NVIDIA GeForce 8800 GT (G92)

Part 2: Features, Synthetic Tests

Direct3D 10: PS 4.0 Tests (texturing, loops)

New RightMark3D 2.0 includes two old Direct3D 9 PS 3.0 tests, rewritten for DirectX 10, and two brand new tests. The first two tests can now enable self-shadowing and shader supersampling, which increase their load on GPUs.

These tests measure efficiency of executing looped pixel shaders with a lot of texture lookups (up to several hundreds of lookups per pixel in the heaviest mode!) and a relatively low ALU load. In other words, they measure a texture sampling rate and branching efficiency in a pixel shader.

The first pixel shader test will be the Fur test. When used with the lowest settings, it uses 15-30 texture lookups from bump maps and two lookups from the main texture. The High Effect Detail mode increases the number of lookups to 40-80. When shader supersampling is enabled - the number of lookups grows to 60-120. And the High mode with SSAA is the heaviest mode - 160-320 lookups from a bump map.

Let's see what happens in modes without supersampling - they are relatively simple, and the correlation of results in Low/High modes must be similar.

The Fur tests with lots of texture lookups show a huge advantage of NVIDIA solutions over the RADEON HD 2900 XT. The previous Mid-End card demonstrates a higher result than the top solution from AMD. This lag is theoretically impossible. Perhaps, AMD still hasn't finetuned its Direct3D 10 drivers.

All results in the High mode are approximately 1.5 times as low as in the Low mode. Results of both GeForce 8800 GTS cards indicate no influence of memory size on the tests. The GeForce 8600 GTS is slower than the G80-based solutions by as much as it should, cut down execution units seriously affect its performance. Judging by the results demonstrated by the GeForce 8800 GT and the GTX cards, performance in this test depends not only on the number and speed of TMUs, or the disparity would have been different. It's 20-25% in the test. Only the fill rate and memory bandwidth differ by this value.

Let's have a look at the results in this test with enabled shader supersampling, which quadruples the load. Perhaps it will change the situation:

Only top GPUs from NVIDIA can cope with such complexity. The GeForce 8600 GTS still outperforms the best of AMD, being two or three times as slow as the G80-based cards. And the GeForce 8800 GT is still outperformed by the GTX card. However, it closes the gap as the shader gets more complex and the GPU load grows. Supersampling quadruples the load. But it slows down G8x-based cards approximately by five times, and R6xx-based cards - only by 3.5 times. So the HD 2900 XT almost catches up with the GeForce 8600 GTS.

The second test that measures efficiency of executing complex looped pixel shaders with many texture lookups is called Steep Parallax Mapping. With low settings it uses 10-50 texture lookups from a bump map and three lookups from main textures. The heavy mode with self-shadowing doubles the number of texture lookups, and supersampling quadruples this number. The most complex test mode with supersampling and self-shadowing uses 80-400 texture lookups, that is eight times as many as in the low mode. Let's analyze simple modes without supersampling first:

This test is more interesting from the practical point of view. Various parallax mapping methods have been used in games for a long time already. Heavy modifications, such as our steep parallax mapping, will soon be used, e.g. in Crysis. Along with supersampling, this test can enable self-shadowing that doubles the GPU load (High mode).

Even though AMD solutions have been traditionally strong in our Direct3D 9 tests of parallax mapping, the RADEON HD 2900 XT performs only on a par with the GeForce 8600 GTS in the updated test without supersampling. Besides, self-shadowing causes a bigger performance drop in AMD products, over two times versus 1.5 in NVIDIA solutions.

The GeForce 8800 GT is noticeably faster than the old GeForce 8600 GTS, their performance difference reaches three times. The GeForce 8800 GTX is still in the lead, but the GT card is slower only by 13-16%. Let's see what supersampling will change. Performance drop from supersampling was bigger in NVIDIA cards in the previous test, so it brought the 8800 GT and the GTX cards closer to each other.

FPS values obtained with enabled supersampling and self-shadowing again indicate a very heavy GPU load. These two options enabled together increase the load by almost eight times, causing a catastrophic performance drop. The performance difference between our graphics cards remains. However, when supersampling is enabled, the AMD card improves its results versus NVIDIA, just like in the previous case. NVIDIA drops performance fourfold, AMD - only threefold. But the R600 is still faster only than the G84, being outperformed by other graphics cards.

Both GeForce 8800 GTS cards demonstrate identical results, while the GeForce 8600 GTS is twice as slow. What concerns a comparison of the GeForce 8800 GT and GTX, the new product is outperformed by the GTX card. But the difference reaches 10-12% in this complex test! Modified TMUs with more address units of the G92 actually have no advantages over the G80 in real conditions.

Direct3D 10: PS 4.0 Tests (computing)

The next couple of pixel shader tests contains very few texture lookups to minimize the effect of TMUs on performance. They use a lot of arithmetic operations, so they measure arithmetic performance of GPUs, how fast they execute arithmetic instructions in pixel shaders.

The first computing test is called Mineral. It's a complex procedural texturing test, which uses only two texture lookups and 65 sin and cos instructions.

According to our synthetic Direct3D 9 tests, the AMD R6xx performs better than the competing architecture from NVIDIA in complex arithmetic tasks. The RADEON HD 2900 XT is apparently faster in this test, G80/G92-based solutions cannot compete with it here.

The GeForce 8800 GT performs almost three times as fast as the GeForce 8600 GTS, the traditional ratio again. The new product from NVIDIA outperforms both GTS cards and is slightly slower than the GeForce 8800 GTX, conforming with performance (the number and clock rate) of unified processors - less than by 5%.

The second shader test is called Fire, it's even harder for ALUs. It contains only a single texture lookup, while the number of sin/cos instructions is doubled to 130. Let's see what changes as the load grows:

We cannot compare the new NVIDIA card with the RADEON HD 2900 XT and future cards from AMD, the bug in AMD drivers is not fixed yet. Another possible reason is a hardware problem. In this case AMD is doomed in this test.

What concerns a comparison with graphics cards from the same camp, the situation is the same - the GeForce 8800 GT is three times as fast as the 8600 GTS, it's faster than any 8800 GTS by 40-50%, and it's outperformed by the GTX card only by 3% - exactly the difference in ALUs, their frequencies, and the number of execution units.