AMD or Intel : Dual core

There are plenty of reviews that compare HT to dual core, DDR2 to DDR,
price points, AMD vs Intel, etc, but few that really address everything
wrapped together.

Here's my goal:
Purchase a system that can process (multithreaded) 120GB worth of data
in a process similar to dv encoding. That is, crunch, write, repeat.
The entire process is expected to take days, possibly weeks. So heat
is definitely an issue as well. Both cores will be peaked except for
blocked disk IO.

Here's what I want:
Dual core
1MB L2 cache
Fast memory
Fast disk IO (SATA, drives purchased a while ago)

It's pretty much a stalemate between a P4 D 820 and a AMD Athlon X2 4400+
from the reviews I've read. The biggest differences are DDR2 and price
point. The Athlon is selling for $600. The P4 is selling for $240.
And DDR2 sounds pretty attractive right now.

I'm leaning towards the P4. I think the X2 is attracting the gamer
community (not sure why, a single core would do better) and I just
get an oddball feel that the reviews aren't accurate on it.

Any opinions / advice would be welcomed.

TIA
Brian

Brian wrote:
There are plenty of reviews that compare HT to dual core, DDR2 to DDR,
price points, AMD vs Intel, etc, but few that really address everything
wrapped together.

Here's my goal:
Purchase a system that can process (multithreaded) 120GB worth of data
in a process similar to dv encoding. That is, crunch, write, repeat.
The entire process is expected to take days, possibly weeks. So heat
is definitely an issue as well. Both cores will be peaked except for
blocked disk IO.

Here's what I want:
Dual core
1MB L2 cache
Fast memory
Fast disk IO (SATA, drives purchased a while ago)

It's pretty much a stalemate between a P4 D 820 and a AMD Athlon X2 4400+
from the reviews I've read. The biggest differences are DDR2 and price
point. The Athlon is selling for $600. The P4 is selling for $240.
And DDR2 sounds pretty attractive right now.

I'm leaning towards the P4. I think the X2 is attracting the gamer
community (not sure why, a single core would do better) and I just
get an oddball feel that the reviews aren't accurate on it.

Any opinions / advice would be welcomed.

I'd say that by the price alone, the P-D is more attractive. However,
the X2 will be much cooler running. The DDR vs. DDR2 is a non-issue, the
Athlon extracts much better performance out of the DDR than the Pent
does out of the DDR2, because of its built-in memory controller. You
might be able to save some dollars on getting DDR instead of DDR2, but I
doubt it will make up for the difference in price on the processors. The
bigger price of the Athlon is pretty much a matter of you get what you
pay for -- much better engineered, better integration between the cores,
better heat characteristics, etc. However, if you're not looking for
finesse, then the P-D should be fine.

Yousuf Khan

In article , Brian wrote:

Here's my goal:
Purchase a system that can process (multithreaded) 120GB worth of data
in a process similar to dv encoding. That is, crunch, write, repeat.
The entire process is expected to take days, possibly weeks. So heat
is definitely an issue as well. Both cores will be peaked except for
blocked disk IO.

Here's what I want:
Dual core
1MB L2 cache
Fast memory
Fast disk IO (SATA, drives purchased a while ago)

Why are you married to the idea of dual core for this? Myself, I'd go
with a dual Opteron (single core) system. With a good motherboard, you
get the advantage of each CPU having its own dedicated bank of memory plus
a fast HyperTransport link to the other CPU and its memory (some of the
low end dual Opteron boards only have 1 memory bank). AMD's Hypertransport
based boards are a clear winner over Intel stuff for I/O intensive
applications, IMO.

--
Joshua Baker-LePain
Department of Biomedical Engineering
Duke University

Here's my goal:
Purchase a system that can process (multithreaded) 120GB worth of data
in a process similar to dv encoding. That is, crunch, write, repeat.
The entire process is expected to take days, possibly weeks. So heat
is definitely an issue as well. Both cores will be peaked except for
blocked disk IO.

I'm assuming you have a custom application you wrote? If so, are you
compiling it in 64-bit mode or 32-bit mode? This will make all the
difference. If you are insistent on 64-bits, tests have clearly shown
(at least Anand's and a few others that I read months ago) that Athlon
at 64-bits gets a performance improvement while Pentium 4 at 64-bits
suffers a performance degradation. This is probably just a temporary
situation...first generation parts and all. But if you are doing only
32-bit programs, the Pentium 4 outperforms the Athlon in floating point
computation (be sure to use only SSE2/SSE3).

