at 140 fps. Wonder what something with even more complexity than that would run at... oh, look, there's a benchmark that only did 41 fps.

Or go look at CodeCult.com's Codecreatures, which does a lovely 6 fps on a 2.5 GHz P4 w/ Radeon 9700 at 1600x1200 anti-aliased and ansitropic filtering. And it still doesn't look real.

Until we have holographic imaging that's indistinguishable from reality the cards aren't there yet. If you don't need/want it, then fine, don't buy it. But whining that it's clearly beyond what's needed is, well, stupid.

Re:Hm...

[Vireo]

In all seriousness... In Quake 3, the physics model is tied to your framerate (i.e. a new snapshot of the "world" is computed at each frame). It is well known among avid quakers that the physics is different for different framerate, and that there is an optimum at 125 FPS. This has nothing to do with the visuals. You can go faster and jump higher when getting 125 FPS. In one-player mode, it is possible to separate world snapshots and visual frames, but not in multi-player mode. So most gamers will in some ways try to achieve above 125 FPS and then cap it (using the com_maxfps in-game variable) to 125 FPS. It is then important that the card do above 125 FPS in all maps, all occasions (moreover in heavy battles involving many models and thus many polygons).

I can't talk for other games, but since the Q3 engine is widely used, that may be the case for some of them too. That partly explains the "need" for high-FPS.

Another thing to remember

[eviltypeguy]

The GeForce FX has more hardware capability and increased hardware precision then the *CURRENT* revision of the R300. While the ATi card is only 96-bit effective floating point precision through the pipeline. Hopefully, this oversight will be corrected by ATi with the R350 or R400. As Carmack said:

NVIDIA is the first of the consumer graphics chip companies to firmly understand what is going to be happening with the convergence of consumer real-time and professional offline rendering. The architectural decision in NVIDIA's next-generation GPU to allow full floating point precision all the way to the frame buffer and texture fetch, instead of just in internal paths, is a good example of far-sighted planning. It has been obvious to me for some time how things are going to come together, but NVIDIA has made moves on both the technical and company strategic fronts that are going to accelerate the timetable over my original estimations.

My current work on Doom is designed around what was made possible on the original GeForce, and reaches an optimal implementation on NVIDIA's next-generation GPU. My next generation of work will be designed around what is made possible on NVIDIA's next-generation GPU.

- John Carmack, id Software

So even if the GeForce FX is a bit slower for some things, those games that are using full DX9/OpenGL features will get better looking graphics thanks to the increased hardware precision. People using 3D programs like Maya with the Cg plugin will notice the biggest difference especially IMO. And at this point, NVidia's shaders are far better geared to the professional 3D graphics industry than ATi's *current* offering. This might encourage many developers to take advantage of extra GeForce FX features instead of ATi features.

(Source URL for Quote:
http://www.nvidia.com/content/areyouready/ twimtbp. html)

4xFSAA, Anisotropic filtering?

[stratjakt]

Without the goodies on, even the Ti4600 can "outperform" the R9700.

Hard to imagine a 'serious review' site would neglect to test these features. I don't give a crap about 400 average FPS in quake, but I do care if it drops to 14 with all the enhancements turned on. But then they were trying to make the GeForceFX look like it's leaps and bounds better.

I'd imagine it's still the case - the 9700 is still the bandwidth king. Personally, I don't care about faster (when its already faster than my monitor can display and brain can process). My next upgrade will be motivated because it will look better.

The GeforceFX isn't something thats going to leave the 9700 in the dust - it's something that should have come out 6 months ago to compete head-to-head with ATI.

At any rate, after putting together a couple of cheap flex-atx pcs with onboard S4s (shared memory - Shuttle FV25 in case anyone cares), I'm surprised at how little GPU horsepower is needed to actually play most games.

Even UT2k3 is playable on these little guys (albeit not 1600x1200 with all the goodies turned on, but playable). I'm pretty sure my "outdated" radeon 64vivo will play Doom 3 when it goes gold.

Anyhow, my point is that cards have been displaying 'fast enough' for awhile - I mean we don't measure a cards performance in polygons anymore. They need to "look better", as in more natural, smoother, more TV-like.

Re:4xFSAA, Anisotropic filtering?

[MikeBabcock]

I have to agree; I like the look of ATI's 16x oversampling along edges better than 4xFSAA in general. ATI's rendering just looks better to me.

I can still tell which arcade machines use 3dfx-derived/built chips; the graphics just look different. I want my games to look amazing; what good is moving grass if its pixelated??

Re:dx8 vs dx9

[Sepherus]

As there aren't any DirectX 9 games out at the moment, DirectX 8 tests are still important.

Fast, but not fast enough

[Aqua OS X]

Considering that ATI has a new card coming out soon, I doubt nVidia will get to be king of the hill for very long.

Re:not as many units?

[strictnein]

HP and Compaq now only use Geforce 4 MX 420/440/460s on their higher end machines (HP 763N, 863N, 873N, 883N, Compaq 6370US).

The rest of them use either Intel Extreme Graphics (which are ok (HP 753N, Compaq 6350US)) or a lower grade intel card. They also use cards like the S3 ProSavage (HP 523N, Compaq 6320US)).

