[TETENAL]

http://www.webgraphics.at/spass/mhz_myth_320f.mov

A G4 of half the Megahertz of a Pentium 4 is almost twice as fast. For the G5s we have now that would mean that a G5 at 2.5 GHz is roughly the speed of a Pentium 4 at 10 GHz. Intel is many many years away from reaching 10 GHz. So why is Apple switching processors? Is the Megahertz Myth not true any more? The Pentium 4s didn't reduce the number of pipeline stages since this presentation, did they?

[TheMosco]

Originally Posted by TETENAL

http://www.webgraphics.at/spass/mhz_myth_320f.mov

A G4 of half the Megahertz of a Pentium 4 is almost twice as fast. For the G5s we have now that would mean that a G5 at 2.5 GHz is roughly the speed of a Pentium 4 at 10 GHz. Intel is many many years away from reaching 10 GHz. So why is Apple switching processors? Is the Megahertz Myth not true any more? The Pentium 4s didn't reduce the number of pipeline stages since this presentation, did they?

Its not about what they currently have, its what they are going to have. By the time apple releases x86 powermacs, intel will probably be done with netburst. Pentium Ms don't use netburst so those can used in 2006.

[MallyMal]

Originally Posted by TETENAL

Is the Megahertz Myth not true any more?

I think it's true but only true to a certain extent. I think the truth falls somewhere in the middle of what Apple and Intel have said in the past. Also, keep in mind that the Pentium M runs at less GHz than a Pentium 4 but the Pentium M is faster than a comparable Pentium 4.

[dlefebvre]

Originally Posted by TETENAL

...For the G5s we have now that would mean that a G5 at 2.5 GHz is roughly the speed of a Pentium 4 at 10 GHz...

These numbers are a myth. The differene is not that big.
Here's a real world test we did at work:

We rendered the same multi layers, multi effects, complex After Effects project on 2 systems.

Dual G5 2GHz, 2GB RAM, Radeon 6800. Result : 25 min 43 sec

Dual XEON 3.2GHz, 2GB RAM, NVIDIA QUATTRO FX. Result : 26 min 19 sec

[ghporter]

Wasn't this presentation done in '03? There has been a lot of development in Pentium processors since then, while there has been almost zero development in the PPC world since then. In the video, they point out that all of the processors depicted are made on a 18 micron process, while the current generation of Pentiums is made on a 13 micron process. Reducing the process size is just one way that these Intel chips have been advancing. They've done some other good things like including multiple piplines, improving predictive execution so that it really does work, and reducing the hit from data dependencies in the pipline. They've also done a lot with predictive data fetching; the P4s today have a data cache (for each pipline) that stores the data needed for predicted branches in the instruction stream, so even if there is a branch, the data is available immediately instead of causing execution to wait until it is fetched.

I think the biggest reason for Apple to dump IBM was that IBM didn't seem interested in improving the PowerPC chip family, while Intel has been steadily improving the Pentium family to the extent that Intel's chips are now really more efficient and powerful than the PPC.

[Randman]

Well, in the coming months we shall see first-hand if the megahertz myth is just that.

[TETENAL]

Originally Posted by ghporter

There has been a lot of development in Pentium processors since then, while there has been almost zero development in the PPC world since then.

I assumed that the Pentium 4 is still the Pentium 4 (ie. identical in construction with differences only in size and clock-speed).

[ghporter]

Originally Posted by TETENAL

I assumed that the Pentium 4 is still the Pentium 4 (ie. identical in construction with differences only in size and clock-speed).

Nope. In addition to increasing speeds (they're over 3GHz now), Intel has reduced die size beyond what they'd get just from using a smaller process, making the chips more efficient, and they've made the internal functions of the processors better too. Kind of makes you wonder what IBM's been doing with processor development...nothing, apparently.

[Apple Pro Underwear]

well since the year 2000, the "advances" apple used in technology are now no big deal. Intel has made their own advances such as hyperthreading/dual cores and all that jazz.

so the megahertz myth is true in the fact that the "MHZ" should be only one factor of many to consider.