The recommendation is really heavily dependant on which mode and what
computations you are doing. Is your code really well threaded? If not,
then why must you have dual core? If it is, then you'd be even better
off with a small cluster of dual processor Opterons or Xeons. If you
are streaming through 120GB of data, you don't need cache. If you are
repeatedly using the same part of the data, then cache *may* help. If,
performance wise everything is close to equal, then I'd recommend going
with the cheaper part. Perhaps you can tell us more about what is
happening to this 120GB of data so we can make a better guess about
which system might benefit you most....

Alex

Alex Johnson wrote:
I'm assuming you have a custom application you wrote? If so, are you
compiling it in 64-bit mode or 32-bit mode? This will make all the
difference. If you are insistent on 64-bits, tests have clearly shown
(at least Anand's and a few others that I read months ago) that Athlon
at 64-bits gets a performance improvement while Pentium 4 at 64-bits
suffers a performance degradation. This is probably just a temporary
situation...first generation parts and all. But if you are doing only
32-bit programs, the Pentium 4 outperforms the Athlon in floating point
computation (be sure to use only SSE2/SSE3).

You did see the thread about where Intel deliberately turns off all SSE
support on non-Intel processors, even if they have it?

Yousuf Khan

Joshua Baker-LePain wrote:
In article , Brian wrote:
Purchase a system that can process (multithreaded) 120GB worth of data
in a process similar to dv encoding. That is, crunch, write, repeat.
The entire process is expected to take days, possibly weeks. So heat
is definitely an issue as well. Both cores will be peaked except for
blocked disk IO.

Why are you married to the idea of dual core for this? Myself, I'd go
with a dual Opteron (single core) system. With a good motherboard, you
get the advantage of each CPU having its own dedicated bank of memory plus
a fast HyperTransport link to the other CPU and its memory (some of the
low end dual Opteron boards only have 1 memory bank). AMD's Hypertransport
based boards are a clear winner over Intel stuff for I/O intensive
applications, IMO.

If you can afford it and get a dual-memory-bank motherboard, this is clearly
the highest performance option (short of 2x dual core opterons *grin*). I
think once you get two 200-series Opterons and a good motherboard, though,
you're well past the cost of the dual core Socket 939 chip and MB... so it's
more a question of whether the performance is worth the cost.

Depending on how CPU-intensive it is, it sounds likely to be disk bound to
me; if so, and if it's not amenable to FS-level parallelism, a good hardware
RAID controller (say a 3Ware or something) rather than SW RAID may actually
do more for performance than any CPU upgrade.

--
Nate Edel http://www.cubiclehermit.com/

"I do have a cause, though. It is Obscenity. I'm for it." - Tom Lehrer

Yousuf Khan wrote:
You did see the thread about where Intel deliberately turns off all SSE
support on non-Intel processors, even if they have it?

Yousuf Khan

No I didn't see the thread you are refering to. It sounds rediculous.
How can intel do something to you when you aren't using any of their
hardware? Oh, you're using their compiler? It isn't the only game
around. You are asking for it if you use half of one company's products
in conjunction with the other half from their main competitor.

Alex

Yousuf Khan wrote:
Alex Johnson wrote:

I'm assuming you have a custom application you wrote? If so, are you
compiling it in 64-bit mode or 32-bit mode? This will make all the
difference. If you are insistent on 64-bits, tests have clearly shown
(at least Anand's and a few others that I read months ago) that Athlon
at 64-bits gets a performance improvement while Pentium 4 at 64-bits
suffers a performance degradation. This is probably just a temporary
situation...first generation parts and all. But if you are doing only
32-bit programs, the Pentium 4 outperforms the Athlon in floating
point computation (be sure to use only SSE2/SSE3).


You did see the thread about where Intel deliberately turns off all SSE
support on non-Intel processors, even if they have it?

FUD alert! Just how do you think Intel would disable SSE on my AMD chip?
Witchcraft? SSE works when the chip is shipped but Intel disables it in
my computer by some secret technology? Did you read about that in the
AMD "unfair practices" lawsuit, or what?

Sorry, when I read that I just get a vision of a poor Opteron plugged
into a sacrificial motherboard, it's dark, and by the light of the
smokey fire some dude in feathers and a loincloth is mumbling and waving
a dead chicken. Like a flashback from a bad voodoo horror flick. Then a
bolt of lightning comes down and fries the SSE.

Or maybe they stick pins in the SSE portion of a circuit diagram? I
assume you were joking, or meant to post to alt.conspiracy. Please
reassure me that you haven't gone over to the dork side.