Sony uses crap S3 (or is it SiS?) video cards in all of their desktop computers. Which is disappointing. For their laptops they use mainly ATI Radeon derivatives.

Suprisingly, you are much more likely to find an ATI or nVidia card on a HP/Compaq/Toshiba laptop than a desktop, probably because their is still some profit margin left in laptop computers (unlike desktops). Compaq's mid-range 1500 series is all ATI Mobility Radeons, as is HP's mid-range 4000 series and high-end 5000 series. Toshiba's base 1410 series uses Geforce 4 420 Go chipsets, their mid-range 1900 series uses the 440, and their high-end 5200 series uses the 460.

And yes... in case your wondering, I do currently sell computers =)

I don't believe this...sorry...

[waltc]

These benches are more of the same passed-around mush that nVidia's been handing out since October. Wake up and smell the coffee, people--these are the same programs nVidia handed out in the October handouts for benchmarking. Did the reviewer have a gun to the back of his head, so that he couldn't mange to run *anything* else? How convenient.

By the author's own words, this was no review. There are no 6x, 8x FSAA tests, at all, although these are supposed capabilities of GF FX--there are no screen shots for comparison--in otherwords, there is absolutely nothing to prove this ever took place. There are no anisptropic filtering tests, we don't know what cpu system the Radeon 9700 benchmarked on--nothing--absolutely nothing of interest that you would normally see in a real review is present. Even if you believe the author--he says unapologetically he was under direct duress by nVidia as far as what he was permitted to show AND SAY.

Already people on the Rage3D forums are talking about how much slower the 9700P speeds are in this promotional propganda piece than they themselves can get with their systems at home.

Also....what, pray tell, would Alienware be doing with a NVIDIA beta prototype? As a small OEM I would expect that if anything Alienware would have an OEM beta version of the card--possibly. Certainly not a nVidia version of a prototype card! If nVidia needs Alienware to beta test its upcoming card this must mean nVidia hasn't even finished the prototype reference design yet and nVidia's OEMs haven't even begun production!

Here's what I think it is: a paid-for promotional piece which is designed to deter people from going ahead and buying an ATI 9700 Pro. What it most certainly is not is an actual review of the product--by the words of the author himself. What I still can't get over is that these are the very same benchmarked programs nVidia was handing out in October!

When nVidia starts sending out cards to reviewers with driver sets and saying, "Have at it--review it any way you like!" that's when I'll start listening.

Framerate jokes and VisSim vs. PC Games

[GoSpeedRacerGo]

(ooops)...

One quick point to address all the 150 fps in Quake jokes:

Frame rate consistancy is what is most important (by far). A game that runs at 30 fps solid will also feel better than a game that runs at 60 fps some of the time but then bops back and forth between 30 fps and 60 fps.

The VisSim industry has done a better job of saying "we only need 60 Hz (fps) but we better never ever see you dip below that or you are out!" This forces hardware and software to be optimized for locking at 60 fps. I can tell you that a 60 fps Air Force flight simulator will always feel higher performance than a soupped up PC running Quake at 100 fps but dipping down to 50 fps or worse when things get hairy.

The biggest evidence of this issue being unimportant in PC gaming is the number of people or games that run with vertical blank (vblank) synchronization turned off. This is wrong wrong wrong in my opinion but most gamers are willing to live with enormous visual artifacts from partially completed frames to get that max fps and lowest input latency when things tough on the system.

So, to all those that mock high fps benchmarks, I challege you to post information on a recent 3D game, gfx card, system, and config that allows you to play with all the gfx features on (or those that are important to you) with vsync on using a 60 Hz display and only double buffering which locks at 60 fps solid without ever dipping below that.

That is when things have become fast enough for _that particular_ game.

Products like the NV30 and R300 help push the bar but are still not overkill. Take the above challenge and now turn on 16x multi-sampled FSAA (same as an SGI Onxy/IR), 8x anisotropic filtering (often more important than FSAA), 1600x1024 (the native resolution of my DFP), 128bit pixel depth (which NV30 can do before scan out), and include very complex vertex and fragment (pixel) programs. With all of that, turn Vsync on (as it should be) and have this entire combination run at 60 fps per second regardless of what is going on in the game at any and every given moment.

When we can do all of that, we are finished. :-)

The problem then becomes the content creators who continue to push the envolope. GeForce FX is launching with a demo that has Jurasic Park/Toy Story quality rendering tied to real-time dynamics and feature film quality animation. However, it is not quite to the level of Gollum in LOTR: Two Towers. Imagine Doom4 with 50 characters on the screen that all look like Yoda or Gollum. My point is that there is always room and applications for higher performance.

100 of these running around locked at 60 fps is the new goal: http://notendur.centrum.is/~czar/misc/gollum.jpg

DX 9

[Anonymous Brave Guy]

Hate to break it to you, but having just spent some days researching this, I concluded that there is nothing that the GeforceFX will support that the Radeon 9000 series won't. nVidia's web site may say differently, but that doesn't make it so.

The FX may do it faster (though this remains to be seen, of course) but it probably won't do it with better image quality. If anything, I'd say ATI cards have historically produced nicer output where there's any difference at all.

Hell, even the drivers for the Radeon 9700 are getting good reviews. I thought the season of miracles was a couple of weeks ago. ;-)