1. At some point, the order changes, Catalyst moves position to between the two mesa results, which is confusing.
2. Why not have vertical indicators at 30, 60, 120 ... fps instead of automatically determined by the results? From my point of view 30 is ok, 60 is the maximum for most displays, and more than 120 is uninteresting in practice.
Also, why not put the Xonotic results directly after Nexuiz if you refer to them anyway?
Originally Posted by enrico.tagliavini
I wonder if enabling S3TC might improve the performance of the radeon driver. Well this can be an idea for the next article
It does, in applications that enable texture compression the fps are higher when libtxc-dxtn is installed.
Thanks for this benchmark. The problem is that I already knew the results. The OSS driver performs bad and misses plenty of features. Catalyst driver performs good for 3D but is a piece of crap under all the other aspects.
Under Linux you can never have 1 thing that works good for all scenarios. You have to choose between 2 or more things depending on your needs and will never be 100% satisfied because one day, soon or later, you will surely need that missing feature that only the other choice has.
So it's pretty much the same old story: despite all the effort spent tweaking the OSS drivers, they're still running at a tiny fraction of the speed of the closed-source binary blobs. Other than supporting older chipsets that Catalyst has put out to pasture, it's difficult to put a positive spin on this.
So? The main problem seems to me to be that there's about 3 developers, 5 managers and 25 lawyers working on this.
Actually 4 developers, a fraction of a manager and no lawyers but don't let me slow you down
IIRC the main differences between radeon and nouveau in terms of performance progress are :
1. the NVidia hardware equivalent of two performance improving features (what we call tiling and HyperZ) are implemented in such a way that "turning them on" requires less driver work than doing the same on ATI/AMD hardware. Jerome and others been making great progress here recently.
2. the VLIW shader core probably requires relatively more shader compiler effort to get to comparable levels of HW utilization. I'm not sure this is an issue but we should know more once the SI open source driver starts to run beneficially.