what do you mean by better color compression? nvidia has been implementing different forms of compression including color compression since the gtx 9000 series. afaik amd implemented color compression with navi.
Color Compression has been around for a good while, on both sides. One specific implementation: Delta Color Compression, has been around since GCN 1.2 (Fiji/Tonga)
gpuopen.com
I stand corrected, but what makes it better than the nvidia solution?Color Compression has been around for a good while, on both sides. One specific implementation: Delta Color Compression, has been around since GCN 1.2 (Fiji/Tonga)
Here's an older supporting document.
Getting the Most Out of Delta Color Compression
DCC is a domain-specific compression that tries to take advantage of data coherence. It’s lossless, and adapted for 3D rendering. The key idea is to process whole blocks instead of individual pixels.
Well, the method at which they were doing lossless compression made it better at the time. They had a higher average compression ratio which meant less bandwidth was required to get the exact same (it's lossless, so results are 100% identical) results. Fast forward a few years.... nvidia now does a better job at compression than AMD and tends to need less bandwidth for a given performance. It's hard to directly measure though. The closest you can get is loading the same application and check memory usage, but even then things like memory management and other driver overhead makes a 1:1 comparison difficult. Add to that architectural differences which make a 1:1 benchmark all but useless (aka, you can't run both memory & GPU at the same speeds from Navi to Turing and expect to be only measuring difference in bandwidth saved by color compression, you're measuring many other things like decoders, branch predictors, cache systems, etc). Back in the day is was a much bigger deal, nowadays, they are all using compression to some extent, but I'm not sure if we know the actual compression ratio's anymore (there used to be tools to actually compress to format required by the GPU so you could see what the original and compressed sizes where for specific images, but I haven't been into graphics programming for a while and don't know (I doubt) if that's still a big thing people would do by hand).
Someone made the point in another thread that AMD is doing a little better than a 2080 Ti in a real ray tracing benchmark was supposedly no dedicated hardware. Could be better, but it allays my concern a little about AMD's solution holding development in ray tracing back.
That quote correction is pretty well true.Fixed
Yeah, AMD needs to be competitive with RT, it would benefit everyone.
To win over people that buy brands out of habit and wont change that until there's financial reasons to do so.
A better card isn't reason enough?
AMD needs a better card and to be at least competitive with everything else.
Yes, I have since corrected them, AMD does have dedicated hardware, please stop the chain of spreading this misinformation. They organized their hardware different than NVIDIA, that doesn't make it non-existent, it just means it's implemented differently. Anyways, I think the general assumption was AMD would be ~2080ti for RT, so the fact that it's slightly ahead is better than most suspected. Most assumed AMD's first try with RT would fall short of nvidia's 2nd attempt. I'm glad it's doing as well as it is, and we have no clue how well optimized anything has been made for AMD hardware (probably ~zero optimizations).
