muzician21 wrote:
On Apr 9, 1:32 pm, "JR Weiss" wrote:
"muzician21" wrote...
I dont need to be on the cutting edge, what Core 2 chips should I look
for that would net me about 3x the speed of that 2.4gig P4?
None.
While you would see improvement from any C2D over 2.4GHz, you won't likely
see anywhere near 3X the speed on anything.
Hmm. Looking at a chart like this
http://www.cpubenchmark.net/common_cpus.html
gives the impression there are CPU's that are many times faster.
What I'm mostly looking at is rendering times for processing video
such as through VirtuaDub and for creating DVD's. You feel I I won't
see "anywhere near" 3x the speed? If that's correct maybe just maxing
out the board with a faster socket 478 CPU isn't such a bad idea.
A magazine article, or a web site now, will tend to use
benchmarks that emphasize processor performance this way.
(clock_speed * instructions_per_clock) * number_of_cores
What they do, is test multithreaded software. Multithreading works
best in multimedia applications, because a number of problems there
(processing large data sets) benefit from a divide and conquer
algorithm.
For example, Photoshop could split a picture in two pieces, and
a processor core could work on each half of the picture.
But the truth is, activities on a computer consist of a mix
of single threaded ones and multithreaded ones. So a typical
user doesn't see the huge speedup the above equation might
suggest. For single threaded computing, you'd see an improvement
proportional to just a single core. The Core2 "instructions_per_clock" is
how some of the speedup occurs.
(clock_speed * instructions_per_clock)
So if you wanted a 3x speedup at all times, I'd have to pick a
processor that offers that improvement at all times. To do that,
I'd use a single threaded benchmark. If your target was 3x performance
increase only while you were rendering or shrinking a movie, then a
multithreaded benchmark would tell you that.
I can pick a "Pentium 4 2.4GHz C Northwood" on hwbot.org, and then
look at the collected benchmarks. The "C" means FSB800 (front side
bus speed), which would be about as good as it gets for a S478
processor. A much earlier processor, say one for socket 423,
might be FSB400, making it harder to get data in and out of the
processor.
http://www.hwbot.org/ResultBrowseByP...puModelId=1425
SuperPI 1M ( 1 million digits) 80 seconds at 2.4Ghz
SuperPI 32M (32 million digits) 58 minutes 59 seconds at 2.4GHz
Now, compare to an E8400 Core2 Duo 3GHz processor.
http://www.hwbot.org/listResults.do?...pplicationId=3
SuperPI 1M ( 1 million digits) 15-16 seconds at 3.0Ghz
http://www.hwbot.org/listResults.do?...pplicationId=7
SuperPI 32M (32 million digits) 14:10 to 15:59 at 3.0GHz
The scaleup there implies a factor of 5, in the 1 million digit
benchmark. But the thing is, SuperPI uses about 8MB of data in
main memory, and the E8400 has 6MB of shared L2 cache. I don't know
what the locality of reference is like in SuperPI, but I would be
a bit suspicious that the benchmark is overestimating the speedup.
A lot of the SuperPI data, might end up stored in L2, giving
an unfair advantage and a less than honest performance ratio.
So I can try the 32 million digit benchmark. This still seems a
little on the high side.
If we compare 58:59 to 15:59, that is a factor of 3539/959 = 3.69
Your P4 consisted of a single core, and it could have had Hyperthreading,
which makes a second, virtual core. The virtual core, on a good day,
contributes only an extra 10% to performance, as it runs when the
other core is "blocked". Now, you can buy quad core processors,
and if the software you use can actually use all four cores, then
you should see a good improvement.
The Q9650, is two E8400s inside the same CPU package. It is a quad for $324.
The Q9550 is comparable, and is 2.83GHz for $270.
http://www.newegg.com/Product/Produc...82E16819115130
core core core core Q9550, Q9650
| | | | Block Diagram
-+----+- -+----+- Two silicon die, joined inside.
| 6MB L2 | | 6MB L2 |
----+--- ---+----
| |
+-----+------+
|
LGA775 FSB (used for memory access and I/O)
Nehalem (Core i7) is the most recent generation, and the motherboard
and RAM for it, may add to the upgrade costs. This is an example of
one of those. Socket is LGA1366 instead of LGA775 for the other one.
The extra pins support a direct memory interface.
Intel Core i7 920 Nehalem 2.66GHz 4*256KB L2 8MB L3 Cache LGA 1366 130W Quad $289
http://www.newegg.com/Product/Produc...82E16819115202
core core core core Core i7 is a single die
| | | |
256KB L2 256KB L2 256KB L2 256KB L2
| | | |
-+-----------+----------+----------+-
| 8MB L3 |----- triple channel memory
-------------------+---------------- interface on processor
| (like AMD does it)
LGA1366 FSB (used for I/O)
Using the HWBOT again... 14.5 seconds for SuperPI 1M (when the
entire data set could fit in L3. That is 14.5 seconds at 2.66GHz.
http://www.hwbot.org/listResults.do?...pplicationId=3
The SuperPI 32M is 12:45 at 2.66GHz, and ratio to P4 2.4Ghz is
58:59/12:45 = 3539/765 = 4.6x single threaded.
http://www.hwbot.org/listResults.do?...pplicationId=7
An E8400 is $165, and a motherboard with DDR2 memory makes for
a more reasonably priced alternative. It really depends on
what your budget is. The pricing is such, that buying low
end Intel platforms may not make much long term sense.
(You'd only be looking at upgrading again.)
As far as I know, all the current benchmarks on Tomwhardware charts
are multithreaded, intended to let the extra cores show their stuff.
It is too bad they don't try to be more balanced, and throw
in a less impressive speedup from a single threaded benchmark.
I've used SuperPI above, as an example of a single threaded one.
Paul