A Mere 'Speed Bump'?
From these games, we can see that there's more than enough evidence to support the claim that NVIDIA simply took the GeForce 7800 design, optimized it some more with minor adjustments for process tweaks and walla - you have the GeForce 7900GTX. Be that as tit may, there were some differences between the results, so please be patient with us. Aside from process tweaks, just in what areas NVIDIA 'played around' with?F.E.A.R
| 7800GTX | 7900GTX | |||||||
| AA penalty (Bilinear) | AA penalty (AF) | AF penalty | AF penalty (AA) | AA penalty (Bilinear) | AA penalty (AF) | AF penalty | AF penalty (AA) | |
| 1024 | ||||||||
| Min | 17.78% | 23.40% | -4.44% | 5.63% | 0.00% | 10.87% | -4.55% | 10.87% |
| Avg | 34.59% | 28.08% | 16.59% | -6.51% | 21.88% | 13.18% | 10.31% | -8.71% |
| Max | 50.58% | 19.03% | 34.30% | -31.55% | 23.25% | -51.19% | 6.69% | -74.44% |
| Average | 34.32% | 23.50% | 15.48% | -10.81% | 15.04% | -9.05% | 4.15% | -24.09% |
| 1280 | ||||||||
| Min | 40.00% | 36.59% | 8.89% | -3.41% | 11.63% | 16.28% | 0.00% | 4.65% |
| Avg | 39.51% | 35.30% | 15.65% | -4.21% | 31.26% | 33.05% | 6.07% | 1.79% |
| Max | 48.64% | 34.36% | 36.58% | -14.28% | 62.84% | 65.99% | -1.83% | 3.15% |
| Average | 42.72% | 35.42% | 20.37% | -7.30% | 35.24% | 38.44% | 1.41% | 3.20% |
| 1600 | ||||||||
| Min | 44.12% | 36.67% | 11.76% | -7.45% | 40.91% | 39.53% | 2.27% | -1.37% |
| Avg | 40.47% | 37.58% | 12.45% | -2.89% | 43.02% | 34.88% | 15.54% | -8.14% |
| Max | 56.08% | 54.49% | 5.82% | -1.59% | 24.91% | 37.20% | 39.03% | 12.29% |
| Average | 46.89% | 42.91% | 10.01% | -3.98% | 36.28% | 37.20% | 18.95% | 0.92% |
| Weighted Average | 41.31% | 33.94% | 15.29% | -7.36% | 28.86% | 22.20% | 8.17% | -6.66% |
We think the changes in Forceware 84.21 is distorting the facts here, but you can definitely see the differences between the two cards at their respective clocks. Look at the 1024 x 768 results. Running without practically any filtering (bilinear is just so plain, ugly and old), the AA penalty on the GeForce 7900GTX is half that of a GeForce 7800GTX. Look at the average fps results, the maximum fps is distorting the average penalty hit here. However, AF only got something like 1/3 faster - 10 percent on the GeForce 7900GTX as opposed to 15 percent on the GeForce 7800GTX. Moving to 1280 x 960, we can still see a similar trend, although it's less pronounced but at 1600 x 1200, we're seeing the same penalty hit on both cards.
With nearly a 50 percent increase in clock, these AA penalty results is another piece of evidence that supports the 'speed bump' theory. Fillrate is the main factor for determining AA performance. So, why only about 30 percent faster with AF? Well, if you're quite the avid reader (you must be to be reading this far), that's not a surprise - it corresponds nicely with the increase in memory clock thus memory bandwidth. If NVIDIA had push the memory even higher, we may get the same penalty rate in AA and AF.
