Was the Pentium 4 poorly designed?

[Random Variable]

Am I the only one who thinks that Intel is masking the Pentium 4's poor design by continually revving up the chip's clock speed?

 

[Drekce]

The Pentium 4 is NOT flawed. The chip was engineered to be able to handle higher clock speeds. The way the architecture is the chip does (in general) a few less functions per clock cycle, but it was made to run at very high speeds.

 

[MCS]

No you are not, but be warned this is flame bait!!

 

[Random Variable]

I know it was designed for high clock speeds, but it has a pipeline that is like eight miles long (20 stages to be exact). BTW, I don't mind "flame bait"

 

[interchange]

By saying something is flawed, do you mean that it is different from it's design intention or a non-ideal use of the same resources?

Either way, there is little that is flawed about the pentium 4. I personally don't agree with the marketing strategy, and therefore do not recommend buying one.

But, there is nothing flawed about a 20 stage pipeline. In theory, if every stage were used every cycle, the longest pipelin would be the fastest. Not only that, but it would also allow you to increase the clock speed. I don't see a problem with a long pipeline if you can maximize the utilization, which is what Intel is having mixed success at.

Therefore, a Quake III benchmark will scream, but a Windows benchmark may not. The architecture is just taylored to different instruction flow -- not flawed.

 

[Random Variable]

Flawed was the wrong word. It just seems that Intel is compensating for the Pentium 4's shortcomings by revving up the clock speed of the Pentium 4 to astronomical frequencies. However, it is amazing that the Pentium 4 is able to function at such high clock speeds without the need for extreme cooling.

 

[dullard]

<< Flawed was the wrong word. It just seems that Intel is compensating for the Pentium 4's shortcomings by revving up the clock speed of the Pentium 4 to astronomical frequencies. >>



This same logic would mean that the 1.0 GHz Athlon is 'flawed' since AMD has already ramped up the speed to 1.4 GHz. The P4 was designed to start at a reasonable clock and ramp up higher than any other chipmaker is currently able to do. We may argue if that is a desireable goal. However, Intel chose that route, and I see the P4 is following that path quite well. Thus, I do not find the P4 (or the Athlon) to be flawed, even if the speeds keep ramping up.

 

[Ryan]

Do you want to know the real idea behind the chip? "Goes really slow, but really fast"

 

[sparks]

The chip is just designed for clock speed. I believe I read somewhere that the P4 core was designed to ramp up to 10GHz, and with the Hyper-Threading technology, I think Intel is going in the right direction. Now if only they were more price competitive.

 

[damocles]

Intel made a "Pit Stop" when the brought out the P4.

For over a year the initial iteration of the P4 allowed AMD to pretty much be the performance Leader. However the P4 was very much a product of design and will 'evolve' over time. When you combine the potential for a very high ramp up with the future prospects we know of (512K L2 cache, Quad pumped 133Mhz) and some we can only speculate about ( more L1 cache, extra FPU unit?) I think it's hard to say that the P4 was poorly designed. It just needed to overcome a certain amount of 'new product inertia'

 

[Random Variable]

Ummm... I said that "flawed" was the wrong word. All I am trying to say is that because of the way the Pentium 4 was designed, it has to operate at extremely high frequencies to compensate for the long pipeline.

EDIT: And it needs more than a 16KB L1 cache.

DOUBLE EDIT: And Intel needs to do something about the Pentium 4's pitiful floating point performance.

 

[dullard]

<< Ummm... I said that "flawed" was the wrong word. All I am trying to say is that because of the way the Pentium 4 was designed, it has to operate at extremely high frequencies to compensate for the long pipeline.

EDIT: And it needs more than a 16KB L1 cache, but maybe Northwood will deal with that.

DOUBLE EDIT: And Intel needs to do something about the Pentium 4's pitiful floating point performance. >>



1) Backwards logic. The P4 has to have a long pipeline in order to operate at extremely high frequencies.

2) I read once that a larger L1 cache would have made performace worse (it isn't the normal use of L1 cache).

3) The offical SPEC numbers for the P4 show it has better floating point performance than the Athlon. Sure it is an artificial benchmark, but if even one benchmark gives the P4 the edge, then it can't be too bad...

 

[Random Variable]

Since it seems that no one is on my side, I must not know what I'm talking about. Yeah, and I know a longer pipeline generates higher frequencies. I'm just tired of this fascination with clock speeds. ( I sound like a marketing executive from Apple)

 

[TunaBoo]

<< Sure it is an artificial benchmark, but if even one benchmark gives the P4 the edge, then it can't be too bad... >>



Ooo, foolish boy.

 

[damocles]

2) I read once that a larger L1 cache would have made performace worse (it isn't the normal use of L1 cache).

