Maybe we don't need it, maybe we do
Well... it doesn't seem very promising to me for complex numerical calculations:
1. At the moment I understand that normal numerical co-processors don't seem to support it. So complex calculations will be slower.
2. Non-standard length of 29 bits... looks weird to me. I am really looking forward to 32 or 64 bit floats with full co-processor acceleration.
3. It fails just the same way as double does. E.g. I have an infinite integral in my dissertation consisting of 2 almost equal "halves" with opposite signs and it's the difference I need which occasionally might be 10 to 30 digits below least significant digit of extended type. unum doesn't seem to be able warn me when I'm getting an absolutely wrong result due to accuracy lack. Variable length float would have been very nice, but unum is fixed
So... I see it simply as something a bit more accurate than extended type but significantly slower and memory-inefficient.
Maybe I'm missing something...