Athlon XP < P4 < A64

[VIAN]

I emailed Anand, maybe he remembers.

As I was writing, I typed in my name as Vian. And I realized how girly it looks like that - good thing I typed it as VIAN in the forums.

 

[Vee]

There is one thing I'd like to add to this discussion. That is that the P4 is highly dependent upon software to be optimized for it. More specifically, loops have to be autovectorized for SSE2. If this is not the case, i.e. application is old, compiled for PIII, or librares are old, or application is not suitable for packed ints and fp, no P4 is competitive with either K7 or K8. I would even hazard a guess that the 'lowly' XP2500+ beats even the 3.4EE.

Primarily, media and benchmarks can be, and are highly SSE2 optimized.

If you run other, more general types of applications, or older applications, typical benchmarks does not reflect the performance you will "enjoy" from the P4.

There's also a reverse side of this. Packed optimization for 3DNow+ have been rather late coming to many media apps. If you happen to run a such media app (lacking 3DNow+, but optimized for SSE2), performance is going to be dramatically better on a P4. This means you are generally much better off with a P4, if your major activity is media. A64 supports SSE2, but again, older apps seem sometimes to fail identifying that.

Personally, I'm not much concerned or interested in these issues any more. 64-bit software is so much more important to me. But I am a latecomer to AMD. And the primary reason I discovered and explored AMD, was the great disappointment in performance from the P4 compared to PIII. AMD suits me better since it's more like a PIII on steroids.

In closing, I think the question "XP<P4<A64?" need to be put in relation to a specific class of applications.
P4 is *only* ">" to XP in regards of supporting SSE2. In all other respects, 386/387, MMX, SSE, scalar math and int, conditionals, AthlonXP is superior to the P4.

P.S. I also note that on these forums, there are frequent comments concerning the great performance of early Athlons and poor performance of early P4s.
Well... Even though P4B and P4C picked up a good general boost from their faster FSB, I can't see much have really changed. - So I can't really help asking the question: How many realize that benchmarks have changed?
I kinda think that fact (and also changing apps) is the primary reason why the perception of AthlonXP vs P4 performance have also changed.

 

[justly]

LOL, Vee comes through again.

I have always believed in what you just said, although I doubt that I could explain it with such eloquence.
I am quite aware that when benchmarks change and that it tends to benefit the P4 design more than the Athlon. Because of this and the fact that I use a variety of older software, switching to the P4 has always made me a little leery. I don?t know how much the performance would be affected anymore considering the speed the P4 has reached but I still cant help thinking about it.

As for the need to specify a class of applications to be able to compare processors, I agree (I thought I mentioned this already) and thought for the sake of this thread we had agreed upon games. What brought me into this thread to begin with was the questionable reference (IMO) use to compare the two.

BTW Thanks Vee, I always enjoy your enlightened responses.

VIAN, good luck trying to find out anything. I tried when the article was current and either got no answer or was treated as if I was harassing the staff.

 

[VIAN]

If software has that much of an influence, then we need CPU APIs.

 

[carlosd]

VIAN, Those are old bechmarks, only the FX-51 and the 3400+ appears. I was talking in PRESENT, show the tests with all the NEWEST processors. I also said applications , not synthetic benchies like Sysmark.

Didn't do your homework well, kid.

 

[carlosd]

Originally posted by: Dman877
3200 XP or A64? The 3200+ XP is 112 on pricewatch and the cheapest 2.4 (800) is 115...

No this is the 2.4B which cost 115.

 

[VIAN]

VIAN, Those are old bechmarks, only the FX-51 and the 3400+ appears. I was talking in PRESENT, show the tests with all the NEWEST processors. I also said applications , not synthetic benchies like Sysmark.

Well, considering AMD has faster chips in terms of Ratings, 3800, which is probably supposed to compete with an unreleased 3.8GHz from Intel, no wonder it's faster.
A synthetic benchmark is an application. I don't like sysmark either, it's retarded now.

 

[carlosd]

Synthetic benchmarks are not applications since you can't do nothing with them, only false benchmarking.

