Intel iMac performance
 

A few very early tests. Kind of disappointing.

http://www.macaddict.com/forums/topic/76536/1

Yeah, i read that the other day. I was pretty disappointed. The most worrisome part of it, to me, was this:

http://arstechnica.com/reviews/hardw...-coreduo.ars/5

That's a nice review of the iMac G5 vs. iMac Core Duo vs. PowerMac G5 2.5GHz

I think one of the results is the most troubling...perhaps echoing the "hesitation" that the guy mentioned whom i quoted in my last post, and that result is the inexplicably poor performance of the iMac Core Duo on the UI benchmark.

http://homepage.mac.com/mbh0ckey/xbench2.png

what is a user interface test?


Also keep in mind that the very first version of OSX for intel based macs is not going to be the best thing out there. Wait for a few updates and for apple to smooth out all the code. It is going to be a process to migrate to new hardware architecture. However, I think apple made the right move.

UI Test is probably opening and closing windows, moving them around, etc.

That is why I hate benchmarks, a lot of time they do not represent real world use of a computer.

Oh well, I know we are going to get some of the intel based macs at some point so I will wait until my work has them before I make any judgement of my own.

The 2x - 4x faster claims seem largely BS.

Why couldn't apple just move their all in one lines/laptops to Intel and leave the PowerMacs PPC?

IBM had a lot of trouble with the 3 GHz line

Maybe.. but the G5 is 64bit vs 32bit dual core... I'm not opposed to a life of fat binaries, and an option to install either...

There are (currently) only a very few apps that are 64-bit capable. Most of these are science-oriented - e.g. Mathematica
Most people therefore are not seeing any benefit at all from the 64-bit capability of the G5.

I don't think Apple will be introducing replacements for the current PowerMac G5s until the Intel versions are more performant than the current G5 PowerMacs. That might take until the end of the year.

Remember how long it took to do simple things like launch programs in 10.0/10.1? It will improve.

I agree. I think the test results are showing that a lot of the OS code is not optimized and barely native on Intel.

Of course Apple has raised everybody's expectations but I'm willing to give them 1 year of basically no performance improvement as they change over to Intel. Once there are native apps and a major OS update then I'll expect much better performance.

I agree...i think Leopard will be tons faster than Tiger for intel machines, since it'll be designed for intel from the start.

Maybe that's when i'll upgrade my TiBook...

Having just got my Powerbook G4 just before the end of the year I'm quite satified with its performance except maybe in using WinXP under VPC... Something I HAVE to do for work.

But I'm currently using RDC to take care of most of the slowness. External PC box works wonders!

I decided to do the PB G4 because it had been around awhile and I wasn't comfortable going out on the edge with the dual core. Like the other poster above I'll probably use my G4 for a year and see how things are with the intels at that point.

M

I think if I really needed to purchase a new iMac or notebook now, I'd go ahead with one of the new Intel line with extra RAM. My sense is that some of the poor results we're reading about will be corrected with software and OS updates within the next few months. In the meantime, it seems performance is very acceptable (already notable improvements over the G4 notebooks).

Not to mention how do we not know that those benchmarks they are running are optimized for PPC hardware? Remember that HUGE PIII Vs. the G4 debate they had years ago when both of them first came out. People always said that you could double the speed of a G4 and thats what it would be equal to when compared to a PIII. Meaning that a 500Mhz G4 would run equally as fast as a PIII 1Ghz.

In all honestly almost all the benchmarks comparing them was moot. That is because some of the benchmarks were optimized for SSE and MMX instruction sets and others were optimized for RISC and PPC based hardware.

Benchmarks can be tweaked. Run 3DMark on a PC with a Nvidia card and an ATI card, and usually the Nvidia card will score slightly higher because the last release of 3DMark (I haven't actually ran the newest version that was just release like a week or two ago, so this could have changed) was optimized for the Nvidia chipset. I would not concern myself with benchmarks this early in the game. Some people just want the intel based macs to fail, while others are embracing the change. Not to mention apple developers may not have the experience to develope decent benchmarks for x86 based hardware.

