Forum OpenACS Q&A: AMD .vs Intel; GCC et all

Collapse
Posted by Louis Gabriel on
Hi,

Anyone have thoughts on the AMD .vs Intel CPUs?

I've been thinking about building a new computer.  I've never had a
system with an AMD CPU....yet.  I may press the new box into service
as a "home" or personal server.

The factor I've been wondering most about is GCC.  I've read that the
compiled code is much better optimized for the Pentium III than the
AMD Athlon; that GCC doesn't really optimize for the AMD chips much at
all.  I don't know how much a performance factor compiling from source
with the GCC on the Athlon .vs Pentium would be.  Does anyone?

Anyone have input on this issue?

Also, I would enjoy reading any thoughts you care to share on the
various chipset, motherboard and memory options for a Linux developer
or server box.  Stability and reliability are what I value
highest...but that's just me.... how about you?  :o)

(I love this stuff!)

Thanks,

Louis

Collapse
Posted by Robert Ezman on
The first time I read your question I thought that the following link had info on exactly what you were asking. Then I realised the twist. You're not asking about compiler performance but rather compiled code performance.

I figured you might be interested in the story anyway. I got the link from slashdot last night. It's a comparison of multi and single processor AMD and Intel machines. One of the benchmarks is building the Linux kernel. (i.e. performance of GCC)

http://www.gamepc.com/reviews/printreview.asp?review=asmpl&mscssid=&tp=
Collapse
Posted by Andrew Spencer on
Actually, in a sense I believe the above link does answer Louis'
question.  Is not GCC compiled with GCC?  Even if GCC does not compile
100% efficient AMD code, the results at the above URL show that
(running GCC compiling the Linux kernel) an AMD chip outperforms it's
comparable Intel counterpart.
Collapse
Posted by Edmund Lian on
Clock-for-clock, an AMD Athlon is generally faster regardless of whether the app is integer or floating point intensive. In fact, an Athlon is faster even if the Intel CPU is clocked at a higher rate. Here, I'm referring to a P3.

Turns out that Athlons are faster than the new P4 too, even if the latter is clocked at a higher speed. It could be partially because of the lack of GCC optimizations for the P4. See http://www.linuxhardware.org/article.pl?sid=01/07/13/145249 for more info.

Collapse
Posted by Sam Snow on
A good source for hardware reviews is http://www.anandtech.com/ -- and they have two articles talking about how their server setup has changed over the years. Interestingly, their current setup is mostly Athlons: http://www.anandtech.com/IT/showdoc.html?i=1456
Collapse
Posted by Edmund Lian on
<i>A good source for hardware reviews is http://www.anandtech.com/</i>
<br>
<br>
Yeah, but take what is written with a pince of salt. Anand does well at benchmarking, but his lack of engineering background shows through now and then--he is absolutely, absolutely wrong on occasion.<br>
<br>
Tom at <a href="http://www.tomshardware.com">www.tomshardware.com</a> and <a href="http://www.sharkeyextreme.com">www.sharkeyextreme.com</a> are generally much more reliable and accurate. They don't over as many boards etc. though.
Collapse
Posted by Pascal Scheffers on
I've been running AMD hardware since 1994 or something. The only problem AMD used to have was floating-point speed. Games suffered, so Celerons sold much better than AMD-K5/6's. However, in my experience, I've always had faster 'desktop' (read Word, database, etc) performance. The new Athlon/Duron cores do have very efficient FPUs, so games run excelent on AMD hardware (if that is your poison). A part of the reason for that is the cheaper CPU allows you to buy more RAM. I had twice as much as my friends :)

On the remark on Athlon out performing the P4 at the same speed, that is mainly processor design, the P4 has a 20 steps deep pipeline, the Athlon has 7 or 8. A deeper pipeline gives you higher clocks, but also a higher price when a branch prediction failes.

If you really want to see good writing on CPUs go to arstechnica.com, 'Hanibal' knows what he is talking about.

Collapse
Posted by Edmund Lian on
Oh yes, I did forget about Arstechnica... that is a site, thanks. I didn't mention the causes of the P4 throughput issues because there are many, and it's a bit much to get into here...
Collapse
Posted by Edmund Lian on
"that is a site" --> "that is a great site" Oops.