X-bit labs

Started by: lanky | Date 05/04/04

Comments: 45 | Last Comment:  08/25/06

[1-20 | 21-36]

1. yea biatch first post

2. So, what will we be looking at for the next big evolution from this card? Then again that will probably take 18-24 months.

3. I'd like to request Min and/or AVG FPS graphs to be included. I enjoy your reviews but some of us would like to make sure the card won't dip to low in some cases.

Also any overclocking info would be nice.

In fact, results we present are actually average FPS numbers. We are considering possibilities to include minimal FPS in our forthcoming reviews.

Both cards we used were very poor overclockers, so, once we got something that is better in terms of overclockability, watch out for a special review covering this.

4. This is the most bias review ever. It's simply anti-nvidia bullshit.

Whenever Nvidia wins by around 10fps you claim that it's a tie between the 2 cards but when they come within 3fps of each other you claim that the "RADEON X800 XT manages to strongly outperform the rivaling NVIDIA GeForce 6800 Ultra. "

This is simply bullshit as the 2 cards are within 3fps of each other.

We typically consider the results on the percentage basis. If the margin betwen 2 cards is less than 5%, even if 5% are actually 10 fps or 20 fps, we believe that the speed is nearly equal. Furthermore, we are also taking into account modes when NV40's trilinear optimisations are switched off.

We encourage our readers to draw their own concusions based on the numbers we deliver.

5. I'm going to kind of nod @ comment #4...

ATI is using 24 bit color downcoloring 32bit mode. Coming from the users who do digital photography on the side, that's not a limitation I'd think is acceptable anymore.

And the BIG thing is, back when you were touting the 9700/9800, you claimed that PS 2.0 was great because you were ready for the next big thing. Now PS 3.0 is in hardware, it makes programmers lives easier with even bigger effects and "ATI works just great with games you can play right now"?

This article is unusual for Xbit, and it is an article of convenience, catering to the ATI side of the fence while highlighting none of the sacrifices you have to accept in going to that side of the fence.

I like both, and I happily applaud good performance numbers, but at least highlight the serious compromises that ATI made to make these performance figures happen. I think your readers ought to be aware that ATI is banking on selling you a $500 card now and then another $500 card in a year to make up for all the things the NV card can do right now.

Its a poor financial choice and you are hiding that fact behind performance numbers, imo. That is NOT a service to your readers.

The Shader Model 3.0 is actually a superset of the Shader Model 2.0. In fact, NVIDIA's shaders 2.xx have been available on the GeForce FX for quite a while were pretty close to the SM 3.0.

We expect that the SM 3.0 to be mostly used for the purpose of performance optimisations in some areas where it's possible. Additionally, the SM 3.0 is likely to be used in DCC apps. That's where NV40 will strongly benefit.

While games may boast with SM 3.0 for the purpose of new effects as well, we do not consider the possibility very likely based on the fact the the additional caps of the GeForce FX-series were not widely utilised. However, we do not proclaim the final conclusions just yet.

Looks like the next big thing is the SM 4.0.

6. Good review again, gentlemen.

I think that both IHVs will be finding their feet with drivers/scheduling for these chips. For Nvidia it's obvious, but ATI has the same problems... These are the first "4 quads" ASICs. ATI has experience with 2 quads processing, but the scheduler in the reviewed driver set is having problems with load balancing of all 4 quads. The X800Pro with 3 quads is performing worse, being a non power of two design. It will be interesting to see what "DEC's compiler team" does for R420 performance...

P.S. To the above posters, "downcoloring" to 24bit is nonsense... & digital photography has nothing to do with this... We're talking about internal fp calcs, not external framebuffer bit depths which are 32bit integer... All calculations for R3x0/R420 are at 32bit fp. PS 2 is 24bit fp. Unless you use extreme shader length, tex coords, you'll not see any difference. Only if you run eg Mandelbrot in PS will the quantitative differences add up to qualitative deviations. Don't forget, Nvidia still need to use pp_fp16 to attain 24bit performance on ATI's architecture. NV40's fp16 tex ability could be very useful, though. I also think many don't understand just what PS3/VS3 brings to the table. Although loops, branches, predicates are useful for CPUs, just how much benefit a stream/SIMD/MIMD processor with >100 pipeline stages can gain is questionable. One pipeline stall for a dynamic branch & it's flush time...

NV40 is an excellent product, but it's cost Nvidia 220m trans, a hefty power budget & correspondingly low chip/wafer yield... ATI have a smaller part on low -k costing less power, which should also yield better... Guess who'll have better margins? That said, a cheap 6800GT would be hard to pass up, but then the X800Pro has better fillrate...

7. I find it odd that a free game like America's Army is not in the benchmarks...I am curious how it performs with ATI's latest products.

We believe that popular games and DirectX 9.0 games brings our readers a lot of useful information. While we do consider adding some new games into the roundup, we doubt it makes a lot of sense to include DX7, DX8 titles.

8. Nice review. I kind of had the feeling the next ATI card would outperform the Nvidia 6800 cards in most of today's applications. However, both cards have their advantages. I'm either getting the X800XT or 6800 Ultra, but don't know which one would be better to get. Both cards look fantastic. However, the drawback of the 6800 Ultra is the amount of power it requires, and the drawback of the X800XT is that it doesn't support SM 3.0, regardless the fact that no application has SM 3.0 implemented into it.

