Tuesday, a friend lituanien (let us stop always showing Russian!) us forwarded some figures obtained on Ultra GeForce FX 5900 as its impressions vis-a-vis the some briefs tests which it had been able to make. We did not publish these figures upon the departure bus they appeared to us "too high". In fact, they were correct but carried out with the drivers Detonator 50.xx. This person could return us the same results but with less advanced drivers, certainly close to those which will be used in the tests which will be published the next week. This thus means that the first tests published will not be carried out with the new drivers which would name Detonator FX.
We added there the score of our Radeon 9800 Pro of test benchée on the same platform (nForce 2), with the same CPU (Athlon XP 3000+). Not having GeForce FX 5800 pennies the hand, we picked in the data base of Futuremark for this chart.
This being said, here figures in question:
3dmark03 Benchmark Graph
If the increase in the performances is only light (but well presents) with "traditional" drivers, it strongly increases with the 50.xx! However, it could be that the scores are not comparable with those of Ultra GeForce FX 5800 which cheating always in 3d Mark 2003. In other words, although the difference between the 2 charts of NVIDIA is thin with the same drivers, it would seem that the 5900, it, cheating not.
As you know it, GeForce FX 5200, 5600 and 5800 have a large handicap during calculations in Floating Point like the 64 bits (4 X 16 bits) or the 128 bits (4 X 32 bits). NVIDIA doubtless "looked after" GeForce FX 5900 in this field. Besides this joined the comments which we received of which here an extract translated into French:
"... of pretty figures! Ca made a long time on behalf of material NVIDIA! The images seem well, more problem of quality to the first glance. Afflicted not to be able to give you figures as detailed as the other time...... but the images seem correct this time. Truth floating not at a decent speed. You know what I think of that, and you divisions my vision of the things I think (?), it is for me a first chart of NVIDIA which will be able to post the graphics conceived in the direction of DX9... "
We agree of course completely with this person as for Floating Point which is according to us one of the major interests of the DirectX9 material. Although putting it strongly ahead, NVIDIA does not hesitate to decontaminate it in its drivers over the remainder of range FX to gain in performances.
Our fellow-member 3DChips-Fr published this evening some details concerning GeForce FX 5900:
Bus report 256 bits
450 MHz for the core
425 MHz for the 256 Mo of memory
UltraShadow Technology
Intellisample HTC
CineFX 2.0
Power in Floating not doubled
More dry hair
CineFX 2.0 seems to be only CineFX with UltraShadow. Technology which makes it possible to accelerate returned shadow volumes. But it is difficult to say if it is of an innovation hardware or a technology software intended for all range FX, but marketée only with the FX 5900 to increase the impact of the advertisement.
In short, GeForce FX 5900 seems to hold its promises! What somewhat changes remainder of the range FX which, it should be acknowledged, had disappointed overall.