I would not worry about some benchmark at this point. Read the specs, look at what the OS can potientally offer you, especially Tiger with dual core processors on a laptop. Centrino technology on a laptop. Hyperthreading technology in processors, dual core processors, etc

Good advice!

I think though I'll hang on to the G4 PB for a while. My speed needs are mainly in the area of VPC and from what I understand there isn't a plan for VPC on the MacIntels and no one has been able to verify if you can simply partition and load XP. Sounds like the answer is no.

Right now the ability to run VPC is killer for me. I can stay using all the nice Mac apps and still have VPC compiling my work stuff. Just have to plan on a few more coffee breaks.

M

For whatever its worth, the intel iMac in the Ars Technica review had only 512 MB of RAM while the G5 iMac had 1 GB of RAM and the Power Mac had 2.5 GB (I believe). That, by itself, is going to cause some differences as well. I don't know why they didn't load it with at least 1 GB of RAM first?

Do the separate cores split the total ram?

IE, did each core in the intel iMac, then, only have access to 256 mb of ram?

I have no idea, but perhaps someone else here knows the answer to that. All I know is that in the specs at the benchmarking portion of the Ars Technica review, it gave

Attachment 1064

So it had half the RAM of the iMac G5 and one-ninth the RAM of the Power Mac G5.

No, there is one physical processor in the iMac that has two cores. Each core will page ram simultaniously for instruction sets. With DDR ram two sets can be paged every cycle. In a way its like a hardware version of hyperthreading technology. Instead of emulating it, there is another physical core now. A lot of the performance is going to be based on how the OS manages resources.

Like myself and others said, just give it a while and I am sure that the intel based macs will be very fast machines.

From what I can tell, it's not. The new iMacs feel FAST compared to the iMac G5s. Certain things do seem to take a second to start, just a very slight pause but that may well be the hard drive spinning up, I haven't taken the time to pin it down. Garage Band, iDVD, and iMovie are definitely snapper and less "laggy" on a core duo iMac than a G5 iMac.

I can't wait to try a universal binary of Final Cut Pro or DVD Studio Pro on the new iMac. If intel can live up to their roadmap, then it's going to be a nice future.

I've personally never liked xbench, but I don't like benchmarks in general.

An improbable profit margin.
 

.
Off topic?
What I have a heard time understanding is the supposed "financial performance" of the Intel-embracing iMac. I don’t see how the figures quoted by MacRumours could be correct:

According to the company <iSupply>, the 17" Intel-embracing iMac which retails for $1299 costs Apple $898 before adding software and boxing it. They estimate that the most expensive component is the new Intel Core Duo chip which is estimated to cost Apple $265 a piece.

That doesn’t jive with usual profit margins, does it?


Best regards,
ArcticStones

My five minutes with an iMac Core Duo 1.83 GHz
 

Everything is very snappy, once windows are open. It does seem to hiccup a little bit right after a task is asked of it, just a split second pause. It's certainly noticeable though.

Safari and page loading was almost instant. I was very surprised with the difference in page loading between my PB and the new intel iMac...the pages really do snap to the screen.

I took the liberty to run BenchJS...and it did real well. The iMac did the entire test in 4.69 seconds...it takes my PB about 14.

So will The next version of OS X (10.5 Leopard I guess?), be overly optimized for the Intel Macs that it won't be that much of a good thing on "older" macs? Like my PowerBook G4, which I just got for Christmas, will 10.5 still be a good update?

Probably yes, if it has features you want. There is an established efficient code base for PPC processors and updating that is much easier than optimizing for a new processor.

PPC hardware is going to be phased out. So eventually like three years from now it will all be intel based. This is going to be a transitional period.

Does anyone find it odd that IBM's PPC chip was the first to break the 1GHz mark but couldn't keep up with Intel and AMD when they got past it?