(and these days, it has reduced even more because a system can be quite great if it reflects what a person uses it for. such as a movie lover who has an average computer but has a top notch DVD burner and monster LCD. or perhaps the music lover who has 400GB hard drive but a weak overall system)

[Randman]

Shouldn't it be the gigahertz myth these days?

[TampaDeveloper]

I didn't see the video.. But keep in mind that a G5 is slower than a G4 at the same Ghz. And a Pentium M is significantly faster than a P4 at the same Ghz.

[Luca Rescigno]

The video is from Macworld New York in 2001, at the introduction of the Quicksilver PowerMac G4s with 733-867 MHz processors.

The video does correctly state that MHz isn't the sole determining factor in performance. At the same time, though, it implies that pipeline stages are the sole determining factor in performance, which of course is equally wrong. Now, I'm not an expert on processor technology, but I know that in general, processors with deep pipelines (20-30 stages for the Pentium 4, ~20 stages for the G5) can go to much higher clock speeds than ones with shallow pipelines (4 for the G3, 4-7 for the two main versions of the G4). I also know that in some cases, a very high clock speed will yield the highest performance, and in other cases, a short pipeline will make more of a difference.

What I don't know is in which situations that happens and why. Still, there are tons of factors in computer performance and neither the MHz nor the number of pipeline stages alone is enough to make any kind of determination of performance. You also have memory communication, cache size, cache speed, die size, and probably a dozen or more other factors that I don't know about.

[TampaDeveloper]

Good point... And I think its important to note an observation... AMD pays alot of attention to benchmarks... Don't get me wrong, I do believe AMD is currently ahead of Intel, but not as much as benchmarks would have you believe. Intel pays more attention to balanced design decisions.

[Eriamjh]

It's not JUST about Megahertz. Its about power. I call it Powerhertz. G5s are too hot. Intel's power/heat to speed ratio is lower than the PPCs was (based on roadmaps). What good is a 10,000 Gigaflop processor if you need a lake to cool it? It still won't fit in a mini or a laptop.

Apple is doing this because they cannot support the aging G4 in powerbooks and they are tired of expensive liquid cooled G5s eating away at margins. A cooler Mac-86 will be better in the long run.

[Millennium]

It is true that clockrate is no real indication of a chip's power, except relative to other processors of the same architecture. However, as Apple has found out the hard way, it is a very good indicator of how well the chip sells.

[dreilly1]

Another thing to consider is the Apple Laptop situation right now. Even if the G5 compares favorably to the state of the art Pentium-based kit right now, it's clear that Apple thinks that the G4 (which is all Apple will be able to shoehorn into laptops for the forseeable future) does not compare well to Intel's current portable chips, or to the future Intel chips on the roadmap. remember Steve's presentation from the other day -- he didn't say that Intel chips performed better across the board, but rather that Intel chips offered more "performance per watt" or some other made-up metric. (heh, when I was learning about processor architecturs back in the day, power consumption was where Intel did worst. I guess they've caught up.)

Apple claims that they will support both PPC and Intel architectures going forward, but I wouldn't be suprised if PPC is reserved for high-end systems that need 64 bits, while the Intel chips go in everything else. Even though I know all about the MHz myth, I don't think it will be possible to market Intel laptops alongside PPC laptops, and convince the public they have comparable performance when the MHz numbers are different.

If nothing else, we know that OS X will feel faster on an Intel system than XP or Longhorn will feel on equivalent hardware. Now that both OS's will be able to run on the same general architecture, Apples to... um... apples comparisons hold more weight. Come the next WWDC, we will see Photoshop bake-offs that show that Photoshop on OS X Intel runs faster than Photoshop on XP with equivalent hardware. Even though I think this move is all about laptops, it's also a shot across the bow to Microsoft, that now they're both on the same playing field. It's the big Steve vs. Bill deathmatch, and may the biggest ego win!

