[H]ardocp Review for the Geforce FX

[Mariner]

Bizarre benchmarking style over at Hexus. They benchmark:

1) without AA or Anisotropic
2) without AA but with Anisotropic
3) with AA but without Anisotropic

Where are the benchmarks with both AA and Anisotropic filtering!?!

On one of the later pages (UT2003 flyby), they get what looks like a bit of an anomalous result for the AA on the R300. Apparently, they couldn't check this, though, because they no longer had the GFFX. I can't quite understand the logic in this!

Overall, this seems to be the most "NVidia-centric" of the reviews I've seen so far.

I've not read the Tomshardware review - can't be bothered. Thought the HardOCP one was pretty good.

 

[Doomtrooper]

There was some rumours around the net about Nvidias PR team getting creative..So far I've seen the graphing program handed out by NV was only used by Hexus, and what settings to use to show fillrate advantages

 

[Bambers]

I think somethings obviously wrong with the FSAA on the FX from the screenshots. The drops in fps are all fine but the 2x and quincux aren't doing anything in the UT screenshots (are the screenshots just not showing it or is it really not doing anything? someone with an FX want to tell us this? )

Look at the bridge in the NFSHP2 shots. Here 2x and quincux show up as expected as does 4x and even 4xS. 6xS shows 2 distinct middle graduations on the vertical edges (expected) but only a single one on the horizontal edges. 8x is similarly affected, 3 middle shades on verticles as expected but still only a single one on horiztonal edges.

Getting a gf3 or 4 and using the registry to set the 6xS and 8xS modes on those cards will show much better IQ than those shots from the FX.

:|

 

[Tahir2]

I posted this on the hexus forums.. I hope someone listens over there.

quote:
--------------------------------------------------------------------------------
The benchmarks were only run once. Unfortunately due to the extreme time constraints available to us, and in an effort to bring you this very early, exclusive UK look, the benchmarks were only run once; NVIDIA have however assured us that they will be able to make available a card for extended testing in the coming days. As it is we believe that HEXUS has undertaken the widest, 'hands-on' platform testing yet published.
--------------------------------------------------------------------------------




quote:
--------------------------------------------------------------------------------
Anisotropic filtering was tested at a resolution of 1024x768 only.
--------------------------------------------------------------------------------



What Anisotropic settings did you use? Performance? Quality? Performance-Balanced? Aggressive?


quote:
--------------------------------------------------------------------------------
Across screen resolutions tested the GF FX demonstrated a 4.5% improvement beating the Radeon in every test except when using anisotropic filtering.
--------------------------------------------------------------------------------



4.5% is not a lot, and adding that with AF enabled the GFFX is slower, that is really not a good sign. We all know how AF helps with IQ (I'll come back to this later as well).


quote:
--------------------------------------------------------------------------------
In the Bot Match benchmark which is the closest simulation to actually playing UT2003 things did not look so good for the GF FX. The Radeon was quicker at 1024x768 and only outpaced by the GF FX at 1600x1200. The GF FX excelled when playing UT at 1600x1200 with 4xFSAA which is exactly how the game should be played.
--------------------------------------------------------------------------------



That is great that the GFFX beat it at a high resolution with 4XAA.


quote:
--------------------------------------------------------------------------------
Although Q3 is somewhat older and less taxing than some of the other tests it is always good to see some large frame rates on display and on a more scientific note Q3 does test data throughput and bandwidth very well. The GF FX performs very well except in the tests with 4xFSAA enabled.
--------------------------------------------------------------------------------



I thought this is how (4XAA and high resolutions) UT2003 should be played .. but obviously not Quake3 OK that was a bit of a cheapshot but shows fair weather reviewing.


quote:
--------------------------------------------------------------------------------
Across screen resolutions tested the GF FX demonstrated a 1.2% improvement beating the Radeon in most of the tests, but only just.
--------------------------------------------------------------------------------



Now it is only 1.2% on the AMD machine in a synthetic benchmark, but everyone still uses it for bragging rights.

