Forums
New posts
Articles
Product Reviews
Policies
FAQ
Log in
Register
What's new
Search
Search
Search titles only
By:
New posts
Menu
Log in
Register
Install the app
Install
Forums
General Discussions
Switcher Hangout (Windows to Mac)
Benchmark Tests?
JavaScript is disabled. For a better experience, please enable JavaScript in your browser before proceeding.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an
alternative browser
.
Reply to thread
Message
<blockquote data-quote="mmoy" data-source="post: 95894" data-attributes="member: 9872"><p>With the very complex architectures today, it is very hard to do an apples to apples comparison because so much depends on software today. If you wrote a simple program to add numbers from 1 to a million, it's likely that the machine with the highest clock</p><p>rate would win. But a programer could use the Altivec unit to do the additions much,</p><p>much faster or a programmer could use the SSE2 units of Pentium 4 and Athlon64 units</p><p>to do the same thing. The Altivec unit has more registers but could a programmer keep</p><p>them all busy?</p><p></p><p>The Apple world appears to assume that application developers will do some Altivec</p><p>tuning of their applications and Apple is pretty supportive of using Vector instructions</p><p>in their Apple Develoer web pages. There is much less support on the Intel/AMD side</p><p>and far fewer programs are optimized using the vector instructions there. Programmer</p><p>support is so-so as Microsoft and the Gnu folks have different syntax for using</p><p>vector instructions so you don't have standardization similar to what Apple has (if</p><p>you only have one compiler, then there's a defacto standard).</p><p></p><p>One of those rating sites shows a PowerMac 1.8 (single) as being a bit slower than</p><p>the Athlon64 3200+. I have both machines and the Athlon64 is more than a little</p><p>faster than the PowerMac G5. But I have the Athlon64 pretty well tuned.</p><p></p><p>For the responsiveness of my usual desktop applications, though, the difference in speed doesn't really matter. It does on builds with PowerMac builds taking well over</p><p>an hour and Athlon64 builds taking about 40 minutes. But I consider those background</p><p>tasks so the time to completion doesn't really matter. Both machines do a very good</p><p>job of not letting long-running background tasks affect foreground task performance.</p><p></p><p>What I've heard from the science/engineering types that need to crunch a lot of single-precision floating point numbers is that you can't beat Altivec for that kind of application. The interesting thing is that the Pentium 4/Athlon 64 systems also support</p><p>double-precision floating point math in their vector processors while the PowerPC</p><p>only supports scalar double-precision math via f87. Pentium 4/Athlon 64 also supports</p><p>double-prevision in f87. But I think that most heavy-duty number crunching is done</p><p>in single-precision.</p><p></p><p>The Pentium 4 architecture seems to me to be better at handling applications that</p><p>work on bytes instead of words. Maybe that's a CISC vs RISC thing. I note that the</p><p>pixel scheme on the Mac tends to be four bytes while it tends to be three bytes on Windows. Four bytes lends iteself to better alignment and better use of vector instructions. To use vector instructions with three-byte pixels requires a lot of extra</p><p>shifting and masking. Vector instructions generally work with objects that are sized as</p><p>powers of two.</p><p></p><p>At any rate, I've done a lot of assembler programming in x86 and am now doing the same for PowerPC so I'm sure that my opinions on various things will change with</p><p>experience.</p></blockquote><p></p>
[QUOTE="mmoy, post: 95894, member: 9872"] With the very complex architectures today, it is very hard to do an apples to apples comparison because so much depends on software today. If you wrote a simple program to add numbers from 1 to a million, it's likely that the machine with the highest clock rate would win. But a programer could use the Altivec unit to do the additions much, much faster or a programmer could use the SSE2 units of Pentium 4 and Athlon64 units to do the same thing. The Altivec unit has more registers but could a programmer keep them all busy? The Apple world appears to assume that application developers will do some Altivec tuning of their applications and Apple is pretty supportive of using Vector instructions in their Apple Develoer web pages. There is much less support on the Intel/AMD side and far fewer programs are optimized using the vector instructions there. Programmer support is so-so as Microsoft and the Gnu folks have different syntax for using vector instructions so you don't have standardization similar to what Apple has (if you only have one compiler, then there's a defacto standard). One of those rating sites shows a PowerMac 1.8 (single) as being a bit slower than the Athlon64 3200+. I have both machines and the Athlon64 is more than a little faster than the PowerMac G5. But I have the Athlon64 pretty well tuned. For the responsiveness of my usual desktop applications, though, the difference in speed doesn't really matter. It does on builds with PowerMac builds taking well over an hour and Athlon64 builds taking about 40 minutes. But I consider those background tasks so the time to completion doesn't really matter. Both machines do a very good job of not letting long-running background tasks affect foreground task performance. What I've heard from the science/engineering types that need to crunch a lot of single-precision floating point numbers is that you can't beat Altivec for that kind of application. The interesting thing is that the Pentium 4/Athlon 64 systems also support double-precision floating point math in their vector processors while the PowerPC only supports scalar double-precision math via f87. Pentium 4/Athlon 64 also supports double-prevision in f87. But I think that most heavy-duty number crunching is done in single-precision. The Pentium 4 architecture seems to me to be better at handling applications that work on bytes instead of words. Maybe that's a CISC vs RISC thing. I note that the pixel scheme on the Mac tends to be four bytes while it tends to be three bytes on Windows. Four bytes lends iteself to better alignment and better use of vector instructions. To use vector instructions with three-byte pixels requires a lot of extra shifting and masking. Vector instructions generally work with objects that are sized as powers of two. At any rate, I've done a lot of assembler programming in x86 and am now doing the same for PowerPC so I'm sure that my opinions on various things will change with experience. [/QUOTE]
Verification
Post reply
Forums
General Discussions
Switcher Hangout (Windows to Mac)
Benchmark Tests?
Top