We have made significant changes to the rating calculation method for several reasons. Firstly, there was a number of math errors, so we decided to use geometrical average for all test results. Secondly, the quality value had little to no influence on the results. This was unacceptable due to newer games, in which AA or AF are very important. Thus we made AA and AF influence floating. The geometrical average performance is divided by 60 FPS (as a standard of average minimal sufficient performance) to obtain a coefficient of suitable performance. It's multiplied by a pre-fixed AA or AF coefficient that depends on certain card specs. This gives us a coefficient of card's AA or AF potential, that we finally take into consideration. Otherwise top-end cards would be underestimated, because when you spend $400-500, you get not only the FPS in existing games, but some potential for some higher quality in the future. Otherwise $10 cards made in 2002 could get high marks due to their low price).

Usability ratings are calculated by the following formula:

Kuse = ((K-no-AAA ) ^ (1/53))^ 0.6 * ((Poss ^ 0.4)) / (Price)) * 15000

where:

K-no-AAA - considers performance without AA+aniso.

K-no-AAA = (C10xp * C12xp * C16xp) *
      (HL110xp * HL112xp * HL116xpxp * HL120xpxp) *
      (SC10xp * SC12xp * SC16xp * SC20xp) *
      (CD10xp * CD12xp * CD16xp) *
      (FC110xp * FC112xp * FC116xp * FC120xp) *
      (D310xp * D312xp * D316xp) *
      (G110xp * G112xp * G116xp * G120xp) *
      (G210xp * G212xp * G216xp * G220xp) *
      (G310xp * G312xp * G316xp * G320xp) *
      (F10xp * F12xp * F16xp * F20xp) *
      (CR110xp * CR112xp * CR116xp) *
      (G1610xp * G1612xp * G1616xp * G1620xp) *
      (G2610xp * G2612xp * G2616xp * G2620xp) *
      (G3610xp * G3612xp * G3616xp * G3620xp) *
      (G4610xp * G4612xp * G4616xp * G4620xp)

K-with-AAA - considers performance with AA+aniso. Calculated using the same formula considering AA+aniso performance values.

The similar calculation (but for 1600x1200 resolution only) indicates the prospect of a videocard. The formula contains the same indexes as those used for usability rating calculations, but accelerator's price isn't considered. So, this rating indicates the capabilities of the current 3D cards in the current applications and future games.

You can also compare two ratings to see the correspondence of product prices to their capabilities and features.

According to the same scheme, but without 3DMark05/06 results, we calculated K-no-AAA-only-games and K-with-AAA-only-games coefficients for a games-only rating.

Legend:

• Kuse - usability rating. The higher it is, the better card's mark is.
• Kpro - prospect rating. The higher it is, the better card's mark is.
• HL110xp - Half Life 2 (demo ixbt01) at 1024x768x32, maximum quality;
• HL112xp - Half Life 2 (demo ixbt01) at 1280x1024x32 maximum quality;
• HL116xp - Half Life 2 (demo ixbt01) 1600x1200x32 maximum quality;
• HL120xp - Half Life 2 (demo ixbt01) at 2048x1536x32 maximum quality;
• FC110xp - FarCry (RESEARCH level) at 1024x768x32;
• FC112xp - FarCry (RESEARCH level) at 1600x1200x32;
• FC116xp - FarCry (RESEARCH level) at 1600x1200x32;
• FC120xp - FarCry (RESEARCH level) at 2048x1536x32;
• D310xp - DOOM III (demo ixbt01) at 1024x768x32;
• D312xp - DOOM III (demo ixbt01) at 1600x1200x32;
• D316xp - DOOM III (demo ixbt01) at 1600x1200x32;
• G110xp - 3DMark05 Game1 at 1024x768x32;
• G112xp - 3DMark05 Game1 at 1280x1024x32;
• G116xp - 3DMark05 Game1 at 1600x1200x32;
• G120xp - 3DMark05 Game1 at 2048x1536x32;
• G210xp - 3DMark05 Game2 at 1024x768x32;
• G212xp - 3DMark05 Game2 at 1280x1024x32;
• G216xp - 3DMark05 Game2 at 1600x1200x32;
• G220xp - 3DMark05 Game2 at 2048x1536x32;
• G310xp - 3DMark05 Game3 at 1024x768x32;
• G312xp - 3DMark05 Game3 at 1280x1024x32;
• G316xp - 3DMark05 Game3 at 1600x1200x32;
• G320xp - 3DMark05 Game3 at 2048x1536x32;
• G1610xp - 3DMark06 Game1 at 1024x768x32;
• G1612xp - 3DMark06 Game1 at 1280x1024x32;
• G1616xp - 3DMark06 Game1 at 1600x1200x32;
• G1620xp - 3DMark06 Game1 at 2048x1536x32;
• G2610xp - 3DMark06 Game2 at 1024x768x32;
• G2612xp - 3DMark06 Game2 at 1280x1024x32;
• G2616xp - 3DMark06 Game2 at 1600x1200x32;
• G2620xp - 3DMark06 Game2 at 2048x1536x32;
• G3610xp - 3DMark06 Game3 at 1024x768x32;
• G3612xp - 3DMark06 Game3 at 1280x1024x32;
• G3616xp - 3DMark06 Game3 at 1600x1200x32;
• G3620xp - 3DMark06 Game3 at 2048x1536x32;
• G4610xp - 3DMark06 Game4 at 1024x768x32;
• G4612xp - 3DMark06 Game4 at 1280x1024x32;
• G4616xp - 3DMark06 Game4 at 1600x1200x32;
• G4620xp - 3DMark06 Game4 at 2048x1536x32;
• F10xp - F.E.A.R. at 1024x768x32;
• F12xp - F.E.A.R. at 1600x1200x32;
• F16xp - F.E.A.R. at 1600x1200x32;
• F20xp - F.E.A.R. at 2048x1536x32;
• SC10xp - SCCT at 1024x768x32;
• SC12xp - SCCT at 1600x1200x32;
• SC16xp - SCCT at 1600x1200x32;
• SC20xp - SCCT at 2048x1536x32;
• CD10xp - COD2 at 1024x768x32;
• CD12xp - COD2 at 1600x1200x32;
• CD16xp - COD2 at 1600x1200x32;
• CR110xp - Chronicles of Riddick (demo 44) at 1024x768x32;
• CR112xp - Chronicles of Riddick (demo 44) at 1600x1200x32;
• CR116xp - at Chronicles of Riddick (demo 44) at 1600x1200x32;
• Price - card price for the end of the month taken from price lists of a number of companies (for an average card);
• Poss - estimated functional potential and quality of a videocard (see remarks below);

Poss is calculated by the following formula:

Poss = ((AF * AA) ^ (1/2) * (Driver * VS * PS) ^ (1/3)) / 100

Where:

• Driver - estimated drivers quality (including installation difficulties, etc.) on the 10-mark scale.
• AF - anisotropic filtering:
  • = 500 * (K-with-AAA ^ (1/53)) / 60 fps - depends on relative performance.

• AA - antialiasing (on condition of minimal gameplay):
  • = 500 * (K-with-AAA ^ (1/53)) / 60 fps - MSAA;
  • = 1000 * (K-with-AAA ^ (1/53)) / 60 fps - MSAA+SSAA;
  • = 1500 * (K-with-AAA ^ (1/53)) / 60 fps - MSAA+SSAA+TAA;

• VS - hardware vertex shaders:
  • = 20 - version 3.0;
  • = 15 - version 2.0b;
  • = 5 - version 2.0;

• PS - hardware pixel shaders:
  • = 20 - version 3.0;
  • = 15 - version 2.0b;
  • = 5 - version 2.0;

P.S. We understand that such way of making ratings is not perfect, therefore we await suggestions and comments from our readers to be considered in our next 3Digest.