How does the division of labor work and what is Alternate Frame Randering (AFR)?
AFR is currently the most common variant for the render mode in multi-GPU solutions. As the name suggests, the frames are calculated alternately by the respective cards involved in the piece. However, a prerequisite for a fluent and smooth process is that the computational effort for the individual frames is more or less identical and that the cards involved should also be roughly equal in terms of performance. It doesn’t make sense to combine fast and weak cards, because the commonly held belief that the stronger card will match the level of the weaker one is unfortunately not true. The balancing shifts extremely and the result quickly becomes unpalatable.
Let’s look at our very simplified graph for a system with 2 graphics cards. Card 1 renders the first frame, the second card already renders the second frame in parallel. After that, card 1 takes over the third frame and card 2 the fourth… The result of card 2 ends up each time in the buffer of card 1, to which the monitor is also connected. The whole principle works well as long as the work for both cards is kept within identical limits. However, even with two cards, the transfer and buffering of the data place quite high demands on the constancy of the processing speed, and the synchronicity of the output is thus constantly hanging by a thread.
Solutions with 3 or 4 GPUs work similarly; here, too, the calculations are performed “in turn”, so to speak. Theoretical scalings of 100% increase in dual-GPU systems are hardly achievable in practice and are often enough bought by asynchronous rendering behavior, which then leads to micro-stutters as a consequence, which we will discuss in more detail in the next section. Thus, you have to carefully consider which driver-internal balancing you prefer and whether you rather aim for frame rates or consistency.
- 1 - Einführung und Übersicht
- 2 - Wir funktionieren Crossfire und SLI?
- 3 - Mikroruckler - so klein und doch so ärgerlich
- 4 - Stufe 1: CrossfireX mit 2 GPUs
- 5 - Stufe 1: CrossfireX mit 2 GPUs versus SLI
- 6 - Stufe 2: CrossfireX mit 3 GPUs
- 7 - Stufe 3: CrossfireX mit 4 GPUs versus SLI
- 8 - Testsystem und Benchmarkauswahl
- 9 - Benchmarks: 3DMark11 und Sanctuary
- 10 - Benchmarks: DirectX 11 Spiele
- 11 - Benchmarks: DirectX 10 Spiele
- 12 - Benchmarks: DirectX 9 Spiele
- 13 - Leistungsindex und Leistungsaufnahme
- 14 - Zusammenfassung und Fazit
9 Antworten
Kommentar
Lade neue Kommentare
Mitglied
Urgestein
Veteran
Urgestein
Urgestein
Mitglied
Veteran
Urgestein
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →