Intel's Pentium 4 3.4GHz Processors Reviewed

EconolineCrush writes "In one of the most gratuitous benchmarking indulgences I've seen, Tech Report has tested Intel's new Northwood and Prescott Pentium 4 3.4GHz processors against sixteen competitors ranging from the relatively old school Athlon XP to the opulent Pentium 4 Extreme Edition, with plenty of Athlon 64 action thrown in for good measure. Performance is tested in a wide range of applications, including gaming, rendering, image processing, media encoding, speech recognition, and scientific number crunching. Even if you're not interested in Intel's latest Pentium 4s, the review nicely shows where 18 of the fastest desktop chips from AMD and Intel stack up against each other."

Comments

Speed

[Peden]

While all that processor speed is mighty good, who needs top-of-the-line equipment anymore? The new games all rely on the GFX card rather than the CPU. Any suggestions, other than the fact that Intel is keeping up to Moore's law?

Re:Speed

[nacturation]

While all that processor speed is mighty good, who needs top-of-the-line equipment anymore? The new games all rely on the GFX card rather than the CPU. Any suggestions, other than the fact that Intel is keeping up to Moore's law?

Many non-game apps are CPU bound, and speed is always desired in these situations. Examples include rendering, video compression, SETI@Home, etc. Likely you don't need a faster processor, but it doesn't mean that the business world sees it the same way. Heck, maybe some day these processors will power your graphics card too!

Re:Speed

[Kjella]

Many non-game apps are CPU bound, and speed is always desired in these situations. Examples include rendering, video compression, SETI@Home, etc. Likely you don't need a faster processor, but it doesn't mean that the business world sees it the same way. Heck, maybe some day these processors will power your graphics card too!

Not to mention, many of the games are CPU-bound because of the minimum specs - you can up the gfx from 640x480x16bit -> 1600x1200x32bit, but there's no setting the AI to "dumb -> average -> smart". I'm sure there's lots of interesting ideas in AI (groups, formations, tactics, responses to movement/sound, distractions etc.) or game world design (i.e. things happen to the world around you, not just what's being rendered on the screen) that'd love to have more power to throw at it.

Kjella

Missing 400Mhz....?

[inphinity]

Besides this test being ridiculously comprehensive, did anybody else notice the stat differences between the P$ 3.0 Ghz - 3.4 Ghz?

Or, more precisely, the lack of differences?
I wonder, is this just an inability of benchmark software to challenge a processor at such a high clock speed, or are these processors actually the same thing with shinier packaging?

Thoughts?

Re:Missing 400Mhz....?

[DigiShaman]

Some games (and 3D benchmarks) will be bottle-necked by the video card, leaving the CPU with spare CPU cycles to burn. Also, the benchmark may not require much general processing by the CPU, thus all the burden is pushed over to the video card.

Case in point. I was playing Warcraft3 on my P4 2.8 (with Radeon 9800 Pro). Though my framerate dropped down some at high resolution with 4x anti-aliasing, my CPU was only taxed at 15%. I noticed this after exit the game and looked at the task manager CPU usage stats. I was rather shocked.

Basically, if your a gamer, then your better of spending all your cash on a nice 3D card, RAM amount, then CPU ...in that order.

Initial observations

[Pidder]

A quick glance on the system setups shows that they have used RAM with almost the same CAS-latencies in all the setups. The AMD CPUs benefit from low CAS to a greater extent than the P4. When an Intel fanboy site like Tomshardware wants the p4 to beat the Athlon they usually use very slow ram on the Athlon setup, which is of course overlooked by most consumers.

Re:Initial observations

[drinkypoo]

I've certainly noticed that people are more or less willing to take for granted that two things are true about AMD processors. One, they do more per cycle. This should be clear, anyway, because they have more functional units and the functional units are more flexible. Also they have the credibility of having built a multitude of RISC designs over the last few years, most of which have had x86 emulators on the front of them of course. Well, and the back, it wouldn't do to fetch and never retire. Anyway Two, they are much cheaper than intel processors. Sadly intel outstripped AMD in terms of bus bandwidth some time ago and AMD is just now catching up again with the processors with integrated memory controllers - Since that is separate from the bandwidth used to the north bridge. It seems that HT should give about 1/2 the performance of the P4's FSB, but since it doesn't have to carry information from the CPU to main memory (FX-53 has a DDR 400 dual channel memory controller, which should be plenty of memory bandwidth for anyone. Of course DMA still has to occur via HT but in most cases this should not be a serious problem. (Using system memory for AGP textures will still be slow, though of course still faster than loading them from disk all the time.)

So, the hot AMD processor (FX-51) currently beating up on the hot intel processor. The FX-53 is even more destructive (about 10% faster still) and I doubt that it will be substantially more expensive than the P4 EE 3.2GHz 2MB cache, which is already defeated in the benchmarks by the FX-51.

So yes, with the release of the Hammer-core processors, it is unusual for intel to be able to keep up with AMD these days - As it was with the Athlon before it. Remember when the Athlon's double-pumped bus made it two or three times as fast (in terms of FSB) as the intel processors? And how intel processors had less cache, and typically slower cache? Since the release of the K6 intel has been running scared, even in spite of the K6's many flaws. The Athlon was the real sign that AMD was ready to compete with everyone, that's really an amazingly slick chip and there's a multiprocessor version, so AMD targeted basically every space below supercomputing with that processor, and had good success with sales nearly everywhere. (Actually the K6 sold quite a few units also.)

Heat

[shawkin]

With the case open, this thing runs at 178 degrees. In a practical sense, all the other benchmarks are less important.

It is not going to be easy to cool. It is not likely to be suitable for clustered processing. It is not likely to be particularly reliable.

This article illustrates the diminishing returns of the current Intel CPU architecture and processes. Soon, both AMD and Intel will be forced to explore new designs similar to the IBM Power 5.

Given the time, effort and money involved in developing a new CPU architecture, the near and medium term future may lie with IBM.

Re:The problem with all these new processors is

[beswicks]

That is possibly one of the most bizzare comments to be marked interesting, just because a new form factor comes out, it doesn't mean that the processor companies will dump all the current chips.

True they may have a new package for some of the processors to fit a new slot or modified mb chipset, but that is nothing new, we don't just chuck out all the old work when something new comes along.

c.

Re:Nice In-Place Ad

[nacturation]

Planting an obvious ad in the middle of "journalism" is just wrong.

I don't see anything wrong about it. Imagine if you ran a tech review site and couldn't afford to equip all your various test machines with gigs of RAM each. Wouldn't you approach a company and ask if they could perhaps donate (or at least loan) you the equipment you needed? And, if they did such a thing, wouldn't it be nice to credit them for helping you out?

I fail to see how this is a "plant". It would be suspect if this were a review of sound cards and, right in the middle of the article, it said "Hey, your system needs more memory... purchase Corsair RAM today!" then that would be a plant. It would be no different than somebody comparing operating systems and thanking IBM/Dell/whoever for loaning you the equipment to do a side-by-side comparison with realtime parameter tweaking rather than having to tediously reformat a single machine every time you want to test a new config.

It's the lost art of the professional "thank you".

Re:i havent read the article

[tttonyyy]

does it actually have nay insightfull comments, or things we havent heard of already or is it pretty much like the previous cpu review made:
This Just IN, NEW CPU FASTER THAN OLD CPU

Nothing exciting really. Summary is basically this: Amd for 3D games, Intel for MP3 and DivX encoding (and marginally for some scientific software). At the end of the day it all depends what software is running, so there's no clear way to define which is "better" for the masses.

Nothing changes there, I guess.

Summary of the article

[heironymouscoward]

1. AMD64 is better for games
2. Intel Northwood P4 3.4 is good for general use.
3. Intel's new Prescott is too hot.
4. Whatever you buy will be redundant in 2 months.

Plus ca change, plus ca reste la meme chose.

Throw some G5s into the mix

[oingoboingo]

What I'd like to see in a huge multi-CPU benchmark like this are some Apple G5 systems thrown in too. Decent cross-platform tests are hard to find, but given OS X's UNIX underpinnings, it may be possible to come up with a set of tests that are run on x86 Linux and OS X which have an identical code base, and which do not artificiallly advantage one architecture over the other. One thing I've found since switching to OS X about 6 months ago...the Mac community still lacks a really good site which does solid, rigorous benchmarks of Mac hardware/software...and there are a lot of myths and misinformation doing the rounds on various Mac forums (as there are on PC forums too). A well controlled multi-CPU benchmark including some Macs could go a long way to alleviating this.

Re:Throw some G5s into the mix

[thatguywhoiam]

This site compares Macs to Macs... its sort of useful.

This site actually has a German G5 vs. Athlon benchmark posted right now.

Neither one is like Tom's (good or bad)... but its something.

Pretty poor LinPack performance...

[nickovs]

Taking the opportunity for a moment to troll, flame bait and be an annoying Apple user, I think it's worth commenting how piss-poor the P4's LinPack performance is. The Apple Xserve G5 gets 4.5 Gigaflops out of each of it's two 2GHz G5 processor when running HPC Linpack, as opposed to the 3.4GHx P4 "Extreme Edition" which peaks at just 1.3 Gigaflops. Anyone looking to do serious scientific calculations rather than just playing Quake should not be using Intel hardware these days; it just doesn't keep up with the PPC G5 for floating point.

Re:Pretty poor LinPack performance...

[jstott]

Taking the opportunity for a moment to troll, flame bait and be an annoying Apple user, I think it's worth commenting how piss-poor the P4's LinPack performance is.

The AltiVec processor on the G5's is a vector coprocessor. If your compiler/library is set up to use it, that's good for a 4-5x increase in floating-point speed. Essentially the CPU does a block of mathematical operations in parallel--Cray mainframes work the same way, only more so. This is different from pipelining in that it's a true parallel operation. I think the AltiVec can do vector integer operations as well, but that won't change the LinPack performance.

Note too that the boost from a vector processor only works on specific types of floating point operations, most notably matrix math, so it's not a magic cure-all. Also, the data has to be in the right format and loaded into appropriate registers, so it helps to have code written specifically to use vector operations (although a good optimizing compiler can still do a lot of the work for you)

.

-JS

Scale matters!

[IceFox]

After reading it I get the following:

1) If you are doing anything in Lightwave by all means don't use AMD's XP :) There must be some major tweak they are missing.

2) Encoding type work XP seems to be the best bang for the buck (right now)

3) I had a difficult time understanding the results because most of the graphs didn't have a scale to go by. Some of them like the games you could figure out that 500fps is twice as fast as the slowest at 250fps, but in either case you didn't care. With lame from the looks of it the slowest was still faster then what I could rip from cd (need to test, but just off the top of my head). Maybe on the larger scale for a particular test all of the cpu's are very close together, but in the view of close up it looks like one is _way_ faster.

4) With all of the tests there wasn't one compiler test :(

-Benjamin Meyer

LaGrande?

[slux]

Has everyone already completely forgotten about LaGrande?

The tech sites certainly don't seem to be making much fuss about the fact that Prescott has this technology already in it. I wonder how they can be that unknowing of it. There was this big Extremetech article on LaGrande though.

Even on Slashdot no-one seems to be bringing it up these days. For me, the benchmarks aren't even worth looking at with the knowledge that these processors are the beginning of the DRM revolution. Seems they're being able to sneak the technology inside every PC just as they've planned it.

Still, sticking with AMD is going to be just a temporary measure. Is there any talk about integrating DRM into the PowerPC? If not, maybe the next motherboard upgrade could be a Pegasos or one could just go with a Mac.

AMD64 testing

[ValourX]

Yet another review that doesn't test in 64-bit mode.

I don't know why this wasn't deemed Slashdot-worthy, but here's an excellent review of a P4 3.2E versus an Athlon 64 3200+ in both 32-bit *AND* 64-bit mode:

AMD64 vs. i386 in FreeBSD

-Jem

compiler comparison

[pwagland]

Hmm. Just had a quick browse of the article, and noticed something a little funny. In the Sphinx speech recognition test they compared all of the chips with both the microsoft and the Intel compiler. What was strange about it though was that for every AMD chip the Intel compiler was faster, by up to 4%. However, for 7 out of the 10 intel processors the microsoft compiler produced faster code than the intel compiler!

Bizzare eh?

Short version: Don't bother!

[Junks Jerzey]

Note: I'm not trolling, nor am I an AMD zealot.

Yes, you can't go by raw clockspeed alone, but in this case its close enough. In short, 3.4GHz P4 is THIRTEEN PERCENT faster in raw clockspeed than the 3.0GHz P4. The actual performance increase is less than that. At the same time, BOTH PRICE AND POWER DISSIPATION have gone up by MUCH MORE THAN THIRTEEN PERCENT.

Bottom line: This is a completely uninteresting processor at the current time.

Unbiased journalism

[pwagland]

Now, sure, we don't expect these people to be totally unbiased, but where did they pull this from?:

The Pentium 4 'E' is an absolute monster in workstation graphics, capturing the top spot in three of the six tests and tying for it in one more. In the other two, the Prescott 3.4GHz is second only to the Athlon 64 FX-53.

By the way, that test that it tied? It tied it with the Athlon 64 FX-53. But then I guess they wouldn't get their advertising budget if they said:

The Pentium 4 'E' and Athlon 64 FX-53 are roughly equal in workstation graphics, with the P4E winning three of the six tests the A64 FX-53 winning two, and they tied one test. Overall though there was less than 2% difference in any test.