Quake 4
| 7800GTX | 7900GTX | |||||||
| AA penalty (Bilinear) | AA penalty (AF) | AF penalty (bilinear) | AF penalty (AA) | AA penalty (Bilinear) | AA penalty (AF) | AF penalty | AF penalty (AA) | |
| 1024 | ||||||||
| Min | 5.80% | 13.64% | 4.35% | 7.84% | -5.00% | -6.67% | 0.00% | -1.67% |
| Avg | 21.77% | 28.74% | 1.28% | 6.98% | 6.13% | 8.40% | 0.78% | 2.27% |
| Max | 4.68% | 19.64% | 1.75% | 14.96% | 0.61% | 3.11% | 2.42% | 2.50% |
| Average | 10.75% | 20.67% | 2.46% | 9.93% | 0.58% | 1.61% | 1.07% | 1.03% |
| 1280 | ||||||||
| Min | 42.25% | 47.14% | 1.41% | 4.89% | 7.14% | 20.00% | -7.14% | 12.86% |
| Avg | 45.31% | 45.68% | 8.65% | 0.37% | 26.79% | 31.70% | 0.95% | 4.92% |
| Max | 34.12% | 35.58% | 4.12% | 1.47% | 14.71% | 20.86% | 4.12% | 6.15% |
| Average | 40.56% | 42.80% | 4.73% | 2.24% | 16.21% | 24.19% | -0.69% | 7.98% |
| 1600 | ||||||||
| Min | 59.42% | 54.24% | 14.49% | -5.18% | 25.45% | 34.48% | -5.45% | 9.03% |
| Avg | 53.25% | 51.22% | 11.35% | -2.02% | 44.80% | 46.12% | 3.09% | 1.32% |
| Max | 47.47% | 40.74% | 14.56% | -6.73% | 34.78% | 38.41% | -1.86% | 3.63% |
| Average | 53.38% | 48.73% | 13.47% | -4.64% | 35.01% | 39.67% | -1.41% | 4.66% |
| Weighted Average | 34.90% | 37.40% | 6.89% | 2.51% | 17.27% | 21.82% | -0.34% | 4.56% |
These results are very interesting to explain. Remember, fillrate and bandwidth limits don't kick in as early on the GeForce 7900GTX. It's makes much more sense to compare 1024 x 768 on the GeForce 7800GTX and 1280 x 1024.on the GeForce 7900GTX. OK - why are we seeing a slightly higher AA penalty rate on the GeForce 7900GTX? That's just one of the 'quirks' we told you about, the GeForce 7900GTX seems to be slower than the GeForce 7800GTX.
What's interesting here is AA penalty with and without AF. Look at the 1280 x 1024 results on both cards. That's about 2 percent difference on the GeForce 7800GTX, but about 8 percent on the GeForce 7900GTX! This 'may' be normal variations between runs, but lets look a little deeper. Since the difference here is AF, let's look at AF penalty rate. Again, we're seeing the same numbers (2 vs 8 percent). We believe this is the drop point - where memory bandwidth is not keeping up with the core. To confirm this hypothesis, just look at the 1600 x 1200 results: AF penalty with AA is pretty much the same between the two cards (4 - 6 percent), so at this stage the two cards is bandwidth bound and unsurprisingly behaves the same way.
Serious Sam II
| 7800GTX | 7900GTX | ||||||||
| AA penalty (Bilinear) | AA penalty (AF) | AF penalty | AF penalty (AA) | AA penalty (Bilinear) | AA penalty (AF) | AF penalty | AF penalty (AA) | ||
| 1024 | |||||||||
| Min | 0.00% | 0.00% | 2.27% | 0.00% | -2.38% | -2.38% | 0.00% | 0.00% | |
| Avg | 1.49% | 5.51% | 3.52% | 4.02% | 1.20% | -0.47% | 1.20% | -1.67% | |
| Max | 1.23% | 10.81% | 8.64% | 9.58% | 1.28% | -2.60% | 1.28% | -3.88% | |
| Average | 0.91% | 5.44% | 4.81% | 4.53% | 0.03% | -1.82% | 0.83% | -1.85% | |
| 1280 | |||||||||
| Min | 2.22% | 21.43% | 6.67% | 19.21% | 4.65% | 4.76% | 2.33% | 0.11% | |
| Avg | 5.92% | 15.28% | 16.62% | 9.36% | 1.01% | 2.31% | 2.95% | 1.30% | |
| Max | 13.75% | 15.00% | 25.00% | 1.25% | 0.00% | 6.58% | 1.30% | 6.58% | |
| Average | 7.30% | 17.23% | 16.09% | 9.94% | 1.89% | 4.55% | 2.19% | 2.66% | |
| 1600 | |||||||||
| Min | 16.67% | 18.75% | 23.81% | 2.08% | 4.76% | 23.81% | 0.00% | 19.05% | |
| Avg | 28.59% | 18.54% | 36.90% | -10.05% | 3.50% | 18.74% | 8.15% | 15.24% | |
| Max | 30.26% | 16.28% | 43.42% | -13.98% | 7.89% | 18.75% | 15.79% | 10.86% | |
| Average | 25.17% | 17.86% | 34.71% | -7.32% | 5.38% | 20.43% | 7.98% | 15.05% | |
| Weighted Average | 11.13% | 13.51% | 18.54% | 2.38% | 2.43% | 7.72% | 3.67% | 5.29% |
Things really don't get interesting until we hit 1600 x 1200 - the point where bandwidth can't keep up with fillrate. At 1280 x 960, we can see that 'pure' AA penalty is actually very small on the GeForce 7800GTX - 7 percent. It's AF that's the bigger frame rate dropper here at 16 percent. So much so, that the AA penalty is effectively 'hidden' by AF penalty. The situation is similar, but not quite the same on the GeForce 7900GTX. We pretty much still have the bandwidth at 1280 x 960, so AF penalty is accumulative with AA penalty.
[Previous Page]
[Go to top]
[Next Page]