How is your 15 GHz Pentium 5 doing?

civ_king

Deus Caritas Est
Joined
Mar 9, 2006
Messages
16,368
Users can expect to see the processing speed of Intel's desktop processors hit 15 GHz and that of wireless device and PDA processors hit 5 GHz by 2010, the chip maker's chief technology officer said in Tokyo on Wednesday.

The 15-GHz desktop chip, some five times as fast as the company's soon-to-be-launched 3-GHz Pentium 4 chip, will pack one billion transistors, said Pat Gelsinger, vice president and chief technology officer of Intel as he delivered a keynote address to the company's Intel Developer Forum Japan conference in Tokyo.

Gelsinger would not disclose whether he expected these speeds to be seen in Pentium 4 processors or those based around a new architecture. Intel has said previously that the current Pentium 4 architecture is good up to around 10 GHz.

"We haven't been specific," he said at a post-keynote news conference. "You would look a major micro-architecture like the Pentium 4 and it is typically five to eight years that you would operate on that same micro-architecture until you would introduce a major new one. So while I expect that timeline to be fairly similar, we have not laid out a specific new major microarchitecture step that we will be taking."

Going Mobile
In the wireless and PDA space, Gelsinger said, the company is expecting to see speeds rise from the current 400 MHz to 5 GHz over the same eight-year period of time.

That period is longer than the four to five years Intel took moving its desktop processors through the same speed improvements, as there are power concerns that need to be taken into account in designing processors for wireless and handheld devices.

"Desktops today are [consuming power of] 75 to 100 watts and when you go to handheld devices you are typically operating at less than 1 watt," said Gelsinger.

"Obviously, you are optimizing the design for different criteria. So today, if I was going to look at a StrongArm core or XScale core, could I create a 2-GHz or 3-GHz XScale today? Absolutely. Could I do so and deliver the best trade-off of power and performance inside a 1-watt envelope? No. You tend to design the chips differently to live inside different devices."
http://www.pcworld.com/article/106241/when_will_desktop_chips_hit_15_ghz.html
:lmao:
Intel FTW
 
multiple cores on one chip at that time was a pipe dream, and an untested theory.

in terms of raw processing power, a quad core 3Ghz = 12Ghz. That is, all four cores calculating at 3Ghz, can probably do an amount of calculations equal to what one core running at 12Ghz could do. The reason it doesnt go that fast is due to several things, including the speed of your RAM, the motherboard you're using, and so on.

Think about a restaurant kitchen, you have four chefs, all trying to cook at the same time, but your kitchen, or motherboard, only has two or three ovens.
 
Well, no, the better analogy would be that your 4 chefs all need the same knives, you only got one set, and the chef's dont know how to share.

As for the original article, I think the guy who made that statement was talking out his arse. When Intel was releasing their 3gHz P4's they knew damned well that architecture couldnt break past 10 at best. In fact, by the time those chips were being released, the Core architecture was in the planning stage.

If you think about it, even now Intel is already planning chips for 2015. They know appx. what kind of specs they will be.
 
So that's why they're onto multi-core processors nowadays, now it's all about optimizing the data transfer speeds between the cpu cores and ram, I noticed that bus speeds have increased while the core speed remains at 3ghz for single cores.
 
multiple cores on one chip at that time was a pipe dream, and an untested theory.

No they weren't. IBM had even already released the POWER4 by then.

in terms of raw processing power, a quad core 3Ghz = 12Ghz. That is, all four cores calculating at 3Ghz, can probably do an amount of calculations equal to what one core running at 12Ghz could do. The reason it doesnt go that fast is due to several things, including the speed of your RAM, the motherboard you're using, and so on.

Think about a restaurant kitchen, you have four chefs, all trying to cook at the same time, but your kitchen, or motherboard, only has two or three ovens.

That's not the way it is. Processing power doesn't just add up. Any single sequential task will not be executed faster just by adding more processors - it can only be run on one. The vast majority of software used to be written that way. Some must be written that way, as some tasks just cannot be parallelized. In those you can offload I/O, interface, plus the underlying OS tasks, etc, to a different core and thus avoid some context switches/shared processing slowing your calculation, but the performance gain in a dedicated system would be negligible.