 

[Vee]

Originally posted by: VIAN
If software has that much of an influence, then we need CPU APIs.

Well, I think we need SSE2/3. I agree with Intel on that. Code should be SSE2 as much as possible.
The P4's SSE2 is somewhat prone to stall, and A64's is a bit on the wimpy side, so I think we're going to see even more speed from future improvements.

 

[Vee]

Originally posted by: carlosd
Synthetic benchmarks are not applications since you can't do nothing with them, only false benchmarking.

(SysMark and PCMark are just plain lying.)

I sometimes forget how ugly it can get. Just today I dropped my jaw. I did a REAL working app benchmark (I'm going to fix us up with new A64 3400+s ). On one of our apps, a very hightech, very large and demanding, small customerbase, 5 figure $ licence, "sort of" *engineering* app, the Athlon64 runs the job 4.35 times faster per GHz than the P4! (From earlier, I know that even an AthlonXP turns in 3-4 times the performance per GHz.) To be competitive with the 2.4GHz A64 3400+, a P4 would have to run at 10.44GHz! :Q

That's pretty extreme. Seems almost impossible. On most general (non SSE2) apps you get a 50-85% advantage for AMD (per GHz). The P4 stays pretty cool, so it shouldn't be throttling (BTW, does anyone know a simple way to check for throttling? ). I don't know why this is so extreme. But niche science/tech/math/simulation apps seem to be AMD's real forté. I can only guess it's the combination of non optimized code, scalar FP math, together with tons of branches, that breaks the Pentium 4. The memory model is also very large.

The weird thing is that representatives for these apps, rarely recommends AMD. On the contrary, if they say anything at all, they tend to sing the 'Intel-so-stable-compatible-and-reliable-the-professional-choice' BS. We've had zero, 0.00000.... problems with our AMDs.
- Boy, we must be lucky! :roll:

In all fairness to Intel, if you fool around with video apps that haven't been optimized for AMD, it can get just as ugly in reverse. We still use P4Cs too. For things that suits them. And I like them too. I just wish Intel would come out with a 64-bit successor that have more solid allround performance than P4F seem to have.

 

[justly]

Not that I know much about code, but I have thought (for quite some time) that optimizations where the key to performance. The thing is what I considered a big difference seems like a ?margin of error? compared to what Vee just described, that is not big that is inconceivable.

 

[voodoochili]

real ingame benchi's P4 wins
http://www.xbitlabs.com/articles/cpu/display/doom3-cpu_5.html

 

[carlosd]

Oh yeah they used ONLY VIA chipsets boards, we all know that 6800 video cards reach their best with nforce 3 chipset. Show us the same benchmarks with nforce 3 and see what happens:laugh:

Those benchmarks are nothing, since they used fraps. Fraps is good but not always tells the truth. In some games FRAPS shows me 60 - 85 FPS , but they are jumpy. I would like to see the same bechmarks in other page, look for them an show us. Good luck!

 

[FelixDeCat]

3.4EE>everything!

 

[imported_tss4]

Originally posted by: VIAN

"AXP 3200 < P4 2.4C"


!!!!!!!!!!!!!!!!!!!!!!!!!


more like a 2.8C is better or on the same level as a 3200+

http://anandtech.com/showdoc.aspx?i=1834&amp;p=1

In this article, it shows that even in games, the performance was similar if not better than the 3200.

And at that time, I don't think the price from the 3200 could beat the 2.4C price by a long shot. Still today the 3200 is 20-30 bucks more expensive.

There are some where the 3200's performance can reach about a 2.6C level, but overall, the 2.4C is still a better buy, especially when you consider how far you can OC it.

This same article shows the AthlonXP 2500+ was better than P4 3.20 in a general usage tests. You have to look at all the tests. Each processor has its strengths. But, the XP's are definately not in the same league as the athlon 64's. The P4's have a tough time competeing with them too.

 

[Vee]

Originally posted by: justly
Not that I know much about code, but I have thought (for quite some time) that optimizations where the