I figure there are enough sci-fi geeks here to appreciate this (original by me):
"There is nothing wrong with your election. Do not attempt to adjust your vote. We are controlling everything. If we wish to support a candidate, we will increase the votes. If we wish to defeat a candidate, we will tune them out. We will control the House. We will control the Senate. We can roll the Courts, making them useless. We can reduce the issues to a soft blur or sharpen them to crystal clarity. For the next four years sit quietly and we will control all that you see and hear. We repeat: there is nothing wrong with your election. You are about to participate in a great adventure. You are about to experience the awe and mystery which reaches from your life to... The Diebold Limits."
Chris (who knows too much about computers and opted to vote on paper)
Huh?
What does Intel's "approximate" math library have to do with Mathematica?
Sure, Intel got some good speedups by using less precise math. I doubt that Mathematica could use that because of the loss of precision.
But that doesn't mean that Mathematica cannot be sped up.
Have you written any MMX/SSE or AltiVec code? Do you know how many things they can really do?
Almost any math that has SOME parallelism in it can be sped up with the MMX/SSE or AltiVec vector instructions. Lots of data oraginzation operations (interleave, de-interleave, 24bit->16bit pixels, transpose, rotate90, flip horizontally, etc.) can also be sped up.
Plus SSE and AltiVec include some powerful cache hinting instructions that can be applied regardless of the vector instructions.
I'm pretty sure that several important parts of Mathematica could be sped up with MMX/SSE and AltiVec instructions.
Actually, Mathematica could use AltiVec for many operations.
AltiVec does NOT replace graphics cards -- it's additional instructions comparable to MMX/SSE/SSE2 (sans the double precision).
Anywhere you've got some parallelism in the calculations (and don't need double precision), AltiVec is probably going to be useful.
Chris
And how exactly is Adobe "Apple's bitch"?
And how do you know that Photoshop is "written to scream on Macs", and not written to perform best on EVERY platform as Adobe, Intel, AMD, Apple and most magazines claim?
Oh, and I guess to be fair you'll want to disable all MMX and SSE optimizations when benchmarking x86 processors as well? Have you tried that? Have you seen the results?
BTW - you seem to have missed a number of Apple's benchamrks/demos run with the SETI client, DNET client, some large-scale (does a galaxy count as "large-scale") simulations, and a few cluster systems.
Mr. Carmack;
Just how exactly have you optimized the code for both platforms? I mean, you're famous for shipping applications with highly optimized x86 code, and near debug quality PPC code.
Speaking as someone who has optimized a major application for both platforms: PowerPC wins in a fair test, for integer and floating point. (I have to admit, the just announced AMD dual 1.2 and single 1.4 Ghz systems do edge out the G4/733 on speed - but not by as much as most people think)
As for memory systems -- have you really tried them? The only things that a P4 system can do faster in main memory than a G4 system are memcpy, memset and memcmp. Anything more complicated bogs down the system and the G4 flies ahead. P3 was worse, with RDRAM or SDRAM. Even the AMD DDR systems aren't as fast as Apple's G4 with PC133.
As for the implementations in Photoshop -- call me when you're in San Jose. In some cases it's identical C code (with full optimizations enabled), and in many common cases it's highly optimized for each platform. Oh, and Intel has worked extensively on the x86 code.
Chris Cox
Slashdot Mirror
I figure there are enough sci-fi geeks here to appreciate this (original by me):
"There is nothing wrong with your election. Do not attempt to adjust your vote. We are controlling everything. If we wish to support a candidate, we will increase the votes. If we wish to defeat a candidate, we will tune them out. We will control the House. We will control the Senate. We can roll the Courts, making them useless. We can reduce the issues to a soft blur or sharpen them to crystal clarity. For the next four years sit quietly and we will control all that you see and hear. We repeat: there is nothing wrong with your election. You are about to participate in a great adventure. You are about to experience the awe and mystery which reaches from your life to... The Diebold Limits."
Chris
(who knows too much about computers and opted to vote on paper)
Huh? What does Intel's "approximate" math library have to do with Mathematica? Sure, Intel got some good speedups by using less precise math. I doubt that Mathematica could use that because of the loss of precision. But that doesn't mean that Mathematica cannot be sped up.
Have you written any MMX/SSE or AltiVec code? Do you know how many things they can really do? Almost any math that has SOME parallelism in it can be sped up with the MMX/SSE or AltiVec vector instructions. Lots of data oraginzation operations (interleave, de-interleave, 24bit->16bit pixels, transpose, rotate90, flip horizontally, etc.) can also be sped up. Plus SSE and AltiVec include some powerful cache hinting instructions that can be applied regardless of the vector instructions. I'm pretty sure that several important parts of Mathematica could be sped up with MMX/SSE and AltiVec instructions.
Why did you post this garbage twice? It was wrong the first time.....
Actually, Mathematica could use AltiVec for many operations. AltiVec does NOT replace graphics cards -- it's additional instructions comparable to MMX/SSE/SSE2 (sans the double precision). Anywhere you've got some parallelism in the calculations (and don't need double precision), AltiVec is probably going to be useful. Chris
And how exactly is Adobe "Apple's bitch"? And how do you know that Photoshop is "written to scream on Macs", and not written to perform best on EVERY platform as Adobe, Intel, AMD, Apple and most magazines claim? Oh, and I guess to be fair you'll want to disable all MMX and SSE optimizations when benchmarking x86 processors as well? Have you tried that? Have you seen the results? BTW - you seem to have missed a number of Apple's benchamrks/demos run with the SETI client, DNET client, some large-scale (does a galaxy count as "large-scale") simulations, and a few cluster systems.
Mr. Carmack; Just how exactly have you optimized the code for both platforms? I mean, you're famous for shipping applications with highly optimized x86 code, and near debug quality PPC code. Speaking as someone who has optimized a major application for both platforms: PowerPC wins in a fair test, for integer and floating point. (I have to admit, the just announced AMD dual 1.2 and single 1.4 Ghz systems do edge out the G4/733 on speed - but not by as much as most people think) As for memory systems -- have you really tried them? The only things that a P4 system can do faster in main memory than a G4 system are memcpy, memset and memcmp. Anything more complicated bogs down the system and the G4 flies ahead. P3 was worse, with RDRAM or SDRAM. Even the AMD DDR systems aren't as fast as Apple's G4 with PC133. As for the implementations in Photoshop -- call me when you're in San Jose. In some cases it's identical C code (with full optimizations enabled), and in many common cases it's highly optimized for each platform. Oh, and Intel has worked extensively on the x86 code. Chris Cox