Ron_Jeremy

Where did all the MHz go?

Recommended Posts

I just finished building a pc for a client of mine, & was lucky enough to "test" it for a full day before she took delivery. Anyway, the general system specs are as follows:

Abit 865 board

PIV 2.8C

2 x 512MB Corsair DDR (the LL stuff)

Ati 9200

80GB WD w/8MB buffer

Win2000 Pro

My "secondary" pc (before selling it last week) specs were as follows:

Supermicro P3TDLE (Serverset III LE)

1 x P3 1.4

2 x 512MB Reg PC133

TNT2 PCI

Buncha Atlas 10KIII's

Win2000 Pro

Anyway, the only thing I really noticed after using the P4 beast for a day is that it sure doesn't feel any faster than my old P3 system. In fact, general Windows "responsiveness" felt faster on the "half as fast" P3 system.

I fired up a dozen or so browsers, tons of command windows, just about every applet in Control Panel, every app in Office XP Pro, as many media players as possible, & a bunch of other stuff I can't even remember on both pc's at the same time (or as close to the same time as humanly possible). Every time the P3 system (on a 4 month old Windows install) would beat out the 1400MHz faster P4 system. Alt-tabbing thru the windows & opening yet more new apps would again take less time on the system that was "half as fast" on paper. How is this possible? The only time the P4 spanked (by a monstrous margin I must admit) the older system was when I used LAME to encode 2 dozen or so wav files to mp3's.

I understand that there are programs out there that make good use of all the MHz a high end P4 can muster, but I get the feeling the average pc user could still get by with a P3 600.....

Share this post


Link to post
Share on other sites

This is exactly why we are "storage enthusiasts". The Atlas 10kIII's have HALF the seek time of the WD JB's and so a system that is otherwise far slower (especially in memory bandwidth) actually "feels" faster. This suggests that the main bottleneck is, as we all already knew, the disk drive.

I personally keep a bunch of Tualatin systems myself (and each has a 15k.3), because I don't like to heat up the house by leaving the P4 systems on.

Share this post


Link to post
Share on other sites

Depending on what you do with a PC, any PIII is good enough. I borrowed a PII 350 with 384 MB RAM to someone and he said it was surprisingly fast running Windows XP. It did have a fairly recent disk though (60 GXP).

Share this post


Link to post
Share on other sites

Yep. Access time is king when it comes to snappiness. Despite their excellent performance on SR's application benchmarks, 8 MB cache 7200 rpm IDE drives are mechanically inferior -- something that the electornic wizardry cannot compensate for all the time. If the data requested is not in the cache, the slow actuator is exposed for the sloth that it is.

Share this post


Link to post
Share on other sites

Another thing that I would like to mention is that the Atlas 10k3 like its SCSI bretheren are simply much better at handling longer I/O queue depths. Like RAID 0 setups, SCSI drives are born for situations that hammer the drives with requests. Single IDE drives, no matter how much cache they have, do not deliver the goods when you crank up the load. When you kept on opening up new apps and switching tasks, you were generating conditions that SCSI drives and RAID 0 setups shine in.

Share this post


Link to post
Share on other sites

This all appears to be in direct contrast to section 5 of the 'Performance Factors' FAQ: http://www.storagereview.com/php/tiki/tiki...formanceFactors

"...SCSI drives...are not necessarily faster for non-server use. In fact, the safe rule-of-thumb is that given equal spindle speeds, a current-generation ATA drive will be faster than a current-generation SCSI unit."

Maybe theoretical 'non-server' usage is completely different to what actually happens in real life...? Confusing.

Share this post


Link to post
Share on other sites

Ron, looking at that setup, I'm going to guess it was driver relater.

Try out the intel BMIDE drivers (Are they still making them?), the newest ATI Cats, etc... Check the SMART values as well to make sure that the drive performance isn't drgraded due to remapped sectors. Make sure write caching is enabled on the drive.

Just making sure that you have the basics covered.

Thank you for your time,

Frank Russo

Share this post


Link to post
Share on other sites

I replaced my Cheetah 36ES and 36XL with a 60 GXP and a Barracuda 7200.7. last week and really, the performance difference is close to nil. Only when I start several apps at the same time can I notice the difference - barely.

Before anyone says "new windows install" ... I ghosted the SCSI drive to the IDE drive. (and to test it I ghosted the IDE drive back to the SCSI drive as well so they were both equally defragged).

Share this post


Link to post
Share on other sites

Don't forget that the PIII 1.4Ghz IS a fast chip and faster than it's comparable P4

In pure Seti@home terms it's comparable 2200+ AthlonXP (~1.8Ghz) (runnig the same memory)

Mind to be honest, the P2.8 should still be a lot faster...

So i'd go with the drive theory adding to the rest of it.

Although the ATI card should have made the Windows snappier, bah it's just proves how bad the P4's are :rolleyes:

Hopefully they bring the P-M's into desktop or server standard chips...

