SMP in the modern age

[Adcadet]

hey all -
I remember back in the day ('98-'99 ish) when I was running a pair of Celeron 400s which seemed much faster than my P2-450. At that point, I was pretty well hooked on SMP systems. Next came my dual P3-700s, which seemed much more responsive than my friend's P3-800 (though I was running a Cheetah 18XL and he was running an IMB 75 GXP - same RAM, same vid card). Then came my current system, dual 1.2 GHz Athlon MPs. These have served me well, but it's coming time to upgrade (since my current sytem is getting pretty flakey and my computer is starting to feel slow again). My wife's old P2-400 is getting ready for an upgrade, and I'm wondering if I should spent the $200-300 on a new chip/mobo/RAM for her or just give her my old chips/mobo/RAM.

I know some of you (Clcoker?) have used more modern SMP and non-SMP systems lately, and was just wondering if anyone could share their opinions.

My wife and I do some light photoshop, some office stuff, some DVD and CD ripping. Nothing earth shattering. And I play the occasional game.

 

[blakerwry]

I tend to think that the more you multi task, the more an SMP sytem will help you out.

Not since the Pentium II/III has there really been a good selection of SMP mobos designed for "power user" dekstop use. It seems that most of the current SMP systems are designed for Xeon/AthlonMP and offer features that would be more apealing to server use than desktop use. This has made me stay away from SMP even though I would have liked to get into it.

If you don't use alot of SMP apps, especially the ones that get a big boost from SMP like 3D rendering and Audio/vid encoding apps then there shouldn't be any real need to use SMP.

I'm not sure if going SMP is a one way street, "Once you go SMP you can never go back" sort of thing, but overall I think you'd be best off getting a nice 2GHz+ uniprocessor machine and an Nforce2/kt400a or better mobo given the use you described.

 

[Adcadet]

Blake -
given my status as a poor medical student, I think you're right. I'm sure a nice dual Athlon MP 2800 or dual Xeon 2.8 GHz would be very nice, but I clearly can't afford either of those setups. I'm wondering what the difference between, oh, say a XP 2500 on a nForce2 vs. my current system (dual MP 1.2 GHz) and vs. a current lowish-end MP (perhaps an MP 2000, which seems to be in a sweet spot right now at $122) would be. Oh, if I only had $10,000 to play with for my computer addiction....err, hobby.

 

[time]

James found that a single faster processor was more useful. But I don't think he was talking about Photoshop. Still, with either an overclocked P4 or Athlon, it's hard to go wrong.

If you can wait a bit, the newer AMD Bartons might overclock better and become a really terrific bargain for someone on a budget. As it stands, you'd be hard pressed to double your clockspeed, which I suspect is the primary issue in Photoshop performance.

Right now, maybe a P4 2400/800 FSB overclocked to 3200 on a Canterwood board? But with your usage profile, an Athlon XP2000-XP2400 would be a better buy; no point in paying a premium for MP if you're not going to use it.

 

[Adcadet]

and most of what I do in Photoshop is pretty basic - crop, resize, adjust contrast/color. Those things are 98% of what I do. I would assume that for such things I would need to get a CPU that is twice as expensive to see a noticable improvement.


Back in the day it seemed like most stuff I was doing was CPU-limited rather than memory speed-limited, and thus going dual was a reasonable way to get better performance. Now-a-days I'm not so sure that's true. From what I remember seeing, FSB speed seems to be much more important, and thus perhaps going from my dual MP 1.2/266 to, say, a 2500/333 or even a 2X00/400 would give me a big bump in speed. Now here's a question: for ~$300US I can get an nForce2 board, 2*256 MB RAM, and an XP 2400/2500/2600 running at 333 (maybe higher) FSB. For about the same price, I could get a pair of MP2000, but still on the 266 MHz FSB. I'm wondering what would give me the best performance for what I do.

No, I do mostly office work, but I am usually doing multiple things at once. Now, I realize that everybody thinks that they stress their computer harder than everybody else, and that even if they are using just office stuff, they claim to multitask harder than the others. But I suspect I might be on the higher end of things. Is there any way to get Windows (XP) to log CPU, memory, and disk usage? It'd be interesting to see how often I max out my current system. If it never goes >80%, perhaps a faster FSB would be in order. Or perhaps a faster HDs (I do notice that I'm often waiting for things to load like Word and Mozilla....I wonder if I'm waiting on the HD).

