Dual Core? BFD
 

So Intel has finally announced the release of its dual core Pentium.

I see much media hype about how, since each core is hyper-threaded, it will
run 4 threads at once.

Whoopee. BFD.

Nowhere is there any discussion of what FSB contention or cache misses will
do to the overall performance. Will these so-called 4 simultaneous processes run even twice as fast a just a single process?

This dual-core computer clearly will not have the same performance as two
separate computers, running independently.

Exactly what performance increase will be obtained seems to be a topic that
is carefully avoided by Intel.

I am in the market to buy a new computer. I went to CompUSA with
a CD containing code and data. I wanted to benchmark my Number Field
Sieve code on a new 3.8G P IV vs. an Athlon. The sales people would
not allow it!!

<expletives deleted>

Auto dealers let one test drive a car before you buy! Why can't I #&~&@*
test compare computers before I buy???

I'd like to try out one of these dual core machines before buying one......

Bob :furious: :furious: :furious: :rant: :rant: :rant:

[quote]I am in the market to buy a new computer. I went to CompUSA with
a CD containing code and data. I wanted to benchmark my Number Field
Sieve code on a new 3.8G P IV vs. an Athlon. The sales people would
not allow it!![/quote]

Look for a shop that allows it - and then get your computer there. Afterwards, send an eMail to a representative of CompUSA telling him/her that his/her company just lost a lot of $$$ and a customer.
Maybe some others also join in. If there are enough, it might make them rethink. After all, you shouldn't be able to damage anything...

[QUOTE=Mystwalker]Look for a shop that allows it - and then get your computer there. Afterwards, send an eMail to a representative of CompUSA telling him/her that his/her company just lost a lot of $$$ and a customer.
Maybe some others also join in. If there are enough, it might make them rethink. After all, you shouldn't be able to damage anything...[/QUOTE]
Sound advice.

I would also give serious consideration to buying an Athlon-64 system. The 3500+ seems to be near the sweet spot of price/performance at the moment.

However, you will need to find someone with a AMD-64 system to let you build your benchmarks, which may be slightly tricky.

I'll be in the market for a replacement system RSN now that I'm going to be employed again. I still don't know whether it will be a chunky AMD-64 box or a baby cluster.


Paul

[QUOTE=xilman]Sound advice.

I would also give serious consideration to buying an Athlon-64 system. The 3500+ seems to be near the sweet spot of price/performance at the moment.

However, you will need to find someone with a AMD-64 system to let you build your benchmarks, which may be slightly tricky.

I'll be in the market for a replacement system RSN now that I'm going to be employed again. I still don't know whether it will be a chunky AMD-64 box or a baby cluster.


Paul[/QUOTE]

Yes, I have considered it (and rejected it). The problem is the lack of 64 bit
software, especially the OS and a good *** COMPILER ****.

I have doubts about the reliability of 64-bit Linux [it hasn't been around long]
and as for 64-bit Windoze, the less said the better.........

Maybe when the software gets here I will consider it.......

[QUOTE=R.D. Silverman]Yes, I have considered it (and rejected it). The problem is the lack of 64 bit software, especially the OS and a good *** COMPILER **** [/QUOTE]

Bob,

Personally, I think that you are placing too much emphasis on the wrong components. You can run the machine as a 32-bit system and still use 64-bits for the inner loops of your sieving code where it will make a significant difference. (I assume that those loops will still be hand coded in assembler anyway) This should bridge the gap until you are comfortable with the 64-bit version of whatever OS you choose to run.

Richard

[QUOTE=Wacky]You can run the machine as a 32-bit system and still use 64-bits for the inner loops of your sieving code where it will make a significant difference. [/QUOTE]

You cannot run any 64-bit code using a 32-bit Operating System.

I'm using the Win64 OS and beta MS C compiler with few problems thusfar. Of course this isn't a very thorough test of the OS.

My big worry is the C compiler will end up costing $2K when the beta period expires.

[QUOTE=Prime95]You cannot run any 64-bit code using a 32-bit Operating System[/QUOTE]

Perhaps that is true with that architecture -- All the more reason that I am happy that I use a PPC G5 instead. :)

actually there are pluses to duel cores A its easier to get duel core chips if a xeon version comes out and drops consederablly prices will be cheaper hence more bang for the buck but the biggest thing i can see is 2 cores with ht your dissing it the core might only run at started speeds but its a duel core hence you dont need to buy xeon board and other things that are uasally requiered plus you dont need duplicates of each system hence you would proibly spend for for 2 comps...

[QUOTE=R.D. Silverman]Yes, I have considered it (and rejected it). The problem is the lack of 64 bit
software, especially the OS and a good *** COMPILER ****.

I have doubts about the reliability of 64-bit Linux [it hasn't been around long]
and as for 64-bit Windoze, the less said the better.........

Maybe when the software gets here I will consider it.......[/QUOTE]
Fair enough. I'm rather more willing to take a chance on these things, to get some use out of it and some experience with the system straight away, in the belief that things will improve during that lifetime of the machine.

A case in point: I bought a PPro, when Intel thought everyone would stop running 16-bit software in short order. It's still giving me very good service. Its primary roles are NFS sieving and hosting the in-house ECMNET server. Almost all its long life it has run some version or other of Linux.


Paul

[QUOTE=R.D. Silverman]Yes, I have considered it (and rejected it). The problem is the lack of 64 bit
software, especially the OS and a good *** COMPILER ****.

I have doubts about the reliability of 64-bit Linux [it hasn't been around long]
and as for 64-bit Windoze, the less said the better.........[/QUOTE]64bit Linux is working in many commercial systems since April 2003 and for 64bit Windows there just the RC2 became available to the public (earlier for MSDN subscribers and beta testers).

And you could install 32bit Windows, 64bit Windows and Linux on the same machine. As far as I read (in MS statements) the .NET compilers achieved some impressive boosts (although compared to older versions). And for Linux there are also many Compilers with impressive results including GCC (wouldn't be a bad start). See new SPEC results from Pathscale [URL=http://www.pathscale.com/pr_021505.html]here[/URL] and a nice 64bit compiler comparison [URL=http://www.digit-life.com/articles2/insidespeccpu/insidespeccpu2000-part-h.html]here[/URL].

Although it does do 64-bit, don't discount the Athlon64 as a 32-bit CPU. As one data point, in my laboratory I have two Athlon-XP's and one Athlon64, all with about the same PR rating of around 3100+ (don't remember exactly what each is but they're all in the ballpark) that in their spare time do primarily sieving for NFSNet or using GGNFS. With the NFSNet line siever, the Athlon64 runs about twice the speed of the Athlon-XP's. (I've always been curious about this...this is the largest difference I've seen for any program!) For example, here's my stats for one number in which all three machines (escatter1, 2, and 3) were all running during the entire project:

[URL=http://www.nfsnet.org/stats2/statsreporter.cgi?template=userdetails.html&sortorder=d&project=11_199M&sortcolumn=3&pagesize=&page=0&userid=jgchilders&]NFSNet stats[/URL]

Using Franke's lattice siever, the Athlon64 runs about 25% faster than the Athlon-XP's.

Greg