Nope thats how IBM rolls. They develope something over everyone else and then shelf it to work on something else. Then sell the technology off to another company. Not to mention intel and AMD was making their chips in a way more cost effective way. IBM could never get a handle on that, and the cost never really went down over the years. IBM is more concerned with servers and enterprise business than consumer business, which is why they never really needed to develope PPC hardware much further than it was. It just wasn't a good investment for them, or they at least thought that way.

I want to by a new Mac... My G4 has give up the ghost and my MacMini is struggling! I like the iMac's and don't know what to do!!! do I by the PowerPC model or get an Intel model???? right now all my software is PPC (Quark, Adobe Collection etc) - If I by an Intel Model, I will be chucking away some very expensive software!

Any advice would be appreciated!

The thing with that particular example is was difficult to accurately guage PC performance against Mac performance simply because of the OS difference.

No, the memory controller is seperate to the core of the CPU so therefore they full access to the entire allocation of RAM.

Think about a dual processor system, that is basically what a dual core chip is.

Personally I would go for the Intel simply because it is the newest and the current line and theoretically should have a longer life.

Have a look at Rosetta your software might run under it - http://en.wikipedia.org/wiki/Rosetta_%28software%29

Sorry if this has been covered, I just don't find anything about it.

I've been using a Mac Pro since the first month it was released. I run Parallels Desktop with one or two apps that only exist for Windows. They are insanely fast by comparison to all previous VirtualPCs and SoftPC's I've hacked around with over many years.

Meanwhile, the Mac side performance is erratic, to put it politely. Basically when it boots, or when not much is running, everything is lightning quick. Then the longer I run, the slower it gets. Finally, it acts exactly like Windows. Massive disk thrash, I mean you go to launch or quit Word or Excel and it pounds the disk for 5, even 10 seconds of pure thrash, before responding to whatever you just clicked on.

I've run iDefrag and it helps a lot, but does not change the result.

I added a second GB of RAM, and it helps but does not change the result, it just delays the onset.

I found that if I quit out of apps I am not using, that are VM hogs, like Excel (850 MB), Adobe Reader (800 MB), Parallels Desktop (2 GB), then things zip right along again.

I don't think rebooting has anything to do with it, it simply has to do with the total amount of VM in use.

Almost everything on my system is native Intel according to the Activity Monitor, except for Eudora, and HP All-in-One Communcations (which I was one of the original developers of).

Also, AM shows about 14 GB of total VM. That is just crazy. Apparently programs like Firefox or Safari, as well as Eudora, and all of the MS Office apps, appear to me to leak memory. The longer they are used, the higher their VM totals get, and the more the system is prone to violent and crippling thrashes. They are each in the 500 to 900 MB range, per program. A memory leak in a VM environment such as OSX of course means an ever-increasing swap file size.

Has anyone else a) seen this problem, b) come up with any other solutions other than throwing more real memory at it, or quitting & relaunching apps all the time like we used to do before Switcher, Servant, and MultiFinder? Do I sound old? ;-)

Also, has anyone been using a separate swap partition, or separate swap drive, to work this issue?

Many thanks in advance

I note that RAM usage ("Real Memory" in Activity Monitor) is more relevant than Virtual Memory. It is RAM usage that will (if excessive) cause swapping.

And I note that others have reported problems with Parallels (and, less often, VNC servers) taking up large amounts of "wired" memory.

Note too that Excel (like all of Office 2004) is non-native.

It could be a complete coincidence, but in another thread people are complaining about sudden slowness on their Mac -- both having HP all-in-one software installed.

You might want to try quitting the HP specific processes to see if that is the culprit.

Very interesting. However at the moment, I have HP All-in-One Comm running in background, and no thrashing. The system won't thrash until I get busy in MS Office apps, Photoshop, Adobe Reader, PD runnings Windows, etc. It doesn't seem to matter what I run, it is the VM itself that seems broken.

I was one of the early developers of the HP All In One Mac stuff, but it has been a long time since then, and I have no idea what all they've done with their software.