 

[Mercutio]

I loved SMP on desktop machines. My first was a 486/33 dualie (OS/2 then Linux for SMP). I moved to dual P5/133s (one NT, one Linux) then PPro200s. My last SMP box to date was a dual P3/550.

Basically, the SMP rigs were more generally responsive and held up well, but over time I found it harder and harder to see any benefit from the second CPU.

I wouldn't mind going to a two CPU solution, but at this point, it would be a luxury. Except video encoding, I don't think there's anything I could do at home that would touch a second CPU enough to matter, and at this point, the cost difference for SMP (expensive motherboard, expensive CPU, expensive RAM) is enough that a whole second PC is nearly as feasible.

 

[time]

IIRC, bigger L2 cache = better multitasking, including the larger effective cache achieved by coupling two CPUs. But it's unsound to directly compare Athlon with P4 in this measure. If you believe OfficeBench, Athlon is inherently superior, but P4 hyperthreading presumably reverses this.

So on the Athlon size, a Barton 2500 or 2800 should be best. Naturally, a Xeon is the best P4.

Don't overlook the fact that most nForce2 boards can unlock the multiplier on any Throughbred or Barton (or an MP of course), or the great bin splits that AMD has been getting (less so with Barton). My daughter's (new) computer has an XP1700 pretending to be an XP2600, only with a 58% higher FSB of 210MHz. And that's not maxed out. :wink:

 

[zx]

I just got an SMP system, a dual Athlon MP 2600. Before I had a single Athlon XP 1800. In "general use" (surfing the net, office work, general system responsiveness) there is no noticable difference between both systems. However, the reason I switched to duallies was for VMWare and there, duallies make a huge difference. When I work with a VM (or more ), there is much less impact on system performance. Also, when I play counter-strike, I can leave my VM's powered on without them affecting gaming performance. Also, you can run two heavy tasks at the same time (like MP3/MPEG decoding, gaming, achiving).

Basically, SMP is a luxury. I really like my duallies, but if I don't have the cash, I would build a nice high-performance single processor machine. Also, you must use the power at your disposal. In other words, the way you use your computer must be able to take advantage of both processors. I'm sure that many people will not feel any difference between a single and dual processor machines.

For the uses you enumerated, I don't think that SMP will help that much. A nice single processor system may be as good or even better than a dually. For you needs, it's better to have a single XP 2500 or 2800 than a dual MP 2000.

 

[Adcadet]

I miss the days when I could buy two Celeron 400s and an Abit BP6 for about the same price as a high end P2/P3 and the performance we better. But now that SMP motherboards are so expensive ($190 for a Tyan Tiger MPX), and that MPs cost significantly more than an XP, it's a luxury I think I'm priced out of.

 

[honold]

http://www.intel.com/design/chipsets/e7505/ is something to consider if you honestly decide you could use smp. like the others, i believe it's a not-so-necessary luxury. put the money in a 9800 pro or something

 

[Buck]

Dullies are nice, but as zx pointed out, they are a luxury. Usually they're a luxury not worth having. I was using a dual Athlon 1.2 Ghz and was happy when I moved away from it. The cost of duallie systems just isn't worth impact on most basic computer activity. I spend my fair share of time working with Photoshop, playing Internet radio, using MS Office stuff, running programs in the back ground like webshots and anti-virus software - all of which can be accomplished singled handed by a higher speed Athlon XP. They key to a faster system isn't just the CPU, but also the disks, RAM, and video card involved.

 

[Buck]

Quick question for lab-mouse-boy, what version of Photoshop do you normally use?

 

[zx]

I miss the days when I could buy two Celeron 400s and an Abit BP6 for about the same price as a high end P2/P3 and the performance we better. But now that SMP motherboards are so expensive ($190 for a Tyan Tiger MPX), and that MPs cost significantly more than an XP, it's a luxury I think I'm priced out of.

You can try to build a dually system with cheap XP's to bring the price down. But the mobo's are just too expensive. Or you can upgrade your MP's on your current mobo. Even better, upgrade with XP's .

 

[Mercutio]

http://www.intel.com/design/chipsets/e7505/ is something to consider if you honestly decide you could use smp. like the others, i believe it's a not-so-necessary luxury. put the money in a 9800 pro or something

Or something that would actually make your computer faster, like more RAM or an X15.

From what I've read, I think even most of the gamers here would be really happy with a $180 Radeon 9600.

