Some Fresh Radeon EXA/GLAMOR 2D Acceleration Benchmarks

[tomtomme]

Because he is. He didn't compare the two technologies at all on that graph.

learntoreadplease

[Ardje]

GLAMOR, SNA, EXA, XAA...
Am I the only person seeing it stupid to reinvent 2d acceleration architecture once in a while?

We used to have vector buffers. (Direct control of the crt beam, *NO* pixels).
Then we used to have framebuffers. (Pixelizing the CRT)
Then we used to have text buffers that indexed into font roms.
Then we used to have text buffers that indexed into font ram(!)
Then we used to have framebuffers again with the ability to do raster operations in hardware. (I had a tektronix serial terminal with a 50cm x 50cm graphics accelerator build with TTL).
Then we used to have framebuffers with overlay buffers. (Sprites)
Then we used to have framebuffers with full 2d operations (font blitting full color pixels).
Then we used to have framebuffers with a full dedicated CPU (TIGA anyone? I had a TMS34020(!) with 2MB VRAM and 12MB DRAM)
Then we used to have add-on overlay cards with YUV buffers.
Then we used to have full add-on cards with dedicated CPU's programmed in microcode (that means: no pipeline stalls, you or the compiler has to figure that out, intel actionmedia cards anyone?).
Then we got real vga cards with 2D acceleration.
Then we got 3d add on cards. (Voodoo 1, or the add on boards in the SGI onyx)
Then we got vga cards with 2D and a 3D add on gpu.
Then we got vga cards with 2D and a 3D add on gpu and YUV overlay buffers
+video decoder
+YUV texture
-YUV overlay
-2D accelerator

That leaves us with the current generation: general purpose GPU's used for 3D rendering, now also used for 2D rendering/acceleration, and a bitstream decoder/encoder. No YUV overlays anymore on current PC hardware.
Arm based hardware usually have about 4 overlay buffers with scaling and alpha blending and different color spaces. Some have 2D acceleration implemented as v4l2 m2m devices.

So yeah, it really changes a lot.
We also used to have multiple graphical seats on a single computer, because the computer was to expensive, but the seats were not. Graphical cards used to have full memory protection, then came the PC and protection was a thing of the past. And now even on arm the cards are all behind IO MMU's.

[Daktyl198]

learntoreadplease

Two sets of cards running two sets of 2D acceleration techniques, with no overlap is not a comparison of EXA and GLAMOUR, it's a comparison of the cards in their default configs.
If he had ran EXA and GLAMOR benchmarks on the same set of cards, then it would be a comparison. He's done it before, I don't know why he didn't do it now.

I'm surprised I have to point out something so simple.

[tomtomme]

Two sets of cards running two sets of 2D acceleration techniques, with no overlap is not a comparison of EXA and GLAMOUR, it's a comparison of the cards in their default configs.
If he had ran EXA and GLAMOR benchmarks on the same set of cards, then it would be a comparison. He's done it before, I don't know why he didn't do it now.

I'm surprised I have to point out something so simple.

Ok I got your point.

However the benchmark for me is still informative. Because the underlying hardware should be capable enough of handling any 2d workload in 2016. And if I remember correctly, there is not even a dedicated 2d-hardware part since GCN. Thus only glamour is possible on gcn.
The comparison that you want is r600-gpus with exa and then with glamour and this comparison is interesting for people having this type of gpu. I have a GCN gpu and am quite happy to know now that the 2d experience a big difference compared to the one on old exa cards.
Another interesting comparison would be with intel SNA.