GeForce FX reviews

[Slougi]

http://www.hardocp.com/article.html?art=NDIx
http://www.tecchannel.de/hardware/1109/index.html

Seems like nVidia blew this one, pretty badly too.

 

[Trotterwatch]

I wouldn't read too much into these previews at the moment - the drivers for example are still Beta, and Nvidia have in the past had a habit of releasing Hardware with drivers that work but work slower than they should.

Still these articles make interesting reading.

 

[Slougi]

I wouldn't read too much into these previews at the moment - the drivers for example are still Beta, and Nvidia have in the past had a habit of releasing Hardware with drivers that work but work slower than they should.

Still these articles make interesting reading.

It is not just speed nowadays with cards of this pricerange, but also quality AA and AF. Look at the AA comparison on Anandtech, I think you will agree that the Radeon 9700 has much higher IQ in this case. AF is tied imho, both look about the same.
Also listen to the mp3's on tom's hardware. It sounds like a damn jet-plane taking off. lain: It exhausts air at 120F in DOS mode. lain: And most likely will cost more than the Radeon 9700 Pro. Imho, a complete disaster of a card. Also had some corruption that did not look like driver problems, more like overheating memory.

 

[Trotterwatch]

Yeah, Beta Drivers can't excuse this cards lacklustre performance (in comparison to what was expected).

I'll check out the MP3 of the exhaust sound now... I've heard it is bad though from many sources. I think Nvidia have made a mistake using DDR 2 with a 128bit bus rather than the generally cooler DDR 1 using the 256 bit bus (like ATI have done).

Will be interesting to see which card performs best in DX9 games though.

from Toms Hardware:

When we pointed out the immense noise generated by the card, NVIDIA responded by assigning this to the prototype status of the card. In the final design, the noise level is supposed to be reduced considerably.

Lets hope and pray that Nvidia are telling the truth here!

 

[Slougi]

Yeah, Beta Drivers can't excuse this cards lacklustre performance (in comparison to what was expected).

Too true.

I'll check out the MP3 of the exhaust sound now... I've heard it is bad though from many sources. I think Nvidia have made a mistake using DDR 2 with a 128bit bus rather than the generally cooler DDR 1 using the 256 bit bus (like ATI have done).

This is one of their mistakes, the other was to rely on TSMC's low k 0.13micron process, which was not ready in time, and still isn't. They had to use the higher k process (more electric resistance = heat), which lead to the huge fan we see now.

Will be interesting to see which card performs best in DX9 games though.

This will actually be quite meaningless for the next year or two
Except for the new 3DMark, which is a benchmark, not game.

 

[Trotterwatch]

WOW, just listened to the MP3 - it sounds like someone is drilling a hole right through the case. It also reminds me of going to the Dentists.

Nvidia need to cut that by at least half and preferably more than 3 quarters - at the moment that sound is absoloutely ridiculous.

 

[Slougi]

Yep, jet plane taking off
Sound like 3dfx? Late, slow, huge power requirements.

 

[Trotterwatch]

Yep, and the FX moniker at the end now seems like fate. I hope this card doesn't bomb, as to be honest we need all the competition we can get in the graphics card market. At the moment there are 2 competitors - if this bombs then, Nvidia will have a tough time coping... they have invested A LOT of money in this.

 

[Slougi]

Competition is a Good Thing™, true. However, if this bombs it will put Ati and Nvidia on equal footing. Again , a Good Thing™. NV35 and R350 should be on par. I just hope TSMC gets their act together.

 

[pj64er]

Yep, jet plane taking off
Sound like 3dfx? Late, slow, huge power requirements.

dont be so harsh. This has been the first time nVidia has been late. have a little faith. concerning the huge power consumption, doesnt the R9700Pro require its own power source as well?

 

[Slougi]

dont be so harsh. This has been the first time nVidia has been late. have a little faith. concerning the huge power consumption, doesnt the R9700Pro require its own power source as well?

So that is a reason not to critizise them?
R9700 does require an extra power connector, but the geforce FX draws around 70W, while the r9700 pro draws 49. The geforce is right up there with a p4 2.8Ghz in terms of heat production(!) (!) Seriously, listen to the mp3 on tom's hardware. Look at the corruption on hardocp's site. Look at AA/AF quality. Look at price. Look at performance with AA/AF on. Look at the eta of the gffx. I think the r9700 is the clear winner here.

 

[ra5555]

great news for me, knowing that the geforce FX won't be 1000000000000 times faster than my card and forcing me to get a new one

 

[AlphaWolf]

In all honesty, I have found the geforce ti4200 and 4400's to be the absolute best cards for the money, and I don't see any need for anything better for at least a year or so.

I am still waiting for a card to finaly do hardware FSAA before I ever get anything beyond that.

 

[Tagrineth]

dont be so harsh. This has been the first time nVidia has been late. have a little faith. concerning the huge power consumption, doesnt the R9700Pro require its own power source as well?

Voodoo3 was early.

And AlphaWolf: ATi Radeon 9500/9700 series.

Technically the only cards with "Software" hacked AA were GeForce256, GeForce2, Kyro/II, Radeon R6, and Radeon 8500... all others do it in "hardware"... but it's sooooo subjective really

In any case, everyone, read the GFFX review at AnandTech, they really go in-depth with the quality comparison, and GFFX's Performance AF is really embarrassingly poor-quality compared to even the ATi R300's Performance AF.

edit: Fixed a board code...

