Wich card is the king of the hill (nv40 or R420)
- Thread starter Rican
- Start date
- Status
The advantages between the NV40 and R420 as I see them.
R420
1. Less power draw and subsequently less heat, one molex
2. Smaller PCB
3. 3Dc
4. More efficient AA/AF
5. Temporal AA
6. Better DX9 performance
NV40
1. shader model 3.0
2. Better OGL performance
3. Full hardware accelerated encoding/decoding
Considering the R420 outnumbers the NV40 in advantages, I consider it to be the winner. And I haven't even mentioned the fact that ATI plans on introducing completely rewritten OGL code and even further optimized shader compilers in a future driver release. Shader model 3.0 offers very little besides performance increases as well. Which only leaves number 3.
Anything I'm missing?
R420
1. Less power draw and subsequently less heat, one molex
2. Smaller PCB
3. 3Dc
4. More efficient AA/AF
5. Temporal AA
6. Better DX9 performance
NV40
1. shader model 3.0
2. Better OGL performance
3. Full hardware accelerated encoding/decoding
Considering the R420 outnumbers the NV40 in advantages, I consider it to be the winner. And I haven't even mentioned the fact that ATI plans on introducing completely rewritten OGL code and even further optimized shader compilers in a future driver release. Shader model 3.0 offers very little besides performance increases as well. Which only leaves number 3.
Anything I'm missing?
jimmyjames123
Regular
Yes, you are.
The 6800 GT and each 6 series card below it will all be single slot/single molex. The 6800 GT is said to have power requirements no greater than 9800XT/5950 Ultra.
Also, it is not really accurate to say that the R420 has more efficient AA. Actually, it seems that the NV40 may have slightly more efficient AA, while the R420 may have slightly more efficient AF, generally. In fact, some reviewers think that NV's AF algorithm is slightly cleaner/sharper than ATI's AF algorithm, even though they are both angle dependent.
You also missed a checkpoint for the NV40, being Ultrashadow II and Doom 3 style games that are shadow intensive. This should be a very strong genre for the NV4x cards.
Keep in mind that NV's drivers are also somewhat raw, and not as mature as ATI's drivers at the moment. NV is moving to a very different architecture this time around with the 6 series, and I imagine it will take some time for them to learn how to effectively optimize for the 6 series while deemphasizing optimizations that were made for the FX series.
jimmyjames123 said:
There's something about this that troubles me. If the 6800 GT is essentially just a downclocked Ultra then wouldn't that mean it would suffer from the same power requirements as well? If this is the case how can one expect to overclock it to 6800U levels and above without the aid of another molex? I believe eVGA or BFG ran 3DMark03 for hours on one molex and they claimed there was some artifacting and visual discrepencies.
I disagree completely here. From what I have seen in the various benchmarks, performance levels drop farther on the NV40 when 16x AF is enabled then on the X800. You could debate this, however it's a fact that the NV40's performance drops ten fold when 6x or 8x AA is enabled. And they both look about the same image quality wise.
I did forget about Ultrashadow.
jimmyjames123
Regular
No. The 6800 Ultra and Ultra variants require the second molex connector because they need to be able to handle core clock speeds well in excess of 400Mhz core. See the Techreport review on how power requirements go up significantly from 400Mhz core to 450Mhz core on the 6800 cards, while going up very little from 350Mhz core to 400Mhz core.
You didn't carefully read what I said. I said that the X800 tends to have a more efficient AF algorithm. The 6800 tends to have a more efficient AA algorithm, up to 4xAA of course. NV's 8xAA is not comparable because it includes a mix of SS + MS. ATI's 6xAA is naturally an advantage as it is something that NV does not offer. You also forgot about the comment regarding how NV's AF algorithm is even now considered to be slightly clearer/sharper than ATI's.
Yes AA up to 4* is more efficient on 6800. From B3D @ 16*12 +4*AA
Code:
UT2k4:
6800: -0.1%
X800: -0.4%
Tomb Raider: Angel of Darkness
6800:-29.5%
X800:-30.3%
Call of Duty
6800:-4.5%
X800:-8.1%
Serious Sam: Second Encounter
6800:-4.5%
X800:-12.8%
RightMark - Lighting (Phong) - PS2.0
6800:-0.2%
X800:-4.4%
3DMark03 GT2, 1600x1200 FSAA
6800:-44.4%
X800:-60.0%jimmyjames123 said:
"eVGA or BFG ran 3DMark03 for hours on one molex and they claimed there was some artifacting and visual discrepencies." This test was done at the request of nvidia after someone asked what would happen if only one molex was used on the Ultra. Obviously at stock Ultra speeds there are some problems in the rendering in this situation. Thus the GT is bound to suffer the same problems at a core speed of 400 MHz and be unusable at higher speeds. For this reason I don't see it being a very good card for overclockers.
Yes the X800 has a more efficient AF algorithm, that is what I said in the first post. NV might have a slight advantage in 2x and 4x AA but it's just that, slight. And at 6x and 8x there's no comparison. The fact of the matter is the XT will outperform the 6800U by a factor of two in almost every test when both cards are run at 1600x1200, 8x AA and 16x AF. The 6800U only outperforms the XT at low resolutions with no AA or AF, and even then it's by a marginal percentage. NV's AF algorithm being sharper then ATI's is debatable.
Sabastian said:
1. I am completely agnostic when it comes to ATI/Nvidia (which is why I am getting more and more annoyed by the blatant ATI fanboism of some people around here, it really starts to impact the usefulness of this forum)
2. I have no clue which one is supposed to be better or worse here. They don't look identical but I am at a loss judging which one is supposed to be better.
Maybe you could point out what exactly makes NV40's AA worse, other than the fact that the line looks pretty solid on the 6800 image while it looks like it has gaps on the X800Pro image.
- Status
Similar threads