The multi-core push is now getting ridiculous. For desktops and other "smaller" computers 2-4 cores are "enough" nor any normal workload (for which people now have unfortunately been trained to demand that crap marketed as "anti-virus"). I'm now seeing systems with far too many "cores" being sold for use with normal "desktop" workloads. Waste. The unnecessary cores are nos supposed to be powered off when not in use, there is however the problem of how the OS schedules tasks between cores to allow for this - it takes time for the processor to adjust power use, and the OS can redistribute processes too fast for gains to be made. Intel markets "turbo-boost" as an effective and OS independent scheme, but I still don't trust their marketing releases.

Oh, and to be fairer with the author of that piece: the P4 architecture eventually did technically achieve high frequencies. It was demonstrated at ~8GHz before it was dropped. The problem was power consumption and heat dissipation. It just wasn't efficient.
 
See Megahertz Myth.

So they got the GHz wrong, but did they get the performance wrong? How does a modern Intel CPU compare to a 3GHz P4? I would have thought that we've got _at least_ five times the performance (given that simply on number of cores you can get at least four times, and each individual core will be significantly faster).

And the "Pentium 5" issue is just a trademark/marketing decision, of course.

It does seem odd that this guy was making such statements about the specific direction of their CPUs (when surely Intel should have had a better idea), but I don't think we have anything to be disappointed about how things have turned out.
 
Processor won't get much faster in terms of Ghz. The higher the speed the lower the response from the most "inner" parts of the chip itself (voltage-wise).

Programmers have to learn how to parallelize their operations, which they don't like to do (belive me, I don't, but it pays big time).

Another thing is most chips have symmetrical cores. So you have 4 cores but they aren't fully utilized if you're just doing the obvious thing like: one core for the main applicaiton loop, another for physics, the third for music and sound. Each core is full with SIMD capable hardware and they aren't used.

The time has come to use SIMDs (Single Instruction Multiple Data, like adding two vectors/tables of data, those subsystems can "crunch" up to serveral items in one cycle as opposed to just on if you're doing it in a standard "loop"). The problem with this is not every task easily convertable. The best example is PS3 and it's cell processor, which in itself, is a very powerful thing. But adapting your algorythm to the raw SPU (only mathematical vector operations core) can be a problem in some situations. If you do manage to do it though - you've got seven of those.

Some guys from IBM (I think it was IBM) managed to hook up 3 of those to render a ray-traced image of a car in full HD with 30fps!

So graphics-wise, SIMD operations are a good way to go. Especially if you have to move lots of units == move around lots of vertices to make up an animation, because essentially that's what's happening.
 
One of Intel's Westmere quad cores turbos to 3.73GHz! Isn't Intel's fastest CPU to date operating a 3.73GHz?
 
Weren't you the one who posted about the 4.8ghz overclocking Intel processor?

Anyway, define 'fastest' cause if you mean clock speed, 4ghz P4 still wins....
 
Weren't you the one who posted about the 4.8ghz overclocking Intel processor?

Anyway, define 'fastest' cause if you mean clock speed, 4ghz P4 still wins....

Yes I did post about a 4.8GHz however massively overclocked≠stock frequency
 
One of Intel's Westmere quad cores turbos to 3.73GHz! Isn't Intel's fastest CPU to date operating a 3.73GHz?

Nope; the Pentium 4 reached 3.8 GHz factory-clocked. Other close contenders include the Pentium 4 Extreme Edition (3.73 GHz), and the Core i5-680 dual-core (3.6 GHz). The Core 2 line topped out at 3.5 GHz with the Xeon X5270.

Clock speeds are increasing, though - check out IBM's 5.2 GHz quad-core.
 
Clock speed is largely irrelevant nowadays though. A high clock but low IPC processor can easily be comparable to a low clock but high IPC processor.

In fact, there is a huge amount of design that goes into a cpu that increases its performance without touching clock speeds. Comparing with only the speeds is like comparing two nice cars by how high their speedometer goes.
 
I have a feeling an i7 (quad) @3 GHz would destroy eight Pentium 4s at 3GHz
 
Exactly. Newer architectures can do so much more with every clock than older ones.
 
wait, if a P4 gets ~2 IPCs and C2D gets ~4 IPC and i7 gets ~5 IPC right?
 
Top Bottom