Low power, cool Rack's anyone.

Perfect for a 'mini' or 'media' PC

Hey a P-M 1.4Ghz will crunch an average seti unit in about the same time as a P4 3Ghz

Share this post


Link to post
Share on other sites

An easy test would be to fire up taskman and see how often your CPU stays above 10% for more than an instant. It's pretty rare in most desktops (except lame encoding and a few other applications, as you noticed).

I've tested XP on a dual Pentium II 266 with an X15-36LP drive (2nd gen 15K Cheetah) and 192MB of EDO ECC RAM. It was actually quite snappy (I think the RAM and the built-in video were the limiting factors). These days, I'm more concerned about the noise of the CPU fan than the speed of the CPU. Pretty much any processor manufactured since 1999 will do fine for me (e.g. 500MHz and above).

Share this post


Link to post
Share on other sites
Ron,  looking at that setup, I'm going to guess it was driver relater.

Most certainly not driver related. I had the latest Intel drivers for the chipset/nic installed & the newest ATI drivers. Dynamic disks, Microsoft critical updates, & SP4 too.

We all know if you were to run the 2 pc's thru the standard slew of benches that the older system would barely win one (if any) event. The only advantage the P3 has is in its disk subsystem, while the P4 has twice the the advantage everywhere else on the spec sheet. I guess my disappointment stems from the fact that Windows "snappiness" would never lead me to believe one system has a one thousand four hundred megahertz disadvantage.

Share this post


Link to post
Share on other sites

it's rather difficult to diagnose a system using only a subjective analysis. if i were to venture a guess, i would think that the drive wasn't running in dma mode. suffice it to say that i would rather have a single 2000jb than a few atlas 10k3s, and noise is not a concern of mine.

Share this post


Link to post
Share on other sites

Windows shell is memory and disk bound, it is not processor bound. I doubt that you are seeing high memory utilization on either machine.

The only cpu intensive thing that I can think it does is unzipping files.

Also it typically only reads metadata. This causes a large number of small disk reads.

IE is typically network bound. You will find that loads significantly faster if the home page is "about:blank"

Share this post


Link to post
Share on other sites

Where the P!!! utilizes a pipe of 10 stages the

p4 have 20.

The benefit of having a P4/2.8G compared to a

P3/1.4G is only theoretically double MIPS.

Ofcourse only by looking at the CPU perspective.

/casa

Share this post


Link to post
Share on other sites
I just finished building a pc for a client of mine, & was lucky enough to "test" it for a full day before she took delivery. Anyway, the general system specs are as follows:

2 x 512MB Corsair DDR (the LL stuff)

2 x 512MB Reg PC133

I fired up a dozen or so browsers, tons of command windows, just about every applet in Control Panel, every app in Office XP Pro, as many media players as possible, & a bunch of other stuff I can't even remember on both pc's at the same time (or as close to the same time as humanly possible). Every time the P3 system (on a 4 month old Windows install) would beat out the 1400MHz faster P4 system. Alt-tabbing thru the windows & opening yet more new apps would again take less time on the system that was "half as fast" on paper. How is this possible? The only time the P4 spanked (by a monstrous margin I must admit) the older system was when I used LAME to encode 2 dozen or so wav files to mp3's.

Both have the same amount of memory. The performance of a PC doesn't depend for 100% on the CPU and as stated the storage subsystem can have a great impact too.

Share this post


Link to post
Share on other sites
e_dawg, i don't believe firing up a few applications (in typical power user fashion) does anything to generate above-ide queue depths

Although "typical" is subjective, I do belive that our resident porn star was not using his desktop in everday typical power user fashion, but rather purposely pushing the system to its limits to see how it could handle a heavy multitasking load. I can generate I/O queue depths of 2-3 when I am doing some heavy multitasking on my desktop (hit 4 on my laptop despite having a "fast" Toshiba 5400 rpm 16 MB cache 2.5" drive). Not worthy of OLTP servers, I know, but it is still something that a SCSI drive or striped IDE's would be more comfortable with than a single IDE drive. I noticed the difference myself with my WD800JB, striped WD600BB's, and single Atlas 10k3 drives at home. The WD800JB is the slowest drive in terms of light usage snappiness and heavy multitasking responsiveness.

Despite the praise lavished upon the 8 MB cache 7200 rpm drives around here, I find that they just don't compare in terms of snappiness and will never settle for one as my system drive as long as I can afford a decent 10k+ drive.

Share this post


Link to post
Share on other sites
This all appears to be in direct contrast to section 5 of the 'Performance Factors' FAQ: http://www.storagereview.com/php/tiki/tiki...formanceFactors

"...SCSI drives...are not necessarily faster for non-server use.  In fact, the safe rule-of-thumb is that given equal spindle speeds, a current-generation ATA drive will be faster than a current-generation SCSI unit."

Maybe theoretical 'non-server' usage is completely different to what actually happens in real life...?  Confusing.

I agree with the fact that the proverbial typical desktop PC usage modeled in SR's application-level benchmarks may not be entirely representative. Completely different? Nah, I'd say it's reasonably close, but benchmarks are just benchmarks -- that is, they measure specific performance metrics in specific situations. By their very nature, they cannot model everyone's usage patterns nor can they measure all factors that contribute to the user experience because some of these are not readily quantifiable in any reliable or meaningful way. Eugene has done a good job with the latest version of SR's application level benchmarks, but there is a limit to what you can measure and what you can include in your model. Anyone who has done quantitative analysis knows that even the best models are falliable and are not always representative of the complexity that is reality.

For example, I still feel that access time is important when it comes to pure snappiness. Indeed, an 8 MB cache and intelligent firmware can hide a slow actuator to an extent, but that only works if the requested data is in the cache. If it's not, you will be relying on the mechanics of the drive... if your drive is (e.g., notebook drives), everyone will know. Even a 5400 rpm 16 MB cache notebook drive is very noticeably slower than a mechanically superior 7200 rpm 2 MB cache desktop drive. And "snappiness", of course, is not quantifiable... or at least not with the traditional I/O per sec measurement.

But nevertheless, this "theoretical non-server usage" in Eugene's FAQ and SR's benchmarks is representative enough of what happens in real life that it's better than most benchmarks out there, and we should be glad that we have it. When I look for a new drive, this is the first place I come. I take the SR application-level benchmarks, measured access time, net drive temp, idle noise, and Eugene's subjective noise comments ... put those factors against price and availability, and I've got my decision.

Share this post


Link to post
Share on other sites
IE is typically network bound.  You will find that loads significantly faster if the home page is "about:blank"

i believe this has more to do with cache lookup than network speed. the actual launch of ie doesn't 'pre-fetch' the page afaik.

Share this post


Link to post
Share on other sites

i would expect a given 10k drive to feel snappier than a given 7.2k drive (so long as both have good cache/firmware), but i don't think it has anything to do with interfaces or queue depths. i wouldn't expect a wd jb to feel like a slug.

Share this post


Link to post
Share on other sites
For example, I still feel that access time is important when it comes to pure snappiness

You may be right, you may be wrong on this. I have not seen much saying that reduced access times directly lead to a noticeably quicker OS on the desktop PC. Personnally i have often found that believing in something tends to influence what one seems to feel. I am not wanting to annoy you, what i say concerns myself in fact. In my situation what i "believe" is a bit different though. I don't think in terms of "this drive is faster than this other one" concerning the drives i use in my own computer (in case they are reasonably comparable of course). As long as the objective test results are good enough i am OK with that; i just make sure that the disks holding my OS and applications are fast.

In definitive i did not have the "impression" that an Atlas10KIII class drive was definitely faster than my previous IBM 75GXP (Raid 0) although there is quite a difference objectively speaking. Basically, the less i care the less i notice as long as everything is stable and fast in my eyes. I think that what one thinks and feels is quite dangerous and will often lead to contrary positions from different people hardware even being equal.

This is probably valid for CPUs too in respect to higher speed grades if one restricts the usage to what i would call general stuff like internet browsing & (heavy) multitasking included for instance which any power user tends to do regularly. General OS usage needs a quite a stronger CPU to be noticeably quicker.

Second, as far as feelings go, there is also another problem relative to the one or other usage and the one other system etc. Maybe one would not "feel" that the one or other drive is faster because there is plenty of Ram etc. (which indeed is a simplified example but halfway true).

Personnally, that's just my opinion, i prefer objective results like Storagereview's over feeling (after all we have the same reference).

Share this post


Link to post
Share on other sites

To throw another ball in the court..

A system can also 'feel' snappier if you use a decent mouse compared to an poor one.

My MS Optical Mouse makes the computer 'feel' quiker, snappier. The cheapo ball mouse I have makes it 'feel' terrible

:lol:

Tis true though, the screen, keyboard and mouse. The things you interact with do make one feel better than the other as you just prefer them.

Goto a 14" 60Hz monitor and you automatically assume it'll be slow, so you start to believe it's slow (even though it maybe a Quad Xeon setup ;))

Share this post


Link to post
Share on other sites

I think Kenneth and Jeff already more or less addressed this, but by opening a bunch and task-switching a bunch, you haven't done anything processor-intensive. So why would you expect a large difference in performance of these activities based on your processor speed? These activities are more likely to be (as Kenneth said) disk- or memory-bound. Hence, the more "snappiness" with your faster 10Ks. You DID see a performance gain in the one CPU-intensive task you ran: encoding. Games would probably also see some gains (though not on the same scale as encoding, I would wager).

But even then, was there a really significant difference? The Atlas' seek is HALF that of the JB. Shouldn't we see a much greater difference in performance? How big WAS the perf difference you saw? Less than a second's delay for each opening? Perhaps a little worse, with all that grinding away, and hitting the page file at the same time by alt-tabbing.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now