 

[jtr1962]

Thinking about this topic my opinion here is the same as it is regarding extreme overclocking(not to be confused with regular overclocking which is really just running your processor at the best speed it can do without any fancy cooling)-why bother? In six months a single processor system will be just as fast, and six months after that a low end system will perform the same. Maybe when we just can't make a single processor any faster because we can't shrink things further then dual and multiprocessor systems will start to make sense. Until then I really don't see the point other than bragging rights to having the fastest machine made for a few months. Better to invest the money in more RAM or a faster boot drive. We're at the point now where even the slowest processor is faster than needed for about 99% of the population. I wish manufacturers would try to design more systems which are silent and cool while being fast enough for most people.

 

[Adcadet]

Quick question for lab-mouse-boy, what version of Photoshop do you normally use?

The name's "Adcadet," or even "Andy" if you wish. And it's a rat.

I use Photoshop 7.0. Why?

 

[Buck]

Quick question for lab-mouse-boy, what version of Photoshop do you normally use?

The name's "Adcadet," or even "Andy" if you wish. And it's a rat.

I use Photoshop 7.0. Why?

I ask, Andy, because Photoshop 6.0 had some optimization problems with Windows XP. Once you installed the most recent updates for 6.0, it would finally work like it should. Otherwise, you're left with a very slow application, regardless of the hardware performance capabilities. However, since you're beyond that limitation, we'll skip the subject.

 

[Adcadet]

ah. Ya know, now that you mention it, Photoshop 5 seemed OK, 6 seemed really slow and 7 is just fine.

 

[Handruin]

For light Photoshop work as you've mentioned, I doubt you would truly appreciate the benefit of a dual rig. I think each of the filters and functions in Photoshop vary in how much they utilize the SMP ability of your machine.

What I mean is that a Gaussian blur may take slightly less time to apply the filter where as a radial blur set to "best" could take far less time on an SMP system. One filter may have better SMP utilization than the other.

I would agree with Mercutio in that an X15 and more RAM would "feel" more significant to you than a second CPU. Where you would see the difference with this is during an automated "action" in photoshop that does plenty of opens, resizes, filter, and saves.

I haven't tested this; I'm only going on a hunch.

 

[e_dawg]

Is there any way to get Windows (XP) to log CPU, memory, and disk usage? It'd be interesting to see how often I max out my current system. If it never goes >80%, perhaps a faster FSB would be in order. Or perhaps a faster HDs (I do notice that I'm often waiting for things to load like Word and Mozilla....I wonder if I'm waiting on the HD).

Run perfmon to monitor your CPU, memory, and disk usage. I just upgraded my Athlon 650 to an XP2200 + Soltek 75FRN2-L nForce 2 mobo. I am a little disappointed, to tell you the truth. My biggest performance problem is that I get these unresponsive "frozen CPU" periods where it seems like a rogue process has pegged the CPU at 100% or there is some heavy swapping going on when I am playing UT2003. It's definitely not swapping, because I have 512 MB of RAM and monitor my memory usage ... but what disk I/O is definitely tying up the kernel (I log disk activity with perfmon, and these "frozen moments" correlate with disk access).

This system uses an all-IDE storage subsystem. Tried 2x WD600BB in RAID 0 (current setup) and a WD800JB for my system drive with a WD153BA for my data/mp3, etc. Everything is on separate channels.

I seem to notice this problem less with my other system, which is an XP1800 + A7N266-VM nForce 1 system with an Atlas 10k3 SCSI system drive. IMO, my IDE setup is the problem here. Say what you want about how well IDE drives do on SR's OfficeMarks, but it appears that IDE setups are less graceful with multitasking than SCSI setups.

The only other thing I can think of is perhaps the nForce2 southbridge sucks (I've had the "pleasure" of VIA's 686 southbridges before; is Intel the only one that can get it right?).

 

[e_dawg]

Regarding perfmon, once you set up the counters you want, to recall it quickly for future use, save the settings in the MMC as perfmon1.msc (or whatever you want to call it). Then, when you want to run it in the future, just Run: perfmon1.msc.

 

[e_dawg]

... but what disk I/O is definitely tying up the kernel (I log disk activity with perfmon, and these "frozen moments" correlate with disk access).

What I meant to say was...

