AMD architecture performance analysis: DirectX11 Focus

Testing was carried out on a Core i7 8700K system running at stock frequencies, with all titles typically running all cores and threads at 4.3GHz. System memory was supplied by GSkill with 3200MHz CL14 FlareX modules while our motherboard is an MSI Gaming X370 offering. All games were run from SSD, with benchmarks calculated using FCAT data. Essentially, games are run with v-sync disabled at the settings described, with each new frame marked up with a brightly coloured border. By tracking the border in our software, we can determine frame-rate and frame-time. Multi-gigabyte video captures are condensed into frame data .txt files which we feed into the Eurogamer CMS to create the graphs below. Every single frame is available for your perusal, with frame-times viewable in context by pressing play on the appropriate YouTube insert.

For desktop users, the graphs themselves are interactive too. Mouse over a particular entry to get the appropriate reading, or better yet, click the graph for the arguably much more useful percentage differentials. In short, every frame we capture is represented on these pages and you have as much or as little performance data as you desire. At-a-glance or deep-dive, it's all here.

For our first quartet of games, we're going to be looking at DirectX11 titles, a mixture of two established 'classics' and two more recent entries. Typical form for benchmarking involves ramping everything up to ultra and measuring from there, but we've changed tack for this piece. The Radeon R9 280X has a hard limit of three gigs of VRAM. We wanted performance data at both 1080p and 1440p resolutions, without hitting memory limits. In some cases, our existing bench works as is. In others, we've had to dial back texture quality, while some instances required pulling down settings across the board - exactly what we had to do with our first title, the now infamous and indeed legendary Assassin's Creed Unity.

Assassin's Creed Unity

To begin with, remember that all cards are running at 1.0GHz with 256GB/s of memory bandwidth. So Tahiti and Polaris with 32 compute units are delivering 4.096 teraflops, with Polaris and Navi with 36 CUs at the same frequencies handing in 4.6TF. That's an additional 12.5 per cent of compute power, but as you can see here, the difference between our two Polaris cards amounts to just four per cent or so of extra performance. This is why AMD ships RX 580 with higher clocks and faster RAM than the RX 570. Assassin's Creed has never been especially friendly to AMD hardware and across the generations, stacking Tahiti/Polaris and Polaris/Navi gains, we get a 30 per cent increase at 1080p, rising to 31 per cent at 1440p.

Crysis 3

Crysis 3 sees a much more significant improvement in normalised performance across the generations. Tahiti to Polaris sees an 11 to 12 per cent improvement in performance, while Polaris to Navi sees an increase of 16 to 18 per cent. Across seven years of AMD architectural development, chaining the percentage gains across architectures together, we get a 50 per cent increase at 1080p, dropping to 44 per cent at 1440p. This is more in line with the results seen in 3DMark Firestrike on the first page. Again, with RX 570 and 580 matched clock for clock, the extra CUs are delivering very little - an extra four per cent of thereabouts of performance.

Far Cry 5

In Far Cry 5, we're running at normal settings rather than the usual ultra in order to ensure that the game doesn't trouble the 3GB VRAM limitation of our Tahiti-based R9 280X. There are some interesting takeaways here. Firstly, the performance bump between our two Polaris models is higher than our previous tests - a nice result offset by Polaris's really poor scaling against Tahiti. Happily though, Navi fares better in terms of generational gains. Stacking up Tahiti against Polaris, then Polaris against Navi, the end to end gain amounts to 38 per cent at 1080p and 34 per cent at 1440p.

Ghost Recon Wildlands

The ultra setting on this title reduces even some of the most modern GPUs to sub 60fps figures and absolutely soaks VRAM. Happily, just dropping down one 'notch' on the settings to very high keeps us within the 3GB VRAM budget of the Tahiti-powered R9 280X and still stress-tests the hardware intensely. Again, you'll note a paltry three to four per cent differential between RX 570 and 580, but Navi's lead over Polaris at the same spec crosses the 30 per cent threshold. This title at high settings really does stress compute - which is the aim of the exercise, after all. Stacking up the architectural gains we get a 59 per cent uptick at 1080p vs 55 per cent at 1440p. Impressive stuff!

AMD RDNA vs GCN Analysis:

• Introduction, video analysis, synthetic benchmarks
• Gaming benchmarks DX11: AC Unity, Crysis 3, Ghost Recon Wildlands, Far Cry 5 [This Page]
• Gaming benchmarks DX12: Rise/Shadow of the Tomb Raider, Strange Brigade, Wolfenstein 2
• Gaming benchmark problem children: Battlefield 1, Forza Horizon 4, The Witcher 3
• AMD architecture: Navi memory bandwidth scaling and conclusion