Celeron vs. P4

PrimeCruncher · 2003-10-13, 15:09

I know that a Celeron versus a P4 at the same speed for primality testing is much slower, but does anyone know what effect the reduced cache and bus speed have on trial factoring? Thinking of putting together a "monster" to be used for trial factoring (I never do LL tests, takes too long and that's what all my friends are doing so we have a little friendly rivalry going

) and I can save about $100 by going with a Celeron over a P4.

dsouza123 · 2003-10-13, 16:59

PrimeCruncher brings up a critical point both specifically, P4 versus Celeron with SSE2, and in general, what is used/needed to maximize trial factoring speed ?

CPU speed is a major item. Same CPU clocked faster/slower has a direct measurable effect on speed.

SSE2 support for 2^65+ speed optimization is also major. Overriding SSE2 support makes that very apparent.

Cache size ? How big a cache is needed to cache the code path in Prime95 to do Trial Factoring ? Because the OS and other code will also reside in the cache if it is too small does the Prime95 code keep getting kicked out ?

Memory speed/ bus speed ? If it can't be cached what effect does memory/bus speed have ?

The SSE2 Celerons go from 1.7 - 2.7 Ghz all with 128KB and 400 Mhz bus speed.
The P4 go from 2.0 - 3.2 Ghz with 512KB cache and 400, 533, 800 Mhz bus speeds.

The issue of what exactly enhances/reduces TF speed has been on my mind for a while.

Thanks PrimeCruncher for illuminating the issue.

Unfortunately I don't have either processor type so I can't do a comparison test with both types of machines testing the same factor with the results at the different bit depths, also stating CPU (version,speed), memory size, bus and memory speeds.

If the trial factoring code path is cache bound, is the 128 vs 512 of critical importance ?

Saving $100 if the cache size and bus speeds aren't critical to the performance is a significant amount.

Xyzzy · 2003-10-13, 17:22

According to a discussion a long time ago in the TPR thread, frequency is the only thing that matters for factoring...

Of course, it needs to be SSE2 to get the 64-bit or higher speed boost...

So I believe a Celeron will be almost identical in speed to a similarly clocked P4...

PageFault · 2003-10-13, 17:30

The above was true for a wide variety of hardware. I had run trial factoring on PI, PII, PIII and Athlon and the results scaled very well as a function of clock speed. Others reported no difference between snaileron / cumine and athlon / duron. The active code for factoring is probably very small, the whole thing taking maybe a few kb in L2.

PrimeCruncher · 2003-10-13, 17:56

Seeing that I have no intention of running LL tests, I will go with the Celeron.

Thanks everyone for your replies.

E_tron · 2003-11-13, 17:35

If you havn't bought the Celeron already, get the 1.8ghz version, because i have one running at 133mhz FSB (about 2.4ghz)without a problem(with DDR266 128MB).

Plus it was about a $60 CPU.

PrimeCruncher · 2003-11-14, 00:00

Thanks for the tip. Already have a C 2GHz, planning to buy more Celerons when the server for the farm is ready. I would overclock it but I wasn't thorough enough in my research and I found out when I got it running that it won't let me adjust the FSB.

E_tron · 2003-11-14, 02:19

Quote:

Originally posted by PrimeCruncher
Thanks for the tip. Already have a C 2GHz, planning to buy more Celerons when the server for the farm is ready. I would overclock it but I wasn't thorough enough in my research and I found out when I got it running that it won't let me adjust the FSB.

I take that back. you bought the right thing. I wanted to say the 2.0ghz celeron, because it is based off of u.13

2003-10-13, 15:09	#1
PrimeCruncher Sep 2003 Borg HQ, Delta Quadrant 2·3³·13 Posts	Celeron vs. P4 I know that a Celeron versus a P4 at the same speed for primality testing is much slower, but does anyone know what effect the reduced cache and bus speed have on trial factoring? Thinking of putting together a "monster" to be used for trial factoring (I never do LL tests, takes too long and that's what all my friends are doing so we have a little friendly rivalry going ) and I can save about $100 by going with a Celeron over a P4.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Celeron 2.40 too slow?	rudi_m	Hardware	14	2005-10-11 03:31
Celeron D and Willamette owners	Prime95	Software	10	2005-07-08 08:44
New celeron. look, look!	E_tron	Hardware	5	2004-07-13 05:16
Benchmark: P4 and P4 Celeron	E_tron	Hardware	3	2003-11-20 18:31
Intel Celeron 700 MHz?	edorajh	Hardware	13	2003-11-19 15:43

2003-10-13, 16:59	#2
dsouza123 Sep 2002 2·331 Posts	PrimeCruncher brings up a critical point both specifically, P4 versus Celeron with SSE2, and in general, what is used/needed to maximize trial factoring speed ? CPU speed is a major item. Same CPU clocked faster/slower has a direct measurable effect on speed. SSE2 support for 2^65+ speed optimization is also major. Overriding SSE2 support makes that very apparent. Cache size ? How big a cache is needed to cache the code path in Prime95 to do Trial Factoring ? Because the OS and other code will also reside in the cache if it is too small does the Prime95 code keep getting kicked out ? Memory speed/ bus speed ? If it can't be cached what effect does memory/bus speed have ? The SSE2 Celerons go from 1.7 - 2.7 Ghz all with 128KB and 400 Mhz bus speed. The P4 go from 2.0 - 3.2 Ghz with 512KB cache and 400, 533, 800 Mhz bus speeds. The issue of what exactly enhances/reduces TF speed has been on my mind for a while. Thanks PrimeCruncher for illuminating the issue. Unfortunately I don't have either processor type so I can't do a comparison test with both types of machines testing the same factor with the results at the different bit depths, also stating CPU (version,speed), memory size, bus and memory speeds. If the trial factoring code path is cache bound, is the 128 vs 512 of critical importance ? Saving $100 if the cache size and bus speeds aren't critical to the performance is a significant amount.

2003-10-13, 17:22	#3
Xyzzy Aug 2002 2·3²·13·37 Posts	According to a discussion a long time ago in the TPR thread, frequency is the only thing that matters for factoring... Of course, it needs to be SSE2 to get the 64-bit or higher speed boost... So I believe a Celeron will be almost identical in speed to a similarly clocked P4...

2003-10-13, 17:30	#4
PageFault Aug 2002 Dawn of the Dead 235₁₀ Posts	The above was true for a wide variety of hardware. I had run trial factoring on PI, PII, PIII and Athlon and the results scaled very well as a function of clock speed. Others reported no difference between snaileron / cumine and athlon / duron. The active code for factoring is probably very small, the whole thing taking maybe a few kb in L2.

2003-10-13, 17:56	#5
PrimeCruncher Sep 2003 Borg HQ, Delta Quadrant 2·3³·13 Posts	Seeing that I have no intention of running LL tests, I will go with the Celeron. Thanks everyone for your replies.

2003-11-13, 17:35	#6
E_tron Sep 2002 Austin, TX 231₁₆ Posts	If you havn't bought the Celeron already, get the 1.8ghz version, because i have one running at 133mhz FSB (about 2.4ghz)without a problem(with DDR266 128MB). Plus it was about a $60 CPU.

2003-11-14, 00:00	#7
PrimeCruncher Sep 2003 Borg HQ, Delta Quadrant 1010111110₂ Posts	Thanks for the tip. Already have a C 2GHz, planning to buy more Celerons when the server for the farm is ready. I would overclock it but I wasn't thorough enough in my research and I found out when I got it running that it won't let me adjust the FSB.