Ethic-wise, has AMD done anything wrong?

Discussion in 'General Hardware' started by Espionage724, Nov 4, 2013.

  1. sykozis

    sykozis Ancient Guru

    Messages:
    21,801
    Likes Received:
    1,056
    GPU:
    MSI RX5700
    Even a complete rewrite to utilize SSE, will leave GPUs with an advantage. It's still going to be heavily floating point intensive....which GPU's handle quite well. The difference will be that PhysX performance on CPUs won't be artificially gimped. Instead, PhysX performance will be based on the performance of the processor itself.

    It's only partially BS. PhysX was intentionally done in x87 for 2 reasons.
    First, there was no associated licensing for the use of x87.
    Second, it's a deprecated instruction set that doesn't run at full speed on CPU but can easily be handled by a specialized chip (with greater performance results).

    It wasn't NVidia's doing by any means but they have left it running x87 instructions out of necessity (more or less). They couldn't just buy up PhysX and immediately drop support for it while rewriting the entire API to bring it up to date. Personally, I'm impressed with how NVidia has handled PhysX (to an extent). They've continually improved the current, x87 based API while working to rewrite the entire API to use x86/SSE instructions.

    It's not BS that PhysX performance on CPU is crippled by the x87 instruction set. It's only BS to blame NVidia for it. Reportedly, NVidia started working on an x86/SSE based rewrite of PhysX a couple years ago to bring it up to date with current instruction sets.

    x87 refers to the instruction set explicitly used by the PhysX API. The x87 instruction set is (and has always been) separate from the x86 instruction set (to the point of once requiring a separate "co-processor"). The x87 instruction set was superseded by the SSE instruction sets.

    That article isn't entirely accurate. Only the 486DX models supported the x87 instruction set natively. The SX model processors required a "math co-processor" for x87 support. (the math co-processor, coincidentally, was also commonly refered to as an "x87 math co-processor" where the X represented the processor generation.)

    Intel deprecated the x87 instruction set more than a decade ago.(late 90's actually)...

    80486 predated PhysX by more than 20 years.....lol
     
    Last edited: Nov 7, 2013
  2. mbk1969

    mbk1969 Ancient Guru

    Messages:
    10,801
    Likes Received:
    8,048
    GPU:
    GF RTX 2070 Super
    Do you mean that floating point instructions not implemented by modern CPUs? And if they are implemented then can anyone assume that their performance is at the level of 486 ones?
     

Share This Page