Difference between revisions of "Sega Saturn/Hardware comparison"

From Sega Retro

Line 52: Line 52:
 
| 2,400,000 vertices/sec,{{ref|Transformation (21 adds/multiplies),{{fileref|ST-240-A-SP1-052295.pdf|page=8}} projection (4 adds/multiplies){{ref|1=[https://books.google.co.uk/books?id=iAvHt5RCHbMC&pg=PA97 ''Design of Digital Systems and Devices'' (page 97)]}} and perspective division (1 divide){{ref|1=[http://www.cs.virginia.edu/~gfx/Courses/2012/IntroGraphics/lectures/13-Pipeline.pdf#page=50 3D Polygon Rendering Pipeline (page 50)]}} per vertex:
 
| 2,400,000 vertices/sec,{{ref|Transformation (21 adds/multiplies),{{fileref|ST-240-A-SP1-052295.pdf|page=8}} projection (4 adds/multiplies){{ref|1=[https://books.google.co.uk/books?id=iAvHt5RCHbMC&pg=PA97 ''Design of Digital Systems and Devices'' (page 97)]}} and perspective division (1 divide){{ref|1=[http://www.cs.virginia.edu/~gfx/Courses/2012/IntroGraphics/lectures/13-Pipeline.pdf#page=50 3D Polygon Rendering Pipeline (page 50)]}} per vertex:
 
*1,022,727 vertices/sec: 1,022,727 SCU DSP transformations (14 cycles per transformation),{{fileref|ST-240-A-SP1-052295.pdf|page=8}} 4,090,908 SH-2 MULT DSP projection operations (53,181,812 remaining), 1,022,727 SH-2 DIVU divisions (445,804 remaining)
 
*1,022,727 vertices/sec: 1,022,727 SCU DSP transformations (14 cycles per transformation),{{fileref|ST-240-A-SP1-052295.pdf|page=8}} 4,090,908 SH-2 MULT DSP projection operations (53,181,812 remaining), 1,022,727 SH-2 DIVU divisions (445,804 remaining)
*445,804 vertices/sec: 11,145,100 SH-2 MULT DSP transform/projection operations (42,036,712 remaining), 445,804 DIVU divisions (1 divide per vertex)
+
*445,804 vertices/sec: 11,145,100 SH-2 MULT DSP transform/projection operations (42,036,712 remaining), 445,804 SH-2 DIVU divisions (1 divide per vertex)
 
*1,025,285 vertices/sec: 42,036,712 SH-2 MULT DSP operations (41 cycles per vertex)
 
*1,025,285 vertices/sec: 42,036,712 SH-2 MULT DSP operations (41 cycles per vertex)
 
|group=fn}} <br> 1,800,000 polygons/sec{{ref|8 vertices per cube (6 quad polygons){{fileref|ST-237-R1-051795.pdf|page=51}}|group=fn}}
 
|group=fn}} <br> 1,800,000 polygons/sec{{ref|8 vertices per cube (6 quad polygons){{fileref|ST-237-R1-051795.pdf|page=51}}|group=fn}}

Revision as of 11:59, 6 November 2016


This short article is in need of work. You can help Sega Retro by adding to it.


Vs. PlayStation

The Sega Saturn is generally more powerful than the rival PlayStation,[1][2] but more difficult to get to grips with.[2] The Saturn has more raw computational power and faster pixel drawing; the PS1 can only draw pixels through its polygon engine, whereas the Saturn can draw pixels directly with its processors, giving it more programming flexibility.[3]

When both SH-2 and the SCU DSP are used in parallel, the Saturn is capable of 160 MIPS and 85 million fixed-point operations/sec, faster than the PS1's GTE (66 MIPS); when programmed effectively, the Saturn's parallel geometry engine can calculate more 3D geometry than the PS1. The's VDP1 has a fillrate of 28.6364 MPixels/s per framebuffer, compared to the PS1's GPU which has a fillrate of 30 MPixels/s (15-bit RGB) or 15 MPixels/s (24-bit RGB). The fillrate for 8×8 textures is 18 MTexels/s for the VDP1 and 15.28 MTexels/s for the PS1's GPU (4000 8×8 sprites).[4][5][6]

The VDP2 has a significantly higher effective tile fillrate of 500 MPixels/s; if the VDP2 is used for drawing textured infinite planes, this frees up the VDP1's polygons for other 3D assets, whereas the PS1 needs to draw many polygons to construct 3D textured planes (with very limited draw distance compared to the VDP2). The VDP1's quad polygons are drawn with edge anti‑aliasing (for smoother edges) and forward texture mapping (with limited perspective correction), while the VDP2's infinite planes are drawn with true perspective correction, whereas the PS1's triangle polygons have aliased edges and are drawn with affine texture mapping which lacks perspective correction (resulting in perspective distortion and texture warping). The PS1 has more effective polygon transparency than the VDP1, while the VDP2 has more effective transparency than the PS1. The VDP1 is more effective at Gouraud shading than the PS1's GPU, while the VDP2 is more effective at visual effects such as misting and reflective water effects.

The PS1's straightforward hardware architecture, triangle polygons, and more effective development tools and C language support, made it easier for developers to program 3D graphics. When it came to 2D graphics, on the other hand, the Saturn's combination of a VDP1 sprite framebuffer and VDP2 parallax scrolling backgrounds made it both more powerful and straightforward to program 2D graphics, compared to the PS1 which draws all 2D graphics to a single framebuffer.

Vs. Nintendo 64

Vs. PC

The Saturn's VDP1 was the basis for Nvidia's first graphics processor, the NV1, which was one of the first 3D graphics accelerators on PC, released in 1995. Like the Saturn, it uses quad polygons and supports forward texture mapping with limited perspective correction, and several Saturn ports are available for it. However, the NV1 has a fillrate of 12.5 MPixels/s and a rendering performance of 50,000 polygons/sec, less than the VDP1's 28 MPixels/s per framebuffer and more than 500,000 polygons/sec rendering throughput. In comparison, the most powerful PC graphics card of 1995, Yamaha's Tasmania 3D, which was based on triangle polygons, had a 25 MPixels/s fillrate and 300,000 polygons/sec rendering throughput, more than the NV1, but less than the Saturn and PlayStation.

Comparison table

See Sega Saturn technical specifications for more technical details on Saturn hardware
System Sega Saturn (1994) Sony PlayStation (1994) PC (1995)
Geometry processors 2x Hitachi SH-2 (28.63636 MHz),[fn 1]
Sega SCU DSP (14.31818 MHz)
Sony GTE (33.8688 MHz)[9] Intel Pentium (133 MHz)
Arithmetic operations 87 MOPS[fn 2] 66 MOPS[12] 44 MOPS[fn 3]
Additions 71 million adds/sec[fn 4] 66 million adds/sec 44 million adds/sec[fn 5]
Multiplications 71 million multiplies/sec[fn 6] 66 million multiplies/sec 44 million multiplies/sec[fn 7]
16-bit divisions 5 million divides/sec[fn 8] 4 million divides/sec[fn 9] 4 million divides/sec[fn 10]
Geometry transformations 2,400,000 vertices/sec,[fn 11]
1,800,000 polygons/sec[fn 12]
1,900,000 vertices/sec,[fn 13]
1,300,000 polygons/sec[fn 14]
1,100,000 vertices/sec,[fn 15]
360,000 polygons/sec[fn 16]