... but what is really happening is that disk I/O is definitely tying up the kernel

(i.e., it is nothing to do with a rogue process, swapping, etc. ... although swapping implies disk I/O, the disk I/O is not from swapping, but rather from Winamp playing mp3's)

 

[Pradeep]

The problem with current option of dual Athlons is that the MP and MPx chipsets deliver turdy mem bandwidth, compared to the single CPU chipsets now available (nFarce2, etc). So for mem intensive stuff, dual Athlons really won't help much.

Things should get much better with Opteron/Athlon 64, however they will need time to ramp up clockspeed before they become attractive from a pure performance standpoint.

 

[time]

E_dawg, what version of the nForce driver set are you using?

 

[Adcadet]

this AM I ran perfmon while doing my usual early AM surfing and whatnot (I turned off folding@home and prime95). I was pretty shocked to see that while my page depth (HD) was usually very low (avg. <1) and my CPU usage was pretty low (usually <20%, occasionally spiking to ~75%), my memory usage was often pretty high - I would often see spikes of over 75%. I'm wondering if this is good evidence that for a lot of what I do, I'm memory-limited. If my morning workout is often putting the memory usage at 75%, I would assume that when I really get going in the late PM it's the memory holding me back.

 

[CougTek]

Buy a gagabyte of RAM while it's still relatively cheap. This Fall, it won't.

 

[Howell]

I'm wondering if this is good evidence that for a lot of what I do, I'm memory-limited. If my morning workout is often putting the memory usage at 75%, I would assume that when I really get going in the late PM it's the memory holding me back.

It's much easier to use task manager to make this determination. After using your computer normally or heavily for the day, pull up the performance tab. Look to the lower left. If Peak Commit Charge is not more than the amount of RAM you have installed you are OK. If it is more, then you have been hitting the HD cache.

All of this is outside of photoshop because it uses it's own caching scheme. I don't know what the guideline on that should be.

 

[time]

This is as good a time as any to admit to something.

Some time back, I applied the AMD AGP (LargePageMininum) patch to a few computers after experiencing the problem with one.

It seemed like a good idea at the time.

A few weeks ago, I was cleaning out some papers and came across the MS Knowledgebase article about it (Q270715). On re-reading it, I noticed this little gem: "The workaround for this issue prevents Memory Manager from using the processor's Page Size Extension feature and may affect the performance of some programs, depending on the paging behaviour. This registry value also limits non-paged pool to a maximum of 128MB instead of 256MB."

I had wondered idly why increasing RAM to 512MB on a PC here hadn't boosted performance anywhere near as much as I expected, particularly with Photoshop running. A quick look in the registry revealed my pawprints:

HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Session Manager\Memory Management
LargePageMinimum: 0xFFFFFFFF

Deleting it transformed the PC. Programs loaded in about half the time. Photoshop was snappier. Oh dear ...

The trouble is, I was so used to Microsoft's whining warnings that they attach to every patch/fix, I just ignored it and charged on anyway, without absorbing what they were saying.

 

[e_dawg]

E_dawg, what version of the nForce driver set are you using?

2.03

 

[e_dawg]

this AM I ran perfmon while doing my usual early AM surfing and whatnot (I turned off folding@home and prime95). I was pretty shocked to see that while my page depth (HD) was usually very low (avg. <1)

Yeah, that is why Eugene revamped the IOMeter suites to use lower queue depths when he was still using IOMeter for desktop performance testing. Then came IPEAK, and a "new old fashioned" way of using typical application usage patterns (as Winbench did, although run at a lower level) to test desktop performance. While trying out IPEAK at the beginning, Eugene noticed that typical application usage patterns showed ver low queue depths as well -- something that made our habit of looking at IOMeter test results at a queue depth of 256 seem rather foolish.

I'm wondering if this is good evidence that for a lot of what I do, I'm memory-limited. If my morning workout is often putting the memory usage at 75%, I would assume that when I really get going in the late PM it's the memory holding me back.

As Coug mentioned, it couldn't hurt to pick up some more memory at these ridiculous prices. I don't think you would see much of a benefit now, but I'm sure your memory usage will increase a bit every year, so it's good to plan ahead -- especially when you're dealing with God's gift to DRAM manufacturers: Photoshop.

 

[Adcadet]

Yeah, I remember the IOMeter switch to lower que depths, but I still did not expect my avg. que dept to be <1. I guess I was expecting 3-5. But I guess this is a good thing. (FYI - my boot drive is on a Cheetah 18XL, and my swap file is on another 18XL).