[Manuel Calavera]

Ne čitaš temu - to je linkano već nekoliko puta (ja sam bar 2 put stavio), ali test ima problem - ne pokazuje frametime (koji se ne mora odraziti na fps-u).

Dobra je vijest što su ljudi zasad zadovoljni (barem u igrama koje igraju), ali neki se žale na microstuttering (iako je fps odličan)

Još jednom, pa neću više jer sam već dosadan sam sebi, prevedeno sa www.pcgameshardware.de (Total-War-Attila) - bolje prevodi sa ger na eng, tko zna ger bolje nek čita orginal. Grafove je lijepio Nybelung na prošloj stranici http://forum.pcekspert.com/showpost....postcount=1350

"In fact, the GTX 970 behaves striking." A special role, however, takes the GTX 970 with its two-pronged an attached storage. In our benchmark Total War is: Attila the 3.5 GiByte, ie precisely the limit at which it could be critical to the 970 under certain circumstances. And indeed, the GTX 970 behaves striking when compared directly with other similarly powerful graphics cards: When scrolling occurs in our scenario a noticeable micro stuttering on that strikes even when looking at a single locomotive unit: This moves despite ( at least initially) high frame rates are not uniform liquid ahead, but in tiny cracks. A look at the birds buzzing around reveals the tiny Stocker by uneven movement.

A look at the frametimes shows a very clear picture, apparently could Nvidia drivers for the GTX 970 has not yet been adapted, because the 3.5-GiByte problem shows up on the paper in all its "glory". In practice, the problem is noticeable, however only if one pays close attention to the regularity of the output frame. As extreme as restless as suggested by the frametimes, the image will not affect the game. The drop in performance at the end of the measurement is to be noted, however, already clearly has 970 well-Fi is their slow-tailed 0.5 GiByte part the GTX, which manifests itself in a powerful performance loss. This falls out so strong that it is also in our benchmarks, a second measurement after a first leaves against this break the fps by almost half - already in 1080p without anti-aliasing, mind you. In the frame time comparison, the behavior of the map to better estimate the image in practice, notably not quite as dramatic fails. The problem with collapsing frame rate and memory bucking show all the way GPUs with only 3 GiByte or less memory and GTX 780 Ti or R9 280X break sharply after the first measurement. We have therefore started after each measurement the game from scratch.
Total War: Attila - Frame-time measurements and classical benchmarks


Frame Time graph you will get more frequent now, moreover, at our face. These measurements show small deviations in the output image much better than classical benchmarks are able to do. This is especially true for the often tiny Stocker, which may occur through lack of memory. Frametimes to show how much time elapses between two successive frames. This allows much more precise conclusions, as the fluctuations of the frame rate of one second to the next - many small stuttering only take a few milliseconds and do not show up in measurements of the frame rate - but certainly in the frametimes with which also explain the perceived performance leaves.

In the next issue PCGH 04/2015 as we use frametimes to explain the memory problems of the GTX 970, in another article, it turns the other hand, to the main memory: How much RAM for optimum play is needed and how and whether additional memory positively to the feel effect. Without frame time measurements, the differences in these articles would be fine in order to communicate with classical benchmarks. Since the memory and its management may gain importance in several respects again: There would be driven by rising resolutions that should be found in the nearest future a significantly higher proliferation, new and potentially at least memory demanding effects such as global illumination, plus possibly with Direct X 12 is a more direct access to the graphics card memory by the game developers. The latter is also a possible source of error that could have a strong impact on the performance and a review would be worth in any case. Frametimes be so from now on frequently appear in our article. However, traditional benchmarks are to have lasted for frametimes also have several night parts: they are much less reproducible, are unclear and need to be analyzed by the reader to draw conclusions and evaluation of the frame time data is more difficult than in the classical measurements with frames per second. We are currently working to simplify the complicated legal frame time-evaluations and to optimize the time required for the pictures of the graph itself may differ slightly from those in other articles here at least.