X-bit labs

Red Revenge: ATI Radeon X1950 XTX Graphics Card Review (page 10)

Category: Video

by Alexey Stepin , Yaroslav Lyssenko

[ 08/25/2006 | 09:27 AM ]

Performance in Theoretical Tests

Marko Dolenc’s Fillrate Tester

The increased memory bandwidth of the Radeon X1950 XTX affects its fill rate positively when working with the Z-buffer only or when mapping one texture. In other cases the number of TMUs becomes a limiting factor and the GeForce 7900 GTX has three times more TMUs, its core frequency being the same as the Radeon X1950 XTX’s.

The new card from ATI doesn’t have anything new to show us when it comes to processing pixel shaders. The memory frequency isn’t important here while the core frequencies of the Radeon X1900 XTX and the Radeon X1950 XTX are identical at 650MHz. As a result, the cards deliver identical performance, except for the shader that emulates per-pixel lighting and accesses memory often. The Radeon X1950 XTX is considerably faster than its predecessor at executing that shader.

Xbitmark version 0.65

There’s no wonder that the Radeon X1950 XTX delivers results comparable to those of the Radeon X1900 XTX. Their GPUs contain the same number of pixel processors and are clocked at the same frequency of 650MHz. There are improvements, however. You can observe them in shaders that work with textures (Cook Torrance + Texture + Fresnel, 27-Pass Fur or NPR (hatch)) as well as in shaders that contain intensive mathematic calculations (Metal + Phong, Wood). We’ll only see in real-life situations, i.e. in gaming tests, what effect this is going to have on the new card’s real performance.

Among other things we can note that the Radeon X19x0 family is still invincible when executing shaders with dynamic branching. Note also that the GeForce 7950 GX2 isn’t far better than the GeForce 7900 GTX in a majority of shaders included in this synthetic benchmark except for shaders with dynamic branching: the dual-chip solution from Nvidia improves its results here relative to the single-chip flagship thanks to the double number of pixel processors.