[Luca Rescigno]

I think it's too bad no one mentioned that now it's Apple that has to do the water cooling after they showed that old Bunny Man ad. That would have been funny... you know, a little give, a little take. But I guess they didn't want to talk about currently shipping products in that way.

I noticed that the "performance per watt" chart was measured in integer performance. How much of an indication of overall performance is that? Were they just using integer performance because it makes Intel look really good?

[Drakino]

That video was made at a time shortly after the Pentium 4 came out. Everyone on the PC side knew it wasn't as fast as the 1.8ghz made it out to seem. The initial Pentium 4 came out between 1.3-1.8 gHz. A Pentium III at 1.3ghz was nearly as fast as the P4 1.8. However, several things changed after that point. Intel began speeding up the bus that the P4 sits on, and that helped quite a bit. It is now nearly three times faster. The initial P4 ran a bus of "400mHz", while the current ones run "1066mhz". That number is a bit misleading, but does show the growth properly.

Apple chose to only focus on the pipeline lenght for that presentation, but that is only one of many factors. Modern processors (since the early 90s) haven't just had one pipeline. They have many pipelines that can all be cranking away at something, as long as everything comes out the same order it came in. However Hyperthreading can work around that as well, negating the issue of first in first out to a point. Ironicially, the G5 has just as long as a pipeline as the initial P4, but during it's introduction you didn't hear how a G4 would cream it, you heard "it supports 190 (or so) instructions in flight" making it sound impressive. But that number directly comes from the long pipelines and the number of pipelines the G5 has, since thats where the in flight instructions sit.

*edit*
I forgot my biggest complaint about that video. They always show those videos of the pipeline with both pipeliens running at the same speed. They dismiss the "Well, the P4 runs more then twice as fast clock speed wise" by not running the P4 pipeline graph at 2x the speed of the G4. That would have gone counter to the argument they showed. Again, see spin below.

Apple is very good at trying to spin facts to get people to buy their products. They do mostly stay true to the truth, but they typicially don't expose all the aspects to what they talk about, or they start touching on grey areas. All that really matters is if you can get the work done you need to on a Mac, and for me, OS X allows me to do that better then Windows. So even if my next Mac has one of these huge pipeline chips Apple bashed in 03 in it, I don't care, as long as it runs OS X.

[exactopposite]

the pentium m is a lot faster than a pentium 4. i just read a comparison where they overclocked a pentium m to 2.5 ghz (from 2.13) and it was faster than a 3.6 ghz pentium 4. oddly enough, even when it was overclocked it didn't use any extra power (electricity). the pentium m is quite remarkable in that it it is a very fast processor that doesn't generate much heat. the comparison was done using the same motheboard (requires an adapter for the pentium m) in ther same system, so it was an equal comparison. the pentium 4 is on it's way out and the pentium m has a very bright future.

here's the link if anyone wants to read it. it shows the resutls of various intel and amd chips.
http://www17.tomshardware.com/cpu/20050525/index.html

[MindFad]

Approaching cult mindset level: Grasping At Straws™.

Proceed >

[Kerrigan]

Mac users: welcome to a world where you don't have to fret over processor speeds. There is no longer any reason to come up with elaborate, convoluted explanations as to why an 1.2 ghz G5 is ninety times as fast as a 3.7ghz P4... There is no longer any reason to talk about how a G5 can perform a guasian blur slightly faster than a P4. Accept that Intel is the no.1 chipmaker in the world, and move on.

[mitchell_pgh]

Originally Posted by Kerrigan

Mac users: welcome to a world where you don't have to fret over processor speeds. There is no longer any reason to come up with elaborate, convoluted explanations as to why an 1.2 ghz G5 is ninety times as fast as a 3.7ghz P4... There is no longer any reason to talk about how a G5 can perform a guasian blur slightly faster than a P4. Accept that Intel is the no.1 chipmaker in the world, and move on.