Right, sorry about the Office oversight.

I fully expect Parallels to use a lot of memory, that is fine, it is a complete OS with several apps running. But much like Mark Choi reports in a parallel thread (from November of 2006), the intense thrash is reproducible without PD around. If I run Safari or Firefox long enough, and/or Excel, Word, PowerPoint, Adobe Reader, and Photoshop, it will happen just the same. I can close all files in all of those apps and the system will still thrash for 5+ seconds at a time, especially when switching back in to one of the big apps, then quitting it. If I go to quit four or five at once, it is a comical barrage of massive disk i/o.

When this happens, Activity Monitor is quite clear about what is happening. Before thrash, as I am running now with relative quiet in the room, I have 50,000 page ins and only a handful of page outs. Once I get all the apps going for a few hours though, I will see for example 300,000 page ins, and even more page outs than that. It seems that in addition to leaking VM, someone somewhere is writing out dead pages that should be freed, not written to disk.

System again is a Mac Pro with 2GB real memory. 27 GB free on the boot partition, 165 GB free on the 'user' partition. One actual user.

Maybe the simple answer is, if I see 14 GB in Activity Monitor for total VM, then I need to keep quitting apps, or buy more memory (again). I admit that a 14 to 2 ratio of VM to real memory is awfully optimistic =) However, why does it take 350 MB of VM each to run Calculator, Dock, World Clock, Activity Monitor, etc.

In the thread by Mark Choi that you refer to (http://forums.macosxhints.com/showthread.php?t=62798) the problem was established to be due to excessive memory use by some of the apps he was using. There is nothing mysterious about getting a lot of disk activity when the system needs to swap because the apps you are running are using more RAM than what you have.

And note that Photoshop is also non-native unless you are running the beta version of CS.
Even when native, Photoshop takes a lot of RAM. But as has been discussed elsewhere on these forums, any app running under Rosetta (i.e. non-native apps) will take a lot more RAM than usual - as much as twice as much.

As I said above, it is best to ignore the amounts shown for Virtual Memory and just look at how much Real Memory is being used by your apps.
An application can allocate a bunch of virtual memory but never get around to using even a fraction of what was allocated. The system doesn't do anything with such allocations until the point where the memory is actually used.

So you should add up the amounts of Real Memory being used by the apps you are running (as shown in Activity Monitor) and see if the total approaches the amount of physical RAM you have installed. If it does, you should expect to get swapping.

I thought that was just the amount allocated to the app. It doesn't actually use that much.

I believe hayne covered that ;)

What timing! This morning Software Update just happened to come up and ask me if I wanted to some updates; I believe they were Java update, Video firmware, Daylight Savings Time, and the last was a Security Update. I said yes, rebooted, installed the new firmware, rebooted again, and I've been testing all day in disbelief. I think it was the Java update that did it. The thrash problem seems to be completely fixed. I have been pounding on everything all day, and can't cause the thrash problem anymore. Still using 14 GB of VM. But for page faults now, I see 64,000 in with only 1400 out. The room is much quieter, and the system is much quicker. If it was in fact a Java problem, that might explain why this problem was so prevalent for some users, and not much if at all for others. It all depends where you browse and how much you leak that VM thinks needs to be written before the app exits.

It does when it marks all the memory pages were scribbled on for temporary storage of endless Java junk! Ever see 450,000 page outs in Activity Monitor?! I would not have believed it either. It was not pretty.

NORMALLY it works as you say. And my system seems to have finally reached that state. 2 GB of VM for Parallels Desktop is no problem if it isn't using all of it. And it's not.

I even have HP All-in-One Comm still installed and running.

I'm glad that the problem now seems to be fixed.
But if it were just a Java problem, then you would have only experienced the problem with apps that are using Java - e.g. as you say, with web browsers that accessed sites with Java applets. But other apps would not be affected. I.e. the RAM usage of your browser (e.g. Safari) would perhaps grow a lot, but that of other apps would not be affected. And if you quit the web browser, the problem would go away.
So the problem would have been evident to you as a browser-specific problem. But you seemed to be talking about large memory use by all (or many) apps, not just the web browser.

