CCE Encoding AMD Versus INTEL

[DerekVinyard]

I finally had enough, I'm going to update this PC to speed up CCE's encoding.

How do you find the XP2500 (and faster) hold up to the P4 2.4 (and faster)? I'll be overclocking whatever I purchase and I won't be buying a 3gig processor... Your thoughts would be appreciated.

[nFury8]

This is one of those perennial AMD vs Intel questions that gets thrown all the time. Frankly, if you believe in what the actual product is capable of doing design-wise, off-the -shelf raw power against acquired performance through software optimization and extended instruction sets, then Athlon is your baby. I have to assume that you are well aware of how Intel's 'Netburst' stacks up against pure performance of the Athlon family, IPC vs Mhz madness etc. Its simply a matter of personal conviction, do you want to spend this much amount of money for a ceratin level of performance, or do you believe in putting your money where the real deal is, regardless of popularity and perceived and acquired performance edge. Clock speed is not everything. Your call.

[DerekVinyard]

Since the very first computers, there have only been two options for faster computation. The first being a faster algorithm. The second being a faster computer.
Since the first (CCE) is out of my hands and it's always been easier to throw hardware at the problem, I'm upgrading.

[nFury8]

If I'm not wrong, the P4's are more optimized for CCE than the Athlons, just like many apps out there that were tweaked and re-written to take advantage of the Netburst architecture. You can also try and search the CCE section of this forum to compare actual user experiences with the P4 against the Athlon.

[Joe Fenton]

"Netburst architecture" is just fancy Intel double-speak. All it means is a trace cache for micro-ops, a branch predictor for the trace cache, and an extra long pipeline. "Netburst architecture" is just how Intel markets the chip to the ignorant masses.

"Tweaked or re-written" for "Netburst architecture" is just a convoluted way of saying optimized scheduling for the P4. The Intel compiler knows how x86 instructions will be converted into micro-ops, and how those micro-ops will go through the pipeline. Any given high-level language can be compiled a number of ways: to run faster, but use more space; to run slower, but use less space. The Intel compiler uses it's knowledge of how the x86 codes get translated and generates x86 codes that will get converted into micro-ops that do better on the pipelines and trace cache. This will have VERY little effect on how the code will run on x86 architectures that don't have "netburst" - like the Pentium III or the Athlon. It may have a BIG effect on the P4.

So to dumb down the explanation, optimizing for netburst is to keep the P4 from running SLOWER, not to speed it up. That's why P4's don't compare well to P3's - the P4 requires special optimization to give similar performance. It's only the fact that P4's run at greatly higher clock rates that they perform as well as they do. Optimizing for netburst does little to nothing, either positive or negative, for other architecture CPUs.

Where the major speed difference between Intel code on Intel chips vs. AMD chips is the recently discussed fact that the Intel compiler will prevent AMD chips from using SSE/SSE2 code. The XviD people use their own CPUID code and their own hand-optimized SSE/SSE2 code, so this does not affect them. My guess is that DivX is all Intel compiled code, hence the big difference on P4's vs the AMD64 seen in benchmarks.

[Wilbert]

Where the major speed difference between Intel code on Intel chips vs. AMD chips is the recently discussed fact that the Intel compiler will prevent AMD chips from using SSE/SSE2 code. The XviD people use their own CPUID code and their own hand-optimized SSE/SSE2 code, so this does not affect them. My guess is that DivX is all Intel compiled code, hence the big difference on P4's vs the AMD64 seen in benchmarks.
Interesting! I didn't know about that. Do you have a link to these discussions?

[nFury8]

Originally posted by Joe Fenton
"Netburst architecture" is just fancy Intel double-speak. All it means is a trace cache for micro-ops, a branch predictor for the trace cache, and an extra long pipeline. "Netburst architecture" is just how Intel markets the chip to the ignorant masses.

"Tweaked or re-written" for "Netburst architecture" is just a convoluted way of saying optimized scheduling for the P4. The Intel compiler knows how x86 instructions will be converted into micro-ops, and how those micro-ops will go through the pipeline. Any given high-level language can be compiled a number of ways: to run faster, but use more space; to run slower, but use less space. The Intel compiler uses it's knowledge of how the x86 codes get translated and generates x86 codes that will get converted into micro-ops that do better on the pipelines and trace cache. This will have VERY little effect on how the code will run on x86 architectures that don't have "netburst" - like the Pentium III or the Athlon. It may have a BIG effect on the P4.

So to dumb down the explanation, optimizing for netburst is to keep the P4 from running SLOWER, not to speed it up. That's why P4's don't compare well to P3's - the P4 requires special optimization to give similar performance. It's only the fact that P4's run at greatly higher clock rates that they perform as well as they do. Optimizing for netburst does little to nothing, either positive or negative, for other architecture CPUs.

Where the major speed difference between Intel code on Intel chips vs. AMD chips is the recently discussed fact that the Intel compiler will prevent AMD chips from using SSE/SSE2 code. The XviD people use their own CPUID code and their own hand-optimized SSE/SSE2 code, so this does not affect them. My guess is that DivX is all Intel compiled code, hence the big difference on P4's vs the AMD64 seen in benchmarks.

LOL. You really took me to task on that one. :D Couldn't have said it any better. Nice one.

@Wilbert:
Here's (http://groups.google.ca/groups?dq=&hl=en&lr=&ie=UTF-8&oe=UTF-8&threadm=a13e403a.0402091438.14018f5a%40posting.google.com&rnum=1&prev=/groups%3Fdq%3D%26hl%3Den%26lr%3D%26ie%3DUTF-8%26oe%3DUTF-8%26group%3Dcomp.arch%26selm%3Da13e403a.0402091438.14018f5a%2540posting.google.com) the link you were referring to.

[Joe Fenton]

Originally posted by Wilbert
Interesting! I didn't know about that. Do you have a link to these discussions?

A few links on this:
http://tinyurl.com/27lnn
http://www.theinquirer.net/?article=13821

Edit: nFury8 beat me to the punch... didn't notice his link. :)

I wasn't taking you to task so much as just throwing out some info in the hopes that people reading the thread might learn something.