--
bill davidsen
SBC/Prodigy Yorktown Heights NY data center
http://newsgroups.news.prodigy.com

Bill Davidsen wrote:
Yousuf Khan wrote:

Alex Johnson wrote:

I'm assuming you have a custom application you wrote? If so, are you
compiling it in 64-bit mode or 32-bit mode? This will make all the
difference. If you are insistent on 64-bits, tests have clearly
shown (at least Anand's and a few others that I read months ago) that
Athlon at 64-bits gets a performance improvement while Pentium 4 at
64-bits suffers a performance degradation. This is probably just a
temporary situation...first generation parts and all. But if you are
doing only 32-bit programs, the Pentium 4 outperforms the Athlon in
floating point computation (be sure to use only SSE2/SSE3).



You did see the thread about where Intel deliberately turns off all
SSE support on non-Intel processors, even if they have it?


FUD alert! Just how do you think Intel would disable SSE on my AMD chip?
Witchcraft? SSE works when the chip is shipped but Intel disables it in
my computer by some secret technology? Did you read about that in the
AMD "unfair practices" lawsuit, or what?

Sorry, when I read that I just get a vision of a poor Opteron plugged
into a sacrificial motherboard, it's dark, and by the light of the
smokey fire some dude in feathers and a loincloth is mumbling and waving
a dead chicken. Like a flashback from a bad voodoo horror flick. Then a
bolt of lightning comes down and fries the SSE.

Or maybe they stick pins in the SSE portion of a circuit diagram? I
assume you were joking, or meant to post to alt.conspiracy. Please
reassure me that you haven't gone over to the dork side.


I don't know for sure, but I think Yousuf might be talking about Intel
*software* not intel hardware. A lot of people use intel's compiler
because it does a great job optimizing code for intel chips. A lot of
people who don't use intel chips use intel's compiler anyway. I can see
a scenario where a compiler asks the chip "what is your name and
revision" and the answer AMD gives back does not match any of the intel
chips for which SSE2 applies, so the compiler doesn't try to use SSE2.
It doesn't "disable" anything, it simply doesn't write code to take
advantage of it.

That's my guess, not having seen the thread he speaks of.

Alex

Alex Johnson wrote:
Bill Davidsen wrote:

Yousuf Khan wrote:

Alex Johnson wrote:

I'm assuming you have a custom application you wrote? If so, are
you compiling it in 64-bit mode or 32-bit mode? This will make all
the difference. If you are insistent on 64-bits, tests have clearly
shown (at least Anand's and a few others that I read months ago)
that Athlon at 64-bits gets a performance improvement while Pentium
4 at 64-bits suffers a performance degradation. This is probably
just a temporary situation...first generation parts and all. But if
you are doing only 32-bit programs, the Pentium 4 outperforms the
Athlon in floating point computation (be sure to use only SSE2/SSE3).




You did see the thread about where Intel deliberately turns off all
SSE support on non-Intel processors, even if they have it?



FUD alert! Just how do you think Intel would disable SSE on my AMD
chip? Witchcraft? SSE works when the chip is shipped but Intel
disables it in my computer by some secret technology? Did you read
about that in the AMD "unfair practices" lawsuit, or what?

Sorry, when I read that I just get a vision of a poor Opteron plugged
into a sacrificial motherboard, it's dark, and by the light of the
smokey fire some dude in feathers and a loincloth is mumbling and
waving a dead chicken. Like a flashback from a bad voodoo horror
flick. Then a bolt of lightning comes down and fries the SSE.

Or maybe they stick pins in the SSE portion of a circuit diagram? I
assume you were joking, or meant to post to alt.conspiracy. Please
reassure me that you haven't gone over to the dork side.


I don't know for sure, but I think Yousuf might be talking about Intel
*software* not intel hardware. A lot of people use intel's compiler
because it does a great job optimizing code for intel chips. A lot of
people who don't use intel chips use intel's compiler anyway. I can see
a scenario where a compiler asks the chip "what is your name and
revision" and the answer AMD gives back does not match any of the intel
chips for which SSE2 applies, so the compiler doesn't try to use SSE2.
It doesn't "disable" anything, it simply doesn't write code to take
advantage of it.

That's my guess, not having seen the thread he speaks of.

Boy, would that be shooting yourself in the foot and blaming the
shoemaker... I believe you can still specify the target processor, at
the application level the 600 series 64bit code should run on AMD.

I don't promise that, I am assuming.

Related: has anyone looked at the MS C++ compiler for the 64 bit Win?

--
bill davidsen
SBC/Prodigy Yorktown Heights NY data center
http://newsgroups.news.prodigy.com