Quad/Dual Core Duo Questions

 
Hey all,

I understand most of the basics about hardware, like DIMMs and FSB, and
HD/RPM, caches, etc. and this is dangerous, especially when I decide to
build my own computers, lol.  Anyways, here's my tale and questions.

I decided about six months ago to "upgrade" my computer.  I thought I
was in a 89' buick compared to what I could be in.  And the buick, aka,
my system, was actually pretty good.  2GB RAM, 3.2 GHz Pentium with HT
(< this is important later), and a few too many harddrives.  So trying
to get in gear with teh times, I decided to get a new motherboard, new
harddrives (using SATA inplace of my IDE mess), and the crown jewel, a
2.33 GHz Dual Core Pentium.  Now, I had done my research on HT versus
Dual Core, and I was convinced that I was going to see a nice biiiiiig
bump in performace, especially that my FSB was huge compared to the old
FSB, and my RAM had better ratings than before.

So I spend two weeks putting the thing together, working out the bugs,
this and that.  And when its all said and done, it runs like molasses.
My 3.2GHz w/HT ran WAAAAAYYYY better than the dual core.  I am soooo
ticked at myself.

So, here are my questions:  

One, what did I do or think wrong..... and what should I know now that
I should have known before I bought the Dual Core?

Two, The Dual Core Pentiums come in three (or four, maybe) flavors.  Ya
have your Dual Core, Dual Core Extreme, and then Quad Core (and possibly
Quad Core Extreme sometime soon).  Is the quad core simply a dual core
with HT?  or is Dual Come Extreme using HT along with its unlocked
ratio?

Three, should I simply forget that there ever was a thing as HT? or do
you think Intel is going to be using it, or is using it someplace in the
future?

 
<snip>

What operating system are you using? If your new pc runs Vista then it will
run "like molasses" regardless of the hardware. (Hopefully it will be better
when SP 1 comes out.)


My understanding is that Quad core is 4 cores, so its not dual core with HT.
HT is a technique for getting more out of each core by letting the
instruction pipeline work on instructions for other threads. Can't answer
the rest, sorry.


HT always seemed quite a clever idea to me, but it does seem to have
disappeared from the promotional literature. I've no idea what its future
is, but it would seem a shame if it has no place in the future.
--
Brian Cryer
www.cryer.co.uk/brian

 

Not at all -- Hyperthreading was a hack for an insanely poor processor
design in the first place.

Unless we ever see a stupidly long pipeline like the P4 in the future,
HyperThreading will stay buried as a bad memory.

  ??>> HT always seemed quite a clever idea to me, but it does seem to have
 ??>> disappeared from the promotional literature. I've no idea what its
 ??>> future is, but it would seem a shame if it has no place in the future.

 D> Not at all -- Hyperthreading was a hack for an insanely poor processor
 D> design in the first place.

 D> Unless we ever see a stupidly long pipeline like the P4 in the future,
 D> HyperThreading will stay buried as a bad memory.

SMT is widely used to hide memory latencies -- when one thread is blocked
waiting for memory operation (cache miss), another thread gets executed.
this gives significant performance boost on some workloads, and is used in
high-end processors like POWER, Itanium, UltraSPARC. especially in
UltraSPARC T2 -- they run 8 threads per core, have very small cache, and
beat any other processor in performance-per-mhz factor on webserver
workload.

HyperThreading is an extended form of SMT -- it works not only on cache
misses, but in general course, allowing to use execution units
simultaneously. probably other processor architectures doesn't need this
much -- they don't have lots of spare execution units, or it's not worth
complexity. but it's not totally different -- on UltraSPARC T2 one thread
can run expensive FPU operations while other do simple arithmetic.

there are rumours
(http://www.intel.com/pressroom/archive/releases/20070328fact.htm) that
Intel is planning to return some form of SMT on desktop processors, but it's
not clear if that would be mere SMT on cache misses, or HT sharing execution
units..

 ldiaco wrote:

To eliminate any confusion about what you bought, you could quote a
model number, like E6600, E2180, or even an SLxxx code (printed on the
processor box). The thing you bought, could be a Core2 family processor.
For example, it might be an E6550.

http://processorfinder.intel.com/List.aspx?ProcFam=2558

The Core2 2.33GHz, you'd multiply that by 1.5x to 2x, to get some idea of
its "Pentium 4 equivalent" speed. It should feel a little faster
than what you had. And not molasses. More like 3.5 to 4.5GHz or so
in P4 equivalents.

In the past, I've run into a couple people, whose systems seemed to be
stuck at the low multiplier. With EIST (equivalent in its way, to
Cool N' Quiet on AMD), the multiplier changes as a function of system
load. Now, I never did find a definitive step that was missing in their
build. The builders eventually concluded it was fixed, but since they
changed so many things at once, they lost track of what fixed it.

You can use RMClock, to monitor the system. RMClock is supposed to be
able to monitor for throttling, if it is present. I don't really like
the graphical output, but it is better than nothing.

http://cpu.rightmark.org/download.shtml
http://cpu.rightmark.org/download/rmclock_230_bin_upd1.exe
(Screen shot)
http://www.notebookforums.com/attachment.php?attachmentid=8955&stc=1&d=1143767059

"Molasses" can come from a number of things. For example, say your
hard drive was running in PIO mode. The processor would waste a lot
of cycles, any time a file is referenced (since it would use polled
transfer). The bandwidth would be limited (4MB/sec to 8MB/sec).
Programs would be slow to load, but fast once started. You can
test with HDTach or HDTune, and get some idea of the actual speed.
My disk speed ranges from 60MB/sec to 40MB/sec sustained, from the beginning
to the end of the disk. More modern drives manage 70MB/sec or a
little more, in sustained performance. The performance graph
should be a curve - a flat line means cable/interconnect trouble.

Now, if you were to test with SuperPI 1.5 or something similar,
that would spend a good deal of its time using nothing but
processor and memory. Comparing the old system to the new system,
would then give an idea if the CPU and memory was the limitation.
Just the "feel" of the system, may not be narrowing down where the
problem is. (On the machine I'm typing on, SuperPI 1.5 will do the
1 million digits test in 50 seconds. It became that slow, after
I added my antivirus software. I think I might have had maybe
45 seconds at one time, before the antivirus was added. Just if you
want to compare. SuperPI is single threaded AFAIK. My system is a
Northwood P4 3.15GHz/FSB900/DDR450 dual channel. A highly overclocked
Core2 can do the same benchmark in sub 10 seconds, to give some idea
of how good it can be. So my system is 5x out of date.)

http://www.xtremesystems.com/pi/index.php    (see super_pi_mod-1.5.zip)

The multiplier is one issue. Processor throttling (CPU instruction rate
drop, to try to stave off overheating), is another mechanism for
losing performance. You want enough cooling, to stay below 70C. About
65C to 70C, is where some of the newer processors start to throttle.
The cooling solution should stay clear of those kinds of temps, in
order to get 100% of the goodness.

Keep the RMClock window open, then run SuperPI, and watch the graph
while it runs.

You can also have a look through this thread, but first I'd prefer
to see you get a handle on exactly what is slow, before going
crazy with unrelated fixes.

http://forum.notebookreview.com/showthread.php?t=60416

    Paul