P4 Only One Third the speed of PIII and TBird!

[OneEng]

According to Tom's hardware:
http://www6.tomshardware.com/column/00q3/000903/bb-04.html

I had already concluded that the benchmarks Intel had at IDF were more affected by its dual channel rambus and 400Mhz FSB than the processor core, but had not studied the ramifications of the results!

The comparitive benchmarks Tom does demonstrate that an equivelently clocked PIII would simply distroy P4 in non-bandwidth constrained benchmarks. DDR AMD based systems would most certianly wipe up the P4 even when bandwidth was the limiting factor.

This does not look good for Intel......Time for me to get some more AMD stock!

 

[billabong]

I think everyone should wait for the final product before people judge it. The K7 was out performed by the K6-2 in early benchmarks. I personally think the P4 will match the T-bird when it is released and that intel should be able to quickly ramp up the Mhz of the chip due to its deep pipelining. In the benchmarks I saw in real world tests the P4 easily won (Quake 3), but in synthetic benchmarks the current CPU's won (Sandra etc). I personally think real world tests show true performance, synthetic ones like sandra probably need to be updated to run correctly on P4's etc to show true performance.

 

[Sunner]

The Q3 scores posted were no doubt foobar.
No GFX card on the market would score as high as reported on 2cpu.com.

 

[DDad]

Well, if I remember correctly, the original Pentium 60 was beat rather handily by the DX100 and AMD's 486 120. And it had the embarassing incident with the Floating point miscalculations

 

[xtreme2k]

dont take these early benchmark too seriously

most of these software are written optimised for the P6-core (PPro), well optimised for the P6 architecture

the P4 is a TOTALLY new architure and it cannot take any advantage of those P6 optimistaion there for it is very normal to run slower than a equivilent P6

once compliers are optimised for the P4 architure, u will see it fly. and intel will have NO problem getting those compliers out to the developers (as they with the P3/sse instructions)

if u think back
there are actually benchmarks of a P/233MMX outperforming a P2/233
eg. CPUmark16 (yes this is ages ago)
but what i am trying to say is
with optimisation
the P4 will eventually be faster, if not faster from a clock speed point of view, which will make the chip faster than the P3 in the long run

 

[jpprod]

billabong; There's a fundamental difference between pre-release P4 and Athlon benches. At Tom's they are talking about results which Intel was showing off atIDF, whereas leaked Athlon benches were done on an intentionally crippled engineering sample. Aside from tremendous memory bandwidth which really shows the potential of dual-channel DRDRAM, the P4 is yet to impress me.

 

[Rand]

The P4 Quake 3 benchmarks are obviously fake. No graphics card on the face of the earth could hit those scores with ANY processor. The current graphics cards simply are not capabale of pulling numbers like that.

The last few days I've started to become more and more impressed with the P4's potential performance. Previously I had a lot of cdoubts but now I'm beginning to think it may just hold it's own clock for clock against the P3/Athlon.

In memory constrained benches it should have no problems running away from current SDR SDRAM platforms of course.

 

[xtreme2k]

i wont say the P4 Q3 benchmarks are actually fake

P3/1133 does 170FPS already

P4 doing 190FPS isnt totally impossible

http://www.anandtech.com/showdoc.html?i=1290&p=6

i remmeber it is 190FPS it showed, am i correct in the register article?

 

[KarsinTheHutt]

Lets wait for Anand's report before we jump to any conclusions. The people who laughed at the early K7 reports probably felt stupid after the Athlon shipped.

 

[DataFly]

How do we know for sure that Intel has not intentionally crippled the P4s they have been sending out or that the benchmarks that they have released are low?

 

[Soccerman]

yeah at 640X480.. possibly 800X600, those speeds ARE possible if you just increase the speed of the CPU, but remember, this CPU isn't faster, so there's something fishy going on, and it aint the bandwidth producing those scores.. (otherwise AMD would wipe the floor with the P3).

 

[xtreme2k]

soccerman

remmeber Q3 is not optimised for P4 at all?
let alone the

mobo
chipset
drivers
and everything

so gettign 190FPS isnt bad at all i must say

 

[Soccerman]

xtreme2K, you forget the scores this P4 gets in everything else synthetic.

 

[MadRat]

Lets not forget that SiMD cuts down on what needs to be pushed across the data bus, too. The P4 runs alot more SiMD (SSE2) than either the P!!! or K7.

 

[MadRat]

I am starting to think the P4 will rock as a server, possibly rock in P4-optimized games, but flunk my personal tests of running old software. The P4 is going to be seriously gunned with its QDR-RAMBUS bandwidth, its a great design for making a good solid advertisement for RAMSUS in servers.

1. IT pros always pay a premium for performance.
2. RAMBUS is getting closer to the magical price range of 133mHz ECC-SDRAM, but is offering a performance edge.
3. Raw GHz sells.

I think Intel has a real winner in the low-end server market. Personally, though, the P4 is not offering me anything to want. Then again, I don't think QDR-RAMBUS will have an advantage for long "when" DDR ECC-SDRAM hits the market.

 

[Remnant2]

Just a quick note about what people have been talking about "optimizing" for a processor, or about having more SIMD to make things faster. This frankly, doesn't work in real life.

First of all, do you know that the most popular C++ compiler on the planet, MSVC 6.0, optimizes for Pentium-Pro as the highest end target? Granted, that P6 core is used through the P2 and P3, but the cache structure and some opcode behavior is very different.

No compiler in real usage yet even knows what an Athlon is, much less is able to optimize for it efficiently. And the P4 will no doubt suffer a similar fate, EXCEPT that Intel puts billions of dollars into writing their own compilers that will be able to take advantage of it.

The only problem is that Intel's compilers aren't as popular as microsoft's.

So point #1) is that the optimization targets of the compilers are so far behind today's CPUs that it almost doesn't matter.

Point #2) is that you need to write assembly code to take advantage of SIMD, so its almost never used except in low-level code (ie vid card drivers). This might change with Intel's newer compilers, but in general it is still the case.



Point #3) I don't believe for a minute that the P4 performs so low as that. I'm an AMD-fan for a long time now (haven't owned a Pentium since my P54C), but Intel isn't dumb, and they have top-notch engineers. I expect the P4 will rock, but it will be severely price and production constrainded, and with AMD ramping up the Mustang in speed, I expect that it will keep Intel competitive with AMD, but no more.

 

[OneEng]

The chip is due out in October. That is only a month away. These benchmarks that Tom has compiled clearly show what a PIII would perform like if it had the same FSB and memory subsystem as the P4. I guess what I am saying is that these benchmarks do not seem to be pre-release as were the Firing Squad numbers of the early K7 core that was performed on alpha silicon. This is a benchmark publicly shown by Intel to outline the superior performance of the P4 over their previous product.

I can't subscribe to the "it is too early" defense in this case. I am sort of supprised that Intel published these results, thus allowing an early comparison of the core.

As for the "software only needs to be optimised" arguement, I don't think I would be willing to buy all new software so my $2000.00+ new computer would run as fast as a sub $1000.00 one.

This benchmark really dissappoints me. This will seriously slow down the technology race since AMD will not be inclined to release faster processors if Intel can not keep up.

Does anyone know the exact release date?

 

[xtreme2k]

OneEng

i understand about not paying for a 2k computer when a 1k comp is just as fast

but CPu is like that

you will not see how a new core benefits until later its life

with software/hardware optimisation and tuning it will eventually show its strength

intel designed the P4 is not from a High IPC point of view (Instruction per clock) but a core than can be scaled into the few GHz region (i seriously doubt the AMD K7 core can be scaled into 2.5-3G+ region, but i think intel is aiming for that). With a few GHz, with more tweaks, i am SURE it will be a pretty damn fast chip

again, i must stress u DO NOT see the benefits of a new core until later on

 

[ElFenix]

you may be able to stick a p3 on a 400mhz fsb, but theres no way it will hit the mhz that a p4 will do, so that comparison is really a lot of crap. yeah, intel will move the p3 to 200mhz bus, and .13u. .35 was good for 300mhz, .25 for 600 mhz, .18 for 1ghz, maybe, if intel gets its act together, 1.1ghz. so .13u will be good to 1.8 ghz if the scaling trend continues, but thats a 9x multiplier. on a p4, that will be 4.5x, and the p4 should hit 1.5x-2x the mhz on a similar process. great, the p4 is slower per mhz than the p3, but then again every x86 processor is slower per mhz than a k5, and you don't see amd making those anymore, do you?

edit: intel has worked really closely with compiler devs to get ia64 intstructions put into compilers, i would think they would do similarly with the p4.

 

[xtreme2k]

guys taking PER MHz comparison too seriously are too 'short sighted'

 

[OneEng]

OK, Lets just talk Speed for Speed then.

Intel can NOT get any kind of yield at 1.5Ghz on .18um.

Since they can not get .13um copper until H2 next hear, I think their position in a MHz war is quite weak.

AMD can easily shift to .15um and achieve good yields at 1.5Ghz. I have heard that new processes allow the use of AMD's existing FAB technology to do .13um. At this process, I would expect the K7 core to reach 2Ghz+. Now add to that DDR and MUCH better IPC, and it doesn't look too good for Intel for the next year. Surely most of you would agree with at least that?

 

[Warrenton]

Again, the PPro performed like a dog compared to the Pentium at first. But once things were shaken down it took off.

Lets take a wait and see attitude. As it stands now, I intend to get a t-bird or whatever this next time around. P4 is going to be too expensive. But I would have gladly taken a P3 over athlon classic. Untill KT133 athlong was not competing with the P3 in my book.

Now just if I wasn't ripping my entire CD collection again, I could type straight, but the system is overloaded.

 

[SJ]

OneEng, partially correct, they can't get PIIIs at 1.5Ghz. P4s are totally different, they plan to release at 1.4Ghz and quickly move to 2GHz. Intels .13um copper FABs go online Q1 '01 so by mid to late Q2 the PIVs with the die shrink will be out.

 

[KarsinTheHutt]

"Intel can NOT get any kind of yield at 1.5Ghz on .18um."
Hmm, OneEng, so you work for intel, right? How else did you know that? Uh oh chipzilla will send goons to get you! :Q

http://www.theregister.co.uk/content/archive/11842.html for all the dirt on P4 production

 

[ElFenix]

the pro was a damn fine piece of machinery (see sig). it was slower per mhz at DOS, which is why it was postitioned as an NT part. of course, when it was released the fastest pentium was a 133 or 150 compared to the 200 of the pro. the best thing is that a pro has enough fpu power to have garuanteed longevity way back in da' day, i played q3 with a v3 on mine up until last march. it wasn't exactly the highest res or the prettiest, but i could frag like a mofo.