The big graphics card review: from GTX 465 and HD 5830 to high-end
We announced every competitor, got over usage and warming measurements, so tests may come finally!
As we used to start with Vantage:
PhysX was not active
Graph is very nice but there is no surprise. Points are getting higher as we raise cathegories and multi-card systems are scaled very well too. We got matches: GTX 465 against HD 5830, GTX 470 against HD 5850 and GTX 480 fights with HD 5870. Except GTX 480 Radeons was more powerful.
We ran Unigine Heaven’s most fresh version and setted the most agressive settings. Radeons did not like that maybe extreme testellation catch them. GeForces won every matchups, Fermi is better with this argument in every cathegory.
You know DirectCompute test application, but we searched fresh version of it now and got interesting results. The most striking feature is software can measuring OpenCL too. NVIDIA’s cards have problem with DirectCompute modul because it could not start two times and if it could it producted the same result in every controller which is a joke against Radeons. There was not everything ok in red side neither, HD 5830 had suspicious weak result and CrossFireX was not supported (SLI neither). Things went in its way at OpenCL but not fully. We did not get result in the same two case which is not stunning, application could not start. It worked with two GTX 470 but CFX did not. Only good news is more powerful cards had better results than smaller teammates.
We use ComputeMark first time and it has the fresh 2.1 version at testing time. In case of Radeons happened what we waited, scores improves with stronger models nicely, stepwise and CrossFireX worked fine too. We can not say this about SLI because application do not support or like it. Radeons win every cathegory against GeForces again.
That is clear by above to applications: measurement of DirectCompute and OpenCL is not an easy job nowadays, we have problems but results are not authentical a couple of time.
We will arive that level, power of graphic cards are not only for graphical processings its tool are DirectCompute and OpenCL or specific NVIDIA CUDA and ATI Stream which can used for video converting. CyberLink MediaShow supports the latter two technology so we can equal efficiency to each other.
We have to say now measurement are not simple at DirectCompute and OpenCL neither and it can be noticed in results. Sometimes a better card works poorer than smaller brothers and we can not profit multi-cards systems. After repeating we did not get same results in the same configuration so these measurements just informative. Now Fermi + CUDA duo looks more effective, even we speak about CUDA or Stream technology have grounds because everytime we finished sooner converting FULL HD to 720p MP4 than we used only processor.