In both Intel and AMD setups the GFFX stomps the Radeon 9700 Pro.. it is a shame you can't actually play a game like it on the market now or even in the near future.


quote:
--------------------------------------------------------------------------------
Unfortunately until we can have more time with the GF FX we can not re-run the test.
--------------------------------------------------------------------------------



The test seemed to take off 4XAA but even then the resolution was the same and suddenly the Radeon 9700 Pro is performing 4fps faster than before.. This is the reason why you NEED TO re-run tests several times!


quote:
--------------------------------------------------------------------------------
In the Bot Match benchmark which is the closest simulation to actually playing UT2003 things did not look so good for the GF FX. The Radeon was quicker at 1024x768 and only outpaced by the GF FX at 1600x1200. The GF FX excelled when playing UT at 1600x1200 with 4xFSAA which is exactly how the game should be played.
--------------------------------------------------------------------------------



Same conclusion as before this time for the AMD setup - however 7 fps is not a lot. Again UT 2003 is different to Quake 3 then?


quote:
--------------------------------------------------------------------------------
For this test we used the challenging ‘Valley of the Jaguar’ demo. Only time constraints forced us to limit the number of benchmarks used. The results show that again the GF FX out performs the Radeon 9700 Pro.
--------------------------------------------------------------------------------



The difference is 6.9% .. not a lot is it?

The difference in performance between the Radeon 9700 Pro and the GFFX is at times in favour of either card, with AA AND AF enabled the Radeon 9700 Pro has been shown to BEAT the GFFX on several other reviews.


quote:
--------------------------------------------------------------------------------
Some people may mistakenly interpret this very early and UK exclusive first 'hands on' look at the NVIDIA GeForceFX as a definitive guide as to buy or not to buy, and may well be thinking that the GeForce FX is not living up to the hype and over inflated expectation.
--------------------------------------------------------------------------------



I'm glad you said that. Perhaps the first on hands look should be called a preview of pre-release performance as you do not really touch Image Quality at all.

And going back to Image Quality for a second lets see what we have here:


quote:
--------------------------------------------------------------------------------
There can be no question that the GeForce FX out performs the Radeon and produces higher image quality as well.
--------------------------------------------------------------------------------



This is where IMO your review from being mediocre turns into a farce.... and disappointing to say the least.. I love this site normally. I don't think I need to comment why and if you disagree well I have your own words to prove my point.

Slightly noisy.. how about 60dB noisy - if you can measure the noise in your next article that would be great.

I look forward to reading your more indepth review but I have little hope that it will make up for this mess. Running a load of benchmarks does not make a good article. Even Brent from hardocp.com (only cos of Brent's talent) did a good job as did Anandtech.. How ironic.

 

[nggalai]

It was highly rushed, we had 2 days to review the card before the NDA lifted.

Believe me we have much more planned for it, I want to get in cpu scaling, cross platform testing, see how much of a performance hit it takes at AA only and then AF only, and more IQ and then just pure gaming with this card to really get a feel for it...

Still, a very nice review. Well done, Brent.

You can still post a follow-up later on, I suppose, with some hands-on gaming experience under your belt . . .

ta,
-Sascha.rb

 

[Doomtrooper]

I agree with nggalai, Brent has brought a good injection of common sense into [H] in the reviews, little less PR spin..more screenshots, benchmarks that are important.

B3D vs. [H]

Lets get ready to rumble

 

[nelg]

Before I begin let me say for the most part I feel that the fourms on this site are the best place for open minded and serious disscussion on 3D related topics. Thus , can anyone here awnswer this question ? Considering how hot this card runs would it not be possible to water cool it because the the water pressure in such a system would pose such a potenial leaking problem. And if so do you think that this is the reason that Nvidia went the FX flow route instead ?

Thanks

 

[MuFu]

I doubt there would be any problems watercooling it. Even a modest H20 rig would be more effective at cooling the ASIC than the dustbuster (massive reduction in noise aside).

If they did watercool it it'd either have to come with it's own self contained watercooling system or be restricted in terms of useability to those that already have H20 rigs in their PC. Either scenario is utterly implausible, IMO, although I don't discount the possibility of a 3rd party offering such a solution at some stage, be it with NV30 or the forthcoming R350.

MuFu.

 

[Doomtrooper]

No a good water cooling solution would be fine, it would have to be some form of custom job..two water blocks..one for the front and back.
The cost involved adding water cooling with a pump would push this card over $1000.00 where we live...and probably $500 US.
Most kits runs $150 for a good one.

 

[nelg]

Thanks for the replies. I recall hearing that one of the ATI board partners makes a water cooled 9700 card for silent operation. I do not know the cost but was wondering if it can be as an effective cooling solution then why didn't Nvida goe this route, considering the negative response to the noise levels of the FX ?

 

[Althornin]

Thanks for the replies. I recall hearing that one of the ATI board partners makes a water cooled 9700 card for silent operation. I do not know the cost but was wondering if it can be as an effective cooling solution then why didn't Nvida goe this route, considering the negative response to the noise levels of the FX ?

ITs not water cooled, AFAIK.
It uses the massive passive Zalman heat pipe.

 

[MuFu]

I think he is refering to talk of Hercules and/or other AIBs producing a watercooled R350 solution. Whether this means "watercooled card package" or "watercooling-ready" I don't know.

MuFu.

 

[Doomtrooper]

Hercules are developing a high end R350 product. This will be based purely at the high end overclockers, the product will be water cooled and be a complete solution - they are in talks with a few manufacturers which we have talked with and the brief is for a "performance block" over a block which is based around looks.

The 9700 Mod was a custom mod at deviant PC...expensive.

There is risk with water projects, if the end user does not assemble it corectly and we all know what happens when water meets motherboards., then they may try to claim damages etc towards the video card company..

 

[Doomtrooper]

Installed..looks like inside the USS Enterprise panel

 

[OpenGL guy]

Installed..looks like inside the USS Enterprise panel

Bah, they didn't even water cool the sound card

 

[RussSchultz]

Ha. You laugh. My sb-live gets pretty warm to the touch.

 

[WaltC]

Another great review from Brent.


Hmm... that's 50C when idling! WOWSER.

Amazing to think that ATi could release a "dustbuster-equipped" 9700 Pro right now that would take the performance crown back immediately, possibly by a considerable margin. That said, I really hope they do not use the 5800 Ultra as an excuse to launch R350 with a similar contraption. People have been saying "they have to in order to get to 400MHz!" but I really don't think that's true.

Also interesting that the fastest card available right now is probably not the GeForceFX 5800 Ultra, but one of the AIB 9700 Pro boards. Hell... there's one with passive cooling that still costs less than FX.

MuFu.

I really think they are concentrating on better yields and tweaking the pipes and maybe doing some general housekeeping with respect to R300 (afterthought kind of things.) I hope they have *better* cooling for R350, but I doubt they'll have any appreciably bigger cooling for it (could do some work with the shim, etc., to get a better/closer merge between the hs and the gpu. That type of thing. Basically just a tweak of the existing cooling system.)

As far as it goes I think I could simply overclock my own 9700P and do far better than GF FX and not have to listen to a hair-dryer at the same time. Whole thing's pretty amazing actually, and serves to illustrate that ATI was at least ~1 year ahead of nVidia when they shipped the R300 9700P--quite a feat in this market. Don't see this kind of leapfrog often.

 

[Rann]

Long time lurker, first time poster here.

Thanks for the great review. I just have a comment/question regarding the maximum IQ comparison at the end of the review.

If you look at the R300 shot, you can see very clearly the health package in the middle. If you now look at the NV30 shot, you can only make out a blue and white mess at that same spot. So is the AA/AF killing the texture or is it a rendering problem?

Overall I'm very dissappointed with NV's IQ... 8X AA with noticeable jaggies and very poor lighting compared to ATi's (in NV's shot top rail is pitch black - very unnatural looking).

Rann

 

[Brent]

If you look at the R300 shot, you can see very clearly the health package in the middle. If you now look at the NV30 shot, you can only make out a blue and white mess at that same spot. So is the AA/AF killing the texture or is it a rendering problem?
Rann

I didn't even notice that

I'll fire it up again and take a look at it...

 

[Guest]

Toms Hardware review is a complete joke.

AMEN !!