This piece of content is functionally an academic exercise. This means that occasionally we will end cases from the "real world", instead exploring how the map behaves and scales at a fundamental level. This means that you will look at invisible scenarios like 8K resolution just to see if AMD's scaling over NVIDIA is improved. Theoretically, AMD should confirm this based on what AMD said about performance, so we just have to confirm that. Anti-aliasing requires multiple samples per pixel, often 2x, 4x, 8x or 16x. This has a very similar impact on the resolution. It hits the same part of the pipeline.
If the memory bandwidth of AMD is really beneficial in games, we would expect the delta of performance between Radeon VII and the RTX 2080 to decrease with increasing resolution and anti-aliasing ratios, or vice versa. with a worsening or widening gap that RTX prefers, as the resolution decreases. We've already shown how Radeon VII sometimes closed the gap at higher resolutions when we published our review. We talked about how the gap ̵
Superposition AA Testing
Start with shaders that are set to "Low" and "Resolutions" at 4K, 5K and 8K. We saw that NVIDIA had a 27.4% lead over the baseline, improving to 27.9% for 5K and a 28.9% error 8 THOUSAND. There is really no change from 4K to 8K here, at least in terms of scaling, though these resolutions will obviously be more playable with increasing magnification.
Next, we performed a one-time test with "extreme" shields and found that AMD actually gained ground in this scenario. AMD averaged 1566 points and NVIDIA 1504 points. We repeated and averaged this test four times and found that this delta was outside the standard deviation.
After learning that extreme shaders show bigger differences, we repeated the tests from 1080p to 4K. At 1080p, the NVIDIA RTX 2080 outperformed the Radeon VII card by approximately 25.8% in 100% of its underlyings. With 1440p, the advantage of the RTX 2080 fell quickly to 16.1%. 4K results are too close and create a chaotic graph, but the RTX 2080 has actually been surpassed by Radeon VII here, with Radeon VII holding a few percentage points ahead. In terms of the "real" frame rate, the benchmark was now 11-12 FPS AVG. This is obviously not playable, and this benefit is not really recognized because nobody would use these settings realistically, but it's interesting to see how two cards in different test scenarios can flip positions.
GTA V AA Testing  GTA V offers many anti-aliasing options. So we train MSAA from 2x to 8x and then activate 8x reflections MSAA. We start with an AVG-FPS diagram of the different settings, which best illustrates how the gap changes with each test.
At the 4K and Very High and Ultra settings for everything, the value of 2x MSAA is 58.8 FPS AVG for the 2080 and 51.4FPS AVG for the Radeon VII, but the most important clue is that the 2080 has one 14.4% advantage over the Radeon VII card. At 4x MSAA, this lead does not change much and is lowered to 14.1%. At 8x MSAA we start to see some movements. The AVG FPS is 27.5 for Radeon VII and 31FPS for the RTX 2080. The improvement in the RTX 2080 rose to 13% (previously 14.4%). By adding 8 additional MSAA reflections and anti-aliasing, we see that the difference for the 2080 FE changes to a 10.8% advantage. Looking at the line chart as a whole, we can see that AMD can gain some losses to the extreme end of the scale, it's just that we're in an unplayable area for other reasons, passes that we ran, the standard deviation was about 0.3 FPS AVG. These results are very accurate.
Firestrike AA Testing
Firestrike is still one of the best tools for this kind of synthetic workload. For this we will consider GT1 and GT2 separately. GT1 heavily loads the GPU with polygons and tessellations, performs shadow and illumination crunching, and uses rake arrays for post processing and particle physics. GT2 is very much for the computational shader workload and greatly increases per-pixel pixels, but reduces the tessellation workload by more than 50%. GT2 should therefore run better on Radeon VII than GT1, in relative terms.
Based on GT2, we can see that the Firestrike Ultra NVIDIA settings are below AMD performance and are at 100% at 86.7% of Radeon VII's base performance. This is unique in this benchmark so far. At 2x MSAA, this gap shrinks to 90.7% of baseline performance. With 4x MSAA, we reach 94.6% of the Underlyings, with the 8x MSAA eventually causing the RTX 2080 to outperform AMD and account for 104.7% of the base performance.
It makes sense that AMD in GT2, where the processing power and computing power are particularly high memory are both emphasized. Why NVIDIA evolves after 8 times the MSAA strategy, we work with the 3DMark team and others to better understand their specific behavior. Our assessment is that NVIDIA memory compression or anti-aliasing edge detection may have something to do with this particular software implementation or other implementation-level benefit.
For GT1, where tessellation and polygons are heavier We see that the two cards are functionally the same under the same Firestrike Ultra settings. NVIDIA gained a remarkable 16% at 2xMSAA, 21% at 4x MSAA and 27% at 8x MSAA.
At 8K with 8x MSAA we are really useless Fradeate, but finally Radeon VII is far ahead of the RTX 2080. This was the reason to completely exhaust the memory of the RTX 2080, whereupon he started choking as he fetched and dropped memory.
Far Cry 5 HD Textures + AA Testing
Far Cry 5 is next. We tested here with SMAA, TAA and HD Textures under different scenarios.
In the first test at 4K and using HD Textures with SMAA, we measured NVIDIA at 62.8 FPS AVG and AMD at 60.6 FPS AVG, which means NVIDIA is about 3.5% ahead. 1% lows were similarly spaced as 0.1% lows.
To test scaling with different settings, we also ran a test with reduced VRAM consumption but otherwise identical settings. We can now switch to this diagram. The results that disable HD textures were functionally identical to the results for HD textures. It's not surprising that we are within the error range, as textures do not really affect performance unless VRAM becomes a constraint. For both devices, the experience is quantitatively unchanged compared to the previous diagram. Qualitatively, the output is the same on every device. Again, NVIDIA is about 3.3% faster.
When switching to TAA, performance drops slightly on each device, but NVIDIA still has an advantage of 3.3% on AVG FPS. HD textures and anti-aliasing changes in Far Cry 5 are not enough to reduce or change the performance delta from card to card.
Conclusion: Not quite what we expected
Our hypothesis was not entirely correct. It looks like we nailed it in some cases, but Firestrike threw us off track. We contacted the 3DMark team and NVIDIA engineers to understand and understand this specific performance scenario. The only case in which we saw a significant increase for Radeon VII was when we exhausted the RTX 2080's VRAM, which was achieved by operating Firestrike Ultra at a resolution of 8K and 8x MSAA. This, of course, is completely unplayable, and once more we point out that this is really just a research exercise, not a very practical exercise. At 8K / 8xMSAA we have seen that Radeon VII was operated at 4.7-4.8FPS AVG to 2080 ~ 1-1.7FPS AVG. This is a "big" percent improvement. This is because we have exhausted the frame buffer of the 2080, but it is clearly not playable on both devices. Regardless, it is interesting data and may help to better understand other features of the card in future tests.
Editorial, Testing: Steve Burke
Video: Andrew Coleman