Footnotes

  1. [2x CPU cores (28.63636 MHz),
    2x MULT multiplier DSP (28.63636 MHz),
    2x DIVU division units (28.63636 MHz)[7][8] 2x CPU cores (28.63636 MHz),
    2x MULT multiplier DSP (28.63636 MHz),
    2x DIVU division units (28.63636 MHz)[7][8]]
  2. [2x SH-2 MULT DSP: 57.27272 MOPS (million operations per second)[10]
    2x SH-2 DIVU: 1.468531 MOPS (39 cycles per divide)[11]
    SCU DSP: 28.63636 MOPS (add and multiply per cycle) 2x SH-2 MULT DSP: 57.27272 MOPS (million operations per second)[10]
    2x SH-2 DIVU: 1.468531 MOPS (39 cycles per divide)[11]
    SCU DSP: 28.63636 MOPS (add and multiply per cycle)]
  3. [3 cycles per add/multiply[13] 3 cycles per add/multiply[13]]
  4. [2x SH-2: 57,272,720 adds/sec (1 cycle per multiply)[10]
    SCU DSP: 14,318,180 multiplies/sec (1 cycle per multiply) 2x SH-2: 57,272,720 adds/sec (1 cycle per multiply)[10]
    SCU DSP: 14,318,180 multiplies/sec (1 cycle per multiply)]
  5. [3 cycles per add[13] 3 cycles per add[13]]
  6. [2x SH-2: 57,272,720 multiplies/sec (1 cycle per multiply)[14]
    SCU DSP: 14,318,180 multiplies/sec (1 cycle per multiply) 2x SH-2: 57,272,720 multiplies/sec (1 cycle per multiply)[14]
    SCU DSP: 14,318,180 multiplies/sec (1 cycle per multiply)]
  7. [3 cycles per multiply[13] 3 cycles per multiply[13]]
  8. [2x CPU: 3,579,545 divides/sec (16 cycles per 16-bit divide)[15]
    2x DIVU: 1,468,531 divides/sec (39 cycles per divide)[11] 2x CPU: 3,579,545 divides/sec (16 cycles per 16-bit divide)[15]
    2x DIVU: 1,468,531 divides/sec (39 cycles per divide)[11]]
  9. [25 cycles (23 cycles instruction,[16] 2 cycles delay[17]) per 3 divides[18] 25 cycles (23 cycles instruction,[16] 2 cycles delay[17]) per 3 divides[18]]
  10. [30 cycles per divide[19] 30 cycles per divide[19]]
  11. [Transformation (21 adds/multiplies),[20] projection (4 adds/multiplies)[21] and perspective division (1 divide)[22] per vertex:
    • 1,022,727 vertices/sec: 1,022,727 SCU DSP transformations (14 cycles per transformation),[20] 4,090,908 SH-2 MULT DSP projection operations (53,181,812 remaining), 1,022,727 SH-2 DIVU divisions (445,804 remaining)
    • 445,804 vertices/sec: 11,145,100 SH-2 MULT DSP transform/projection operations (42,036,712 remaining), 445,804 SH-2 DIVU divisions (1 divide per vertex)
    • 1,025,285 vertices/sec: 42,036,712 SH-2 MULT DSP operations (41 cycles per vertex)
    Transformation (21 adds/multiplies),[20] projection (4 adds/multiplies)[21] and perspective division (1 divide)[22] per vertex:
    • 1,022,727 vertices/sec: 1,022,727 SCU DSP transformations (14 cycles per transformation),[20] 4,090,908 SH-2 MULT DSP projection operations (53,181,812 remaining), 1,022,727 SH-2 DIVU divisions (445,804 remaining)
    • 445,804 vertices/sec: 11,145,100 SH-2 MULT DSP transform/projection operations (42,036,712 remaining), 445,804 SH-2 DIVU divisions (1 divide per vertex)
    • 1,025,285 vertices/sec: 42,036,712 SH-2 MULT DSP operations (41 cycles per vertex)]
  12. [8 vertices per cube (6 quad polygons)[23] 8 vertices per cube (6 quad polygons)[23]]
  13. [17 cycles (15 cycles instruction,[24] 2 cycles delay[17]) per vertex 17 cycles (15 cycles instruction,[24] 2 cycles delay[17]) per vertex]
  14. [25 cycles (23 cycles instruction,[16] 2 cycles delay) per triangle 25 cycles (23 cycles instruction,[16] 2 cycles delay) per triangle]
  15. [114 cycles per vertex (28 multiplies/adds,[25] 1 divide[22]) 114 cycles per vertex (28 multiplies/adds,[25] 1 divide[22])]
  16. [3 vertices per triangle polygon 3 vertices per triangle polygon]

References