While I still believe the PPC architecture was/is superior in design, x86 has won. I'm tired of fighting. We seem to jump ahead... and then nothing for a few years... then we jump ahead... etc. etc.

Also, while the G5 may be king, most of us aren't using Dual G5 systems... and many of us are using G4 systems...

Perhaps if IBM had a portable G5 and 3GHz chip out with info about new ones on the horizon... I would be singing another tune.

[Millennium]

Originally Posted by Luca Rescigno

I noticed that the "performance per watt" chart was measured in integer performance. How much of an indication of overall performance is that? Were they just using integer performance because it makes Intel look really good?

Whoa; you actually found some docs on what a 'unit of power' is? Do you have a link?

As for using integer performance, this is a trick Apple pulled before. In fact, they pulled it with the G3 versus the Pentium III: in those tests, the G3 came out as much faster.

[mitchell_pgh]

Originally Posted by Millennium

Whoa; you actually found some docs on what a 'unit of power' is? Do you have a link?

As for using integer performance, this is a trick Apple pulled before. In fact, they pulled it with the G3 versus the Pentium III: in those tests, the G3 came out as much faster.

I'm tired of seeing the Graphs with the PPC being faster... only to find out that it's only in a very specific area on the HIGHEST end system, with the wind blowing to the Mac's back... etc. etc.

I would rather ride the tide of the x86.

[olePigeon]

Originally Posted by mitchell_pgh

I'm tired of seeing the Graphs with the PPC being faster... only to find out that it's only in a very specific area on the HIGHEST end system, with the wind blowing to the Mac's back... etc. etc.

I would rather ride the tide of the x86.

I'll miss AltiVec, which is what most (if not all) those benchmarks will be based off of. BioIT will miss it too. Gene sequencing via AltiVec was the best.

[TETENAL]

Originally Posted by Millennium

Whoa; you actually found some docs on what a 'unit of power' is? Do you have a link?

Here you go.

[ghporter]

Originally Posted by TETENAL

Here you go.

Not at all what Millenium was talking about. Anyone with access to a dictionary can look up "power" and find out that the unit for it is named after James Watt, the inventor (supposedly) of the steam engine.

The 'unit of power' Millenium is talking about is what Steve Jobs was using as a comparison between Pentium and PPC chips. I would like to see those charts too.

[SVass]

There seems to be several issues being discussed. Some would like to see a watts per gigahertz chart for the two processor families which would be interesting since speed is mainly limited by cooling capacity. Others should note that speed is like horsepower which used to be measured for cars in a lab with the exhaust system removed and had nothing to do with real life or number of jewels in an analog watch which after a certain number had nothing to do with anything.

An analog computer may be faster than a digital one as all parts are working simultaneously. A parallel digital computer may solve certain problems faster than your typical serial one as in a dual processor computer is faster! In other words, megahertz is one number of many that define a system. sam

[Scooterboy]

Man, sure wish I could buy a Memex! Damn digital world!

Try out the
Memex

Soon we can download a Memex Simulator which looks amazingly Aqua-like

[TheJoshu]

The Myth proved too strong. It's not about anything except the numbers: consumers want to get the highest-rated item for their dollar.

[Scooterboy]

My scooter engine turns at 8,000 rpm, but it's nowhere near as powerful as my '68 Camaro engine was that only revved to 6,000 RPM.

We'll see how these new 3 GHz+ Pentiums in MacIntels and most every other PC stack up against Cell and later multi core PowerPC CPU's. How many gigaflops will the PS3 and Xbox 360 reach?

[Luca Rescigno]

There's no reason to expect a single 3 GHz+ Pentium 4 in PowerMacs. The development machine has a 3.6 GHz P4, but it's not a shipping product. I would expect a minimum of 3.6 GHz dual cores or dual processors in a future Intel PowerMac.

Furthermore, the Cell is really not intended for use in computers. It's great for gaming but that's about it.