CPU archtecture getting rid of FPU

[knutinh]

I've been thinking about CPU architecture.

Why wouldn't we get rid of the FPU and use larger ALU for example 532 bit ALU that can basically get rid of FPU and do pure integer math.

Do we really need more than 512 bits?

I think that you underestimate how much SIMD is used for number crunching. Either because the application programmer used assembler/intrinsics/a vectorizingncompiler (intel) or because they rely on some library (blas, fftw,...) that is vectorized.

If we expect our hardware to do function «A» is it not intuitively most effcicient to implement A rather than using some proxy?

-k

 

[Tuna-Fish]

Not only that but have you actually tried to write a SIMD program?

Quite a few, actually. I'm personally drawn to low-level bit twiddling, and SIMD is one of the very few places where that is still worth doing today.

Note that in my post I was not defending SIMD -- I personally think that Intel has taken the allocation of resources to SIMD too far with AVX-512, and that incremental improvements in SIMD throughput don't really help all that much today. I was merely explaining why the FPU complex (really just the SIMD execution units, they have integer too, after all) is so large and power-hungry even though the FP operations themselves are not really any more demanding than integer ones.

 

[thebcomputerguy]

Just freaking stop. 64bit Windows requires SIMD. Period. News flash - SIMD isn't just floating point.

Here's a clue - you haven't had some revelation that every computer scientist over the last 60+ years has missed.

This is nothing but a troll thread.

Attacking other posters is not allowed.
Markfw
Anandtech Moderator

A troll thread?

Instead of thinking of the possibilities and engineering around the flaws, the mod says I'm trolling.

I wouldn't think a forum like this would be so small minded but instead of trying to overcome big challenges people want to cower in fear.

Enjoy the fruits for paying more for less as two CPU manufacturers eventually turn into 1.

 

[Tuna-Fish]

Instead of thinking of the possibilities

and engineering around the flaws

The problem is that the main flaw is the very idea. There is nothing of value in it.

From the first page:

Sure such large ALU will require more power but the FPU on most CPU use 4x the power of the integer part plus a whole lot of additional complexity.

This betrays that you think that FP computation requires more power and complexity than integer. This is not true. SIMD requires more power and complexity than integer, FP is an *optimization* that requires less power and complexity than an integer implementation that solves the same problem. On x86, SIMD and FP are conceptually mixed, so it's easy to think that FP is power-hungry and complex. It isn't.

 

[IntelUser2000]

I wouldn't think a forum like this would be so small minded but instead of trying to overcome big challenges people want to cower in fear.

Enjoy the fruits for paying more for less as two CPU manufacturers eventually turn into 1.

You are not an engineer though. And even if you were, us acknowledging that you are right isn't going to prevent what you say "two CPU manufacturers eventually turn into 1". Basic market economics will do that.