Adding too much L1 cache may have an aberrant affect on latency. However it may be possible to add a little more L1

 

[dullard]

<<

<< Sure it is an artificial benchmark, but if even one benchmark gives the P4 the edge, then it can't be too bad... >>



Ooo, foolish boy. >>



I use computers mostly for floating point calculations (Fluent). For complex differential equation solvers, the 1.7 GHz P4 is faster than anything except the Itanium (1 chip / 1 chip comparisons). Thus the floating point performance really isn't too bad. Sure it could be better. Everything could be better. I may be foolish to use the fastest floating point processor, but I value my time.

Some benchmarks show the Athlon better at floating point, others show the P4 is better at floating point. Thus they probably are each doing well.

 

[Odoacer]

I remeber a while ago in some other post someone made a good point about the P4 and how its marketing department took precedence over the engineering department. They ramp up to insane clock speeds to impress the public first, then they start actually making the chip useful. The best example I can think of this is Hyper Threading technology... this is in all P4 cpu's, but it is not enabled in any of them yet (not until Northwood comes out). The current P4 is gutted (that is, many of its capabilities arent implemented); you could say it's more of a introductory model. It's a marketing tactic really, and quite a smart one at that.

Let's see what you AMD zealots have to say once Northwood comes out.

 

[Flatline]

Not to mention the fact that SSE optimized applications run incredibly well on the P4...Intel has always had the muscle to "help" programmers optimize their programs for Intel chips. The P4 does have one problem - it needs extremely high memory bandwidth in order to take advantage of the quad-pumped bus; that, along with (until very recently) its high price have been keeping the P4 out of the race. With .13 micron process and 512K L2 cache, though, it's looking more and more interesting...

 

[Remnant2]

Poorly designed? No, not at all. It's not like the Oregon design team suddenly looked at the chip they produced at said "oh my, look, we forgot to add performance!" It would be my contention that the P4 was truly designed with the .13 process in mind..

the die size is quite large in .18, and this is the "stripped down" model. There was an article in EETimes about how in the design process, the engineers realized that they had to modify or remove several features to be able to make a reasonable-sized and reasonably cool chip in a .18 process

Some of the changed features included:

Cut L1 cache size in half
Removed the 2nd FPU pipeline
Removed the 3rd ALU (left the 2 "fireball" ALUs)
Removed all L3 cache (compensated by boosting L2 to 256k)

if all those changed hadn't been made, I think the initial P4 would have been a real screamer. Currently I don't think there are any plans to re-add any of the removed features, but just increasing the L2 cache to 512k should boost performance by a fair bit.

Basically though, the P4 was built for speed. It can be seriously fast when using optimized code, but on stuff not produced by the intel compiler, it suffers quite a bit.

This building for clockspeed hurt the P4's acceptance tremendously, but the lead the P4 now has in clockspeed on the Athlon is ridiculous -- 600mhz behind is almost like comparing the G4 to the P3. We know that in terms of real performance on current code, it's a wash, but the computer buying public doesn't.

 

[NOS440]

can 't believe how civil this thread is. I can tell you first hand from 9 months of use there is nothing flawed about a P4 it runs everything perfect. I'm sure it could of been better but we could list a million things that could be better with the thunderbird or paly or the P4. This is just plain flame bait.

 

[Random Variable]

Please, don't post anymore replies. I just wanted to see if anyone would defend the Pentium 4 if I said a few bad things about it.

 

[Wingznut]

"I remeber a while ago in some other post someone made a good point about the P4 and how its marketing department took precedence over the engineering department."

Nah... that's just silly.

 

[Ben]

The P4 is a marketing scam in itself. It was designed to achieve big MHz numbers with actual performance as a secondary or maybe even tertiary consideration.

The average Joe Blow doesn't know enough to check out real benchmarks or performance numbers. They buy based on quick references they can understand, the speed of the CPU, the size of the hard drive, how much RAM, etc.

They brag about how they just bought a P4 1300 with a 60GB hard drive, 256MB of RAM, and a 32MB video card.

What he/she doesn't know is that the P4 1300 is slower than a 1GHz T-bird or P3, they got ripped with a 5400RPM hard drive, they got ripped with PC600 RDRAM, and they got ripped with onboard S3 video that shares 32MB of the system memory.

 

[damocles]

The P4 is a marketing scam in itself. It was designed to achieve big MHz numbers with actual performance as a secondary or maybe even tertiary consideration

That's bollox. It was designed to ramp up much higher than any x86 processor currently on the market. The P4s design is starting to pay dividends now. I'm sure the extra MHz helps with marketing but you cannot seriously say that performance was secondary

 

[BFG10K]

Nope. The Pentium 4 was designed to scale to high MHz speeds and to operate with high memory bandwidth. As such it excels in areas where total system performance is needed. And now that its reached 2 GHz you have to say that it competes very well with a TBird 1.4 GHz in terms of performance.

Intel now needs to address the pricing and a lack of DDR memory.