 

[AlphaWolf]

Why are every single one of these cards so slow at FSAA then? When you enable FSAA in later games, like ut2k3, it kills the shit out of the frame rate.

 

[Trotterwatch]

Why are every single one of these cards so slow at FSAA then? When you enable FSAA in later games, like ut2k3, it kills the shit out of the frame rate.

Isn't that like saying - why does UT run slower when you have 20 bots. The reason for that is the CPU, agreed? The CPU is hardware though is it not.

The same goes for the graphics card, you enable FSAA it has to do more work, it then slows down. It'd do this regardless of whether it uses a hardware implentation or a software one.

Didn't the Voodoo 5 do FSAA in Hardware? If it did, then why did that slow down when it was enabled.

 

[Tagrineth]

Why are every single one of these cards so slow at FSAA then? When you enable FSAA in later games, like ut2k3, it kills the shit out of the frame rate.

It slows down because the reserved frame buffer has to be much, much larger to accomodate larger sample sizes. Add to that the fact that, say, with 4x AA, you're getting around 4x (peak) more frame/Z traffic across the bus.

The R300 and NV30 do a great job of alleviating this stress with colour and Z compression, but it isn't perfect, and quite simply can't be. Free FSAA isn't possible without deliberately reducing your 'straight rendering' frame rate, like some old cards that deliberately sabotaged 16-bit performance to make 32-bit frame rates look better by comparison.

And for reference, the ATi R300 series (minus the 9500 non-Pro) takes a ~30-40% performance hit for 6x FSAA. That's a far cry from Voodoo5 and such losing 50% for 2x.

Simply put, FSAA is very bandwidth-intensive. The only thing that can be done about it is try to be 'intelligent' about it (Matrox FAA) but even that has many hang-ups and such.

 

[AlphaWolf]

I know jack shit about video cards beyond their basic operation, but imo if what is currently done for FSAA is hardware, its a piss poor implimentation; as in it's more of a marketing scheme than anything. I don't want FSAA so that I can just tell my friends "yeah, this card does FSAA (*mumble*but I never use it because its too fucking slow*mumble*)". I want it to actualy be useful.

It'll probably be more than a year before I ever need a new graphics card, I just hope by that time they will get the speed issues worked out with FSAA.

 

[Tagrineth]

I know jack shit about video cards beyond their basic operation, but imo if what is currently done for FSAA is hardware, its a piss poor implimentation; as in it's more of a marketing scheme than anything. I don't want FSAA so that I can just tell my friends "yeah, this card does FSAA (*mumble*but I never use it because its too fucking slow*mumble*)". I want it to actualy be useful.

It'll probably be more than a year before I ever need a new graphics card, I just hope by that time they will get the speed issues worked out with FSAA.

Dude, "Hardware" does not mean "without a performance hit"!

FSAA implementations have refined like crazy recently, as I said 4x on a Voodoo5, Radeon 8500, or GeForce2 results in losing 75% of your performance, whereas on a Radeon 9500/9700 it only takes of ~30%.

The math and RAM requirements are VERY HIGH, there is NO GETTING AROUND IT, except by attempting an "intelligent" algorithm like Matrox, but that doesn't always work very well.

My Radeon 9500 Pro can run 4x FSAA at pretty damn near any resolution; I was playing Serious Sam the other day with the settings cranked to all hell with 6x AA at 1024x768x32, at 40FPS (kinda CPU limited). And that's with Vertical Sync ON.

They simply can't get rid of the speed "issues" with FSAA altogether. It can ONLY be done for free by Deferred Renderers (PowerVR), and yet even PowerVR cores are restricted by fill-rate (maths capacity).

 

[Slougi]

Dude, "Hardware" does not mean "without a performance hit"!

FSAA implementations have refined like crazy recently, as I said 4x on a Voodoo5, Radeon 8500, or GeForce2 results in losing 75% of your performance, whereas on a Radeon 9500/9700 it only takes of ~30%.

The math and RAM requirements are VERY HIGH, there is NO GETTING AROUND IT, except by attempting an "intelligent" algorithm like Matrox, but that doesn't always work very well.

My Radeon 9500 Pro can run 4x FSAA at pretty damn near any resolution; I was playing Serious Sam the other day with the settings cranked to all hell with 6x AA at 1024x768x32, at 40FPS (kinda CPU limited). And that's with Vertical Sync ON.

They simply can't get rid of the speed "issues" with FSAA altogether. It can ONLY be done for free by Deferred Renderers (PowerVR), and yet even PowerVR cores are restricted by fill-rate (maths capacity).

True true. There are still some improvements that can be done, especially with regards to z-buffer compression and access order. Just hope some FAA-like implementation will take off, current MSAA and SSAA algorythms won't really cut it imho, in the long run. Deferred renderers will remove bandwidth restrictions, but computational overhead is still large.
How the AA samples are arranged have a big visual change as well. Compare GFFX 8x AA to R9700 6x AA and you'll see that Ati's 6x AA generally looks much superior. So there is still some tweaking to do, both for speed and quality.