Moreover, you could have proved that it was a Java-related problem by disabling Java in your web browser preferences.

Really hard to say. The one constant in all of my different setups is that I always did a lot of browsing with either Safari or Firefox. At the moment, I can run Adobe Reader, same time as Photoshop, HP All in One, Eudora, Firefox, Excel, Word, PowerPoint, Terminal, Activity Monitor, Address book, Calculator, and some other stuff. Oh yeah and Parallels Desktop with Windows XP running a few apps. The system runs right down near zero on Real Memory left, but no big thrash, just a light shuffle when needed. I pushed it enough to get to 2 swapfiles, but the amount of paging is still completely reasonable. I can't prove it was the Java update that fixed it though. Wish I knew.

If a similar thing happens again, you now are in a better position to be able to track down the cause:
- check how much RAM each of your apps is using
- quit apps that you suspect to be causing the problem
- monitor your memory use to see when you start swapping so you can see which app might be the cause

By the way, having near-zero "free memory" is not a problem - OS X considers unused RAM to be wasted RAM and so uses it for disk caches etc when not needed by apps.

Funny you mention the Ram total. What I don't get is this. When I look in Activity Monitor, I can use copy & paste, and move all the numbers into an Excel spreadsheet. Now do a global replace and remove the MB label from the memory numbers, and hand-convert anything in KB or GB to MB. Now sum the real memory column. I get around 870 MB. Why is the system paging, I have 2 GB of real memory.

It shows around 950 MB "active", and another 491 wired. Still makes no sense to swap anything. 525 inactive, as you say for disk cache. But why swap the disk cache? Defeats the purpose of a disk cache, doesn't it.

Bottom line, the massive improvement I reported yesterday is still holding, 70,000 page ins vs only 7000 page outs, but I still see a lot of room for improvement. In my case, if I could simply turn off VM, the system would likely run better.

"inactive" memory is not only used for disk cache - it holds memory that was used by apps but not used recently.

Fair enough. Still, to your point, what is the real memory usage. It is 850 MB, plus 450 wired. Why should any paging occur? It is worst with MS Word by the way.

Well, bad news. After a week's worth of uptime (5 days), the system shortly resumed paging its guts out. It has made 6 swap files, total of around 2 GB. Page faults are back up to 150k in, 190k out. Since all the paging is a massive whack in performance, freezing the whole machine for seconds at a time, I am still unimpressed with the VM management in 10.4 at least on a Mac Pro. If anyone is getting different results, I am all ears as to what they did to get it that way. As far as I can see, this is normal behavior.

Again I have apps using a claimed 850 MB of average real memory size in Activity Monitor, and 2 GB physical memory installed, tested, and working properly. Why it should need an extra 2 GB of swap space is goofy. What ever happened to RM (real memory) Unix?!

The "average real memory" is almost totally irrelevant. What is relevant is the peak or maximum real memory that was needed. If at any time the demand for memory exceeded the amount of physical RAM that you have, then you will get swapping.

Note that wired memory (e.g. as reportedly used by Parallels) is not available for swapping, so relatively high amounts of wired memory will make swapping more likely since there is effectively less RAM available for your other apps.

I suspect that you will find that the VM system is working as designed - that the reason for your high amounts of swap is that the apps you are running are demanding (at some time) large amounts of memory.

Bottom line:
You need to monitor what real memory is being used by the apps you run and be sure to quit any apps (e.g. like the browser in that other thread you referenced above) that are taking up undue amounts of RAM.

(If in your case it is the browser that is sometimes taking large amounts of RAM, you might want to use the AppleScript that I wrote and which was the subject of an article on the main macosxhints site a month or two ago. Thsi AppleScript warns you when the browser RAM usage gets above an amount that you specify.)