SpokenTruth said:
But is that a likely situation to ever happen? |
Sure. For specialized tasks and markets like industrial where they need Integer and not floating point capability.
There have been CPU core designs which gave higher preference to integer capability over floating point, which gave them big advantages in scientific or other specialized workloads that rely on integer capability.
AMD's Bulldozer CPU core was such a design where it had two integer units to every floating point unit... It meant that in Integer based tasks it was very competitive with Xeon, but in floating point tasks it was a laughing stock.
And before the 486SX, CPU's didn't do floating point at all, they relied on a separate chip for that, provided they even came with one.
Cyrix M1 had Integer capability that would beat the pants off Intel's Pentium at the time, but fell short in floating point capability to the level of a 486 with a decent co processor, which was fine back then as most tasks relied heavily on Integers anyway.
CPU designs without any floating point capability will often fall back to fixed point arithmetic operations that run on the integer unit anyway, so compatibility is retained even if the performance takes a nose dive.
Intel has some controllers which don't have a floating point unit, like their Quark line.
https://ark.intel.com/content/www/us/en/ark/products/86826/intel-quark-microcontroller-d1000.html
The Vortex86 is another such chip that preferences integer over floating point.
In the GPU space, GPU's tended to use Integer over Floating Point for a very long time until FP32 got cheap enough to implement, in some specialized markets Integer pipelines on GPU's are still used as size and power consumption is more important.
So again, Floating Point is just as useless as using bits in determining a systems capabilities, you neglect everything else that goes into a chip by using just floating point numbers.
--::{PC Gaming Master Race}::--
