Mac OS X Intel Build 8F1111

[mad jew]

My iMac G5 1.8Ghz has a 600Mhz FSB, 512KB L2 cache, 1GB PC3200 DDR RAM

My PentiumM 1.7Ghz has a 533Mhz FSB, 2MB L2 cache, 1GB PC4200 DDR2 RAM

They're comparable most of the time. Sometimes the iMac is faster, sometimes the PentiumM. Sometimes the difference is huge - eg. OpenGL is hugely faster on Windows, anything media related is normally faster on the Mac.

One is apparently PPC and the other is X86. They're running two very different operating systems. Are you saying they're similar in terms of speed or not? Do you really think AltiVec has all that much to do with any speed difference anyway?

Look, the best real-world way of measuring the effectiveness of AltiVec is to sit two similar machines side by side with one having a G3 and the other a G4. When I did this, I thought generally speaking the differences were negligible for CPU related tasks. That's merely my experience and I'm more than happy to accept that you've experienced otherwise but bringing PCs into the equation is slightly misleading and ultimately pointless.

 

[aegisdesign]

USB wasn't even around when they went PowerPC--that was in the mid-90s. But Apple (with the iMac) was one of the first, if not THE first, PC makers to widely adopt USB.

I've a Toshiba laptop from 1996 that has USB on it. I couldn't actually use the port till late 97 though when Microsoft finally got around to fixing USB support. But then I'd have to use Win95B and it only sort of worked and at the time I was using NT which never had USB support until they came out with Windows2000.

Win98 was about the first OS to have it working acceptably. The first iMac was mid 1998. The claim that the iMac introduced it is bogus. Apple would not have introduced the iMac with USB if there weren't already USB printers or other peripherals.

 

[mdavey]

Of course. It's even capable of splitting tasks out to other Macs on the network.

Now, that is an interesting idea!

Most consumers will probably be happy with the performance on the January MacIntel boxes most of the time (and very happy with the "it just feels more snappy" effect every time they update the OS).

I'm guessing that at least initially, most people buying a January MacIntel box for professional use (if there is even anything suitable available) will probably have a PPC Mac box elsewhere on their network. How cool would it be to pass PPC-optimised tasks to the PPC box?

One of the great things about X11 in Unix is the ability to remote display individual applications from one machine on another. Does anyone know if Aqua offers a native equivilent?

 

[aegisdesign]

One is apparently PPC and the other is X86. They're running two very different operating systems. Are you saying they're similar in terms of speed or not?

At most tasks they are similar. Intel is faster at integer code and OpenGL on Windows at least. PPC is slower at integer and way slower at OpenGL but faster at anything that can be vectorised into neat little cache fitting routines.

Do you really think AltiVec has all that much to do with any speed difference anyway?

Yes.

Look, the best real-world way of measuring the effectiveness of AltiVec is to sit two similar machines side by side with one having a G3 and the other a G4. When I did this, I thought generally speaking the differences were negligible for CPU related tasks.

At one point I had a G4 Ti Powerbook 400Mhz and a G3 iBook 500Mhz. Both have the same FSB speed, same graphics hardware, same memory speed. The G4 would rip in iTunes at about 4.5 - 6x, the G3 at only 2-3x.

Here's a graph from back in the days of G4 upgrades at 400Mhz...

See that long bar, that's a G3 running Photoshop without AltiVec.

That's merely my experience and I'm more than happy to accept that you've experienced otherwise but bringing PCs into the equation is slightly misleading and ultimately pointless.

This is a thread about Intel Macs and how fast Rosetta G4 emulation will be no? It's entirely relevant as to how much of a difference AltiVec makes and how fast an Intel chip can run the same vector tasks and how fast it can emulate G4 AltiVec code.

 

[aegisdesign]

One of the great things about X11 in Unix is the ability to remote display individual applications from one machine on another. Does anyone know if Aqua offers a native equivilent?

No. FCP (or more precisely the DVD Studio encoding, Compressor and Shake) uses XGrid on up to 10 network nodes.

You could I supose use Apple Remote Desktop to run a remote PPC Mac headless but that's more like VNC than X.

 

[mad jew]

At one point I had a G4 Ti Powerbook 400Mhz and a G3 iBook 500Mhz. Both have the same FSB speed, same graphics hardware, same memory speed. The G4 would rip in iTunes at about 4.5 - 6x, the G3 at only 2-3x.

Interesting. I remember using the same test and getting roughly 7x or 8x for both my G3 and G4 iBook. I remember thinking at the time that it was a pretty interesting result.

Here's a graph from back in the days of G4 upgrades at 400Mhz...

See that long bar, that's a G3 running Photoshop without AltiVec.

I'm not necessarily arguing that AltiVec technology isn't effective, but rather that it hasn't been implemented as much as it could have been. Photoshop is obviously one of the few apps that has proper support for AltiVec which would explain your results (assuming the other update cards are roughly the same in terms of other specs.

This is a thread about Intel Macs and how fast Rosetta G4 emulation will be no? It's entirely relevant as to how much of a difference AltiVec makes and how fast an Intel chip can run the same vector tasks and how fast it can emulate G4 AltiVec code.

Agreed, it's relevant how fast an X86 chip will run AltiVec emulation but there are too many other factors to take into account before we can bring a standard PC running Windows (I assume) into the equation.

 

[SiliconAddict]

Erm, you presumably weren't aware that the Intrepid2 controller in the new Powerbooks runs at a full 333Mhz then?

Not everything has to go through the FSB. A lot of the time Intrepid2 will shove data from RAM, GPU, Network or the harddisk straight into RAM via DMA at a full 333Mhz without involving the FSB at all. ie. at exactly the same speed as a Pentium-M with PC2700 DDR2 RAM.

FSB speed isn't as important as you imagine.

Umm no I wasn't. If this was the case why aren't we seeing greater performances increases on the latest powerbook. The proof is in the pudding as they say. Certainly faster access to RAM with DMA is going to help but it doesn't change the fact that overall system perfornace is tied to the FSB. ONe thing I noted on Apple site about the controller. . .

and the memory interface is synchronized to the MaxBus bus interface at 167 MHz.

Link

That right there is going to throttle whatever performance benefits you get.

 

[geerlingguy]

Bits may run a bit quicker though now that they have Quicktime 7.0.4 as intel native. I'm not sure however how much in FCP is Quicktime and how much is their own code.

In my understanding, ALL playback, rendering, and video-handling is done by QuickTime. FCP is just a really fancy way to harness the raw video-handling capabilities of QuickTime, by letting you add FX, transitions, color corrections, etc. by means of plugins and tools.

The only thing (as far as I can tell) that might slow down FCP is running its plugins and effects; QuickTime has some built-in effects (that would, presumably, be native for Intels), but FCP doesn't use them. The rendering of these effects and transitions would have to be run through Rosetta.

Of course, I could be wrong... but it would seem that compression, importing, and playing with individual clips (without effects) will be just fine on an Intel, even if you use an older version of FCP or FCX.

 

[gnasher729]

Except processing power will probably scale much faster on the x86 front than PPC. With dual core *mobile* chips, I really see a good future for Intel Macs.

I think you missed the point completely.

If Rosetta indeed supports Altivec (which I doubt) then emulated Altivec code will run slower than emulated G3 code. That's why emulating Altivec is utterly pointless and counter productive. On G4, Altivec is faster. On Rosetta, it is slower. No matter how fast the Intel CPU is, emulating Altivec is slower than emulating G3 code on the same Intel CPU.

 

[gnasher729]

Because it CAN emulate Altivec code (Altivec being the engineering term "Velocity Engine" being the marketing term I remember), does that mean non-Altivec code HAS to be emulated the same way? If not, then why is there any reason for Altivec compatibility to be a deterrant?.

Most Altivec-aware applications have two sets of code: One for running on machines with Altivec, and one for machines without Altivec. Like

if (computer_has_altivec)
run_altivec_code ()
else
run_non_altivec_code ()

If Rosetta doesn't support Altivec, the non-altivec code will be executed. If Rosetta supports Altivec, AND Altivec emulation is slower than emulating non-altivec code, then the application will execute the Altivec code even though it is slower!

Another problem: Altivec is very very fast for things that are very suitable for Altivec. Sometimes it is used to do things where it is not so good: The original programming problem may not be very well suited for Altivec, and a programmer finds an alternative method that does five times more work but uses Altivec, which does five times as much work at ten times the speed, making it altogether twice as fast. Better than nothing on a G4.

But if you run this code under Rosetta, the application will do five times as much work, WITHOUT the advantage that Altivec executes faster, so the application will run considerably slower than it needs to.

 

[AidenShaw]

first to "force", not first to adopt

USB wasn't even around when they went PowerPC--that was in the mid-90s. But Apple (with the iMac) was one of the first, if not THE first, PC makers to widely adopt USB.

USB was common on mainstream PC motherboards and systems for a year or so before the iMac.

The big difference was that the iMac dropped all the other ports, *forcing* the use of USB.

The key reason that Apple was able to do that was that the iMac was introduced around the time of Windows 98 - which had native USB support in the OS (Win95 needed addons, which were often flaky).

Hardware manufacturers were developing USB peripherals for the Win98 market - when suddenly the iMac appeared which required USB peripherals. Many of them quickly adopted translucent plastic (ugh, remember that fad?) and sold them to iMac buyers.

There's no way that USB peripherals would have been ready for the iMac if they hadn't already been in development for Windows systems....

The USB parade was well underway in 1998 - Apple just cut into the front of the parade right before it came around the "Win98 corner", and to the naive viewer it looked like Apple was leading the parade. Not even close....

 

[aegisdesign]

Umm no I wasn't. If this was the case why aren't we seeing greater performances increases on the latest powerbook. The proof is in the pudding as they say. Certainly faster access to RAM with DMA is going to help but it doesn't change the fact that overall system perfornace is tied to the FSB.

I've not seen any decent benchmarks for them yet, Have you?

ONe thing I noted on Apple site about the controller. . .

Link

That right there is going to throttle whatever performance benefits you get.

No. That's normal for DDR(2) RAM. Access is sync'd to both the rising edge of the clock and the falling edge of the clock hence the first D in DDR. 2x 167Mhz = 333Mhz.

 

[sluthy]

Most Altivec-aware applications have two sets of code: One for running on machines with Altivec, and one for machines without Altivec. Like

if (computer_has_altivec)
run_altivec_code ()
else
run_non_altivec_code ()

If Rosetta doesn't support Altivec, the non-altivec code will be executed. If Rosetta supports Altivec, AND Altivec emulation is slower than emulating non-altivec code, then the application will execute the Altivec code even though it is slower!

Ahh, that makes sense.

 

[AidenShaw]

Take a look at how QuickTransit works....

I think you missed the point completely.

If Rosetta indeed supports Altivec (which I doubt) then emulated Altivec code will run slower than emulated G3 code. That's why emulating Altivec is utterly pointless and counter productive. On G4, Altivec is faster. On Rosetta, it is slower. No matter how fast the Intel CPU is, emulating Altivec is slower than emulating G3 code on the same Intel CPU.

And you're talking without understanding how QuickTransit emulates code.... You're assuming a naive one-for-one mapping of each PPC instruction to an equivalent sequence of x86 instructions. It doesn't work that way.

QuickTransit - the technology that Apple is licensing under the marketing name "Rosetta" - does a high-level decompilation of the PPC code, and then produces an optimized x86 stream. It can find a faster output than blindly doing a permute just because it found a permute in the PPC stream.

The better question is not whether QuickTransit can interpret AltiVec instructions in the input PPC stream, but whether it can produce SSE2 instructions in the output x86 stream. If that's the case, then QuickTransit would have a big advantage. In fact, it could even produce SSE2 output from G3 input - in essence rewriting old G3 code to "use" AltiVec.

Take a look at https://forums.macrumors.com/posts/1928394/ for a longer description of this issue, at https://forums.macrumors.com/posts/1924389/ for another, and at http://www.transitive.com/technology.htm for a technical brief on the QuickTransit product.

 

[aegisdesign]

I'm not necessarily arguing that AltiVec technology isn't effective, but rather that it hasn't been implemented as much as it could have been. Photoshop is obviously one of the few apps that has proper support for AltiVec which would explain your results (assuming the other update cards are roughly the same in terms of other specs.

Almost all creative apps use AltiVec extensively for their filters and effects. Those apps will also be the ones that take the longest to port over to native Intel code. If pros are to use Intel Powerbooks then the apps they want to use have to run at least as fast as on their old PowerPC hardware or there is no point in 'upgrading'.

Then again, most creative pros I know use desktops so until there's something faster than a quad g5, we'll be using those.

 

[shamino]

I've always wondered just how effective Altivec is though. I've always thought it was more of a marketing term rather than an engineering one.

SIMD instructions in a processor are critical to many classes of apps. Especially those that involve gaming, image, video and audio processing.

As for effectivity, Apple has been designing OS X to take great advantage of it. The Accelerate framework (introduced in 10.3) allows apps to use AltiVec in a way portable to other processors. Many other frameworks (like CoreImage) use AltiVec internally to speed things up. Any app that uses these frameworks takes advantage of the feature.

As for apps that call it directly, there aren't that many, but that doesn't mean they don't take advantage of the chip.

 

[shamino]

Good question. FCP is definitely an Altivec intense application to say the least. ...

But if it channels all its AltiVec code through the Accelerate framework, all that code will end up as SSE code when compiled for Intel or when running through Rosetta.

The fact that an app uses AltiVec doesn't mean as much as many people here seem to think. Unless you know specifically how the app is using it, you can't say how well/easily it will port to Intel.

 

[shamino]

Having Rosetta _support_ Altivec would actually deter me from ever buying an Intel Mac. Emulating a vec_perm instruction takes sixteen loads and stores, and in good have Altivec code half the instructions will be vector permutes. Expect emulated Altivec code to run much slower than emulated G3 code.

You assume the AltiVec code will be emulated as ordinary sequential instructions. If they are morphed into SSE instructions, performance will be better than you're assuming.

The initial Rosetta demonstration at WWDC should've demonstrated clearly that gut-feeling assumptions can't be trusted in situations like this. I'd wait to see the results before making sweeping claims about how you think it will have to suck.

 

[Thomas Harte]

Almost all creative apps use AltiVec extensively for their filters and effects. Those apps will also be the ones that take the longest to port over to native Intel code.

Surely you can't say that without knowledge of how the applications were written? For example, GCC has vector extensions that allow an application to make some use Altivec or SSE without any native instructions being used. It supports all basic arithmetic and binary operations so should be enough for many applications provided the developers have chosen to use it.

 

[shamino]

Win98 was about the first OS to have it working acceptably. The first iMac was mid 1998. The claim that the iMac introduced it is bogus. Apple would not have introduced the iMac with USB if there weren't already USB printers or other peripherals.

You're memory is flaky.

Yes, Win98 was the first version of Windows to have useful USB support. Nevertheless, it was virtually impossible to find any USB hardware to actually use the port until after Apple started shipping iMacs.

At that time, I was exclusively a PC user, and I remember explicitly telling people to not bother paying extra for motherboards with USB ports, because there's nothing to plug into them anyway. I spent a lot of time shopping in stores for any device that could do something useful with these ports and never found anything.

At the time Apple released the iMac, there were almost no USB devices. There was one Epson printer, a Zip drive, and a floppy drive. All introduced at the same time as the iMac and sold as Mac peripherals (encased in Bondi-blue plastic, to underscore the point).

Until the iMac, PC users were perfectly happy using their PS/2, serial and parallel ports for peripherals. The fact that PC motherboards had ports and Win98 had an infrastructure for USB device drivers does not change this.

 

[laughingboy]

USB and old Powermacs

I should have been clearer. Apple is usually good about creating good computers, but if I recall correctly, their first models after a big switch have been lacking in features, which in the past has cost them, big time. Many say this is why Apple didn't "boom" like Windows because Apple's first computers were lacking in features.
I think after the switch to PowerPC their computers were lacking things like USB and such. (this is coming from someone who has one, can't attach anything to it )

As an aside, I was able to stick a vanilla PC USB PCI card into a Powermac 7500 (2nd generation PowerPC circa 1995) to attach printers, scanners etc under OS 9.

 

[AidenShaw]

At that time, I was exclusively a PC user, and I remember explicitly telling people to not bother paying extra for motherboards with USB ports, because there's nothing to plug into them anyway.

I hope that none of them held that advice against you...

Yes, Win98 was the first version of Windows to have useful USB support. Nevertheless, it was virtually impossible to find any USB hardware to actually use the port until after Apple started shipping iMacs.

So, in the month or two between Win98 intro and the iMac shipment, Apple changed the world?

How about this press release:

http://www.microsoft.com/presspass/press/1998/jun98/availpr.mspx

REDMOND, Wash., June 25, 1998 — Microsoft Corp. today announced the availability of the Microsoft® Windows® 98 operating system...

The native support for USB in Windows 98 also presents hardware vendors with a tremendous opportunity to deliver exciting and innovative peripherals, from realistic "force feedback" joysticks to digital cameras, to consumers. USB dramatically simplifies the installation of these peripherals, delivering to consumers true "plug and play" functionality. Currently, a breadth of USB devices are being developed by the industry's leading manufacturers. According to David Fair, chairman of the USB Implementers Forum (USB IF) and technology initiative manager at Intel Corp., consumers can expect to see a steady increase in the number and type of USB devices available, with more than 150 new products due to hit the market over the course of the next year.

"Thanks to the support for USB in Windows 98, a wide array of easy-to-use USB devices are poised to flood the marketplace," Fair said. "USB technology enables a whole new spectrum of computing options that will help lower the barriers to broader PC adoption, expand the market and create a better end-user PC experience."

It's "flaky" memory (or flaky logic) to attribute the "flood" of USB devices to the iMac.

Apple did influence a lot of manufacturers to change their production lines to use tacky translucent plastic - but the devices were already on the way when Jobs cuddled his "space egg" in May 1998....

 

[mdavey]

QuickTransit ... does a high-level decompilation of the PPC code, and then produces an optimized x86 stream.

I've only quoted a tiny bit of your text, but found your post to be most informative - excellent description and including the image really helped. Thanks!

 

[jholzner]

You're memory is flaky.

Yes, Win98 was the first version of Windows to have useful USB support. Nevertheless, it was virtually impossible to find any USB hardware to actually use the port until after Apple started shipping iMacs.

At that time, I was exclusively a PC user, and I remember explicitly telling people to not bother paying extra for motherboards with USB ports, because there's nothing to plug into them anyway. I spent a lot of time shopping in stores for any device that could do something useful with these ports and never found anything.

At the time Apple released the iMac, there were almost no USB devices. There was one Epson printer, a Zip drive, and a floppy drive. All introduced at the same time as the iMac and sold as Mac peripherals (encased in Bondi-blue plastic, to underscore the point).

Until the iMac, PC users were perfectly happy using their PS/2, serial and parallel ports for peripherals. The fact that PC motherboards had ports and Win98 had an infrastructure for USB device drivers does not change this.

Yep! Exactly how I remember it. I wanted to say the same thing but was too lazy to write out the whole response. I remember that the first usb printers were just parallel port cables on one end and usb on the other nad some software to make it work.

 

[jholzner]

I hope that none of them held that advice against you...



So, in the month or two between Win98 intro and the iMac shipment, Apple changed the world?

How about this press release:


It's "flaky" memory (or flaky logic) to attribute the "flood" of USB devices to the iMac.

Apple did influence a lot of manufacturers to change their production lines to use tacky translucent plastic - but the devices were already on the way when Jobs cuddled his "space egg" in May 1998....

Right, and not flaky at all to refer to a Microsoft press release for proof that Apple didn't do squat