BUT, how will the X800XT perform in the future? :\

9. if you know about ATI's questionable "optimisation" regarding anisotropic filtering (where only the first texture stage gets trillinear filtering, and the rest get bilinear filtering), how come you don't know about rtool (http://3dcenter.org/downloads/rtool.php), which can be used to enable correct trilinear filtering for all texture stages on ATI chips ? even worse, you claim ATI's optimization cannot be disabled, and you perform all tests using this low-quality bilinear anisotropic filtering method. Sure, other reviewers did exactly the same, but they didn't seem to have a clue about ATI's texture stage bilinear filtering optimizations, while you knew and did nothing to change that.
I'm dissapointed because I expected a review which would show the performance of these high-performance cards in a way which suits them best: and that is using the highest quality filtering provided by the chips, disabling all driver optimizations for the chips (just like you did with the Geforce 6800). Now I'll never know what the X800XT's performance using the correct trilinear anisotropic filtering method would be like...

Thanks for the information about the app!
We will try to use the utility you pointed out in our future issues.

10. "The top-of-the-line $499 RADEON X800 XT appeared to be faster compared to its main competitor – the GeForce 6800 Ultra – in plethora of applications where it was pretty natual to expect – the games that broadly use complex geometry and loads of math-intensive pixel shaders"

Excuse me, but this is totally untrue.
As it shown from the Benchmarks of F1 Challenge, ATI's cards are using much more CPU power than Nvidia's cards. So, when a game needs a lot of CPU power, e.g. for AI, ATI's cards are not good. You may check setups with lower CPU speed to verify this. Also, I think it is very bad for ATI not to include Shader Models 3, no matter the present situation, and if they want to stay alive in the mainstream market, they shall include them in the very near future.

We are focused on delivering the results of graphics cards not affected by CPU speed. We honestly doubt that anyone would by a $500 card for a computer with low-end or mainstream microprocessor.

F1 Challenge delivers strange results. If you like the game, looks like you need a GeForce. But the whole picture provided by the 25 benchmarks we used appears to contradict what the F1 Challenge showcases. However, there are loads more games coming, so, the final word is not said yet.

11. Man-
All you guys pissing about SM3.0

It's a marketing gimmick. By the time these SM3.0 games are readily available-the next wave of cards will be available. It seems that you guys are forgetting one very good point about this review. NV40 is a great card and also is able to play every game lightning quick. But If I am spending 500 on a card, it better be able to play fastass at 16*12 with all the eyecandy on, and sorry-x800 can do it-yu get jipped with the nv40. No excuses-the Radeon is exhibiting raw power. I'm pretty sure there are plenty of people who agree with me.
Cheers

12. Bias review ever.
When NVIDIA win comment so short and doesnt reflect to it.
When ATI win comment so long and it makes as it is so superior to NVIDIA.
Overall Unreliable comment

13. Anyway i forgot to add something, i am not a fan of NVIDIA and i admit ATI better but when you making a review then you suppose to be neutral, unlike this review which can bee seen clearly the reviewer seen to like ATI more

14. To 9. is entirely false...

In fact, selection of of AA has no "rules" for tex filtering at all... There's no reason point sampling can't be used... ;) Additionally, stage 1 tri 2-7 bi filtering only occurs when forced through control panel. If the app/game requests AA/tri it is applied correctly to all 1-7, etc.

This contrasts nicely with Nvidia, for which there is no way you can set tri on subsequent tex stages with AA enabled... Everything is now "brilinear" on every stage regardless of control panel/app/game requests for tri... So the comparisons were entirely correct. Also be aware that with Nvidia's latest ForceWare drivers, there's a "bug" where if you disable trilinear optimizations in the control panel, they are still active... Convenient bug, no...?

Nvidia have no choice but to copy ATI's AA method because that's where they perceived they lost out to ATI with R3x0 series... (I do not like their selection of 25 deg angle dependence, though. 22.5 is more "optimal"). Unfortunately 6800 can no longer perform "correct" trilinear as per 5900 series.

BTW. RTool only does what normal games/apps do when requesting AA/tri.

Sorry, that was mistake. Should read "Unfortunately 6800 can no longer perform "correct" AA as per 5900 series."

NV40 is an excellent chip with no need for crappy IQ reducing. In fact I am very impressed by its pixel pipeline ALU power...

Many mistakes, today. Everywhere I say AA, I mean AF. Maybe mods can edit. Sorry.

15. ssive, but do we should buying graphic cards to play another gamr? I have ATI-9700 and i think it's OK

16. would you please put some screenshots on the article

17. hi..
In all the other benchmarksfrom other sites the comparisons b/w the 2 cards looked even. but here it looks so much in favour of ATI
heard Nvidia did come put with some new driver and also relaead extreme series. maybe tht will help

18. i paid 22000 Rs when ti4400 came brand new.....god now my card is not in the same playin field .is it
Guess we can never play catch up

19. "...we believe that the RADEON X800 XT clearly packs the punch over the competitor in terms of performance in applications that use shaders 2.0/2.x and are available today."


Which will perfom better in the long run though, Radeon X800 XT or GeForce 6800 ultra?

20. In firestarter you said "But once the eye-candy features are switched on, the RADEON X800 XT manages to strongly outperform the rivaling NVIDIA GeForce 6800 Ultra".......3 points is strongly outperform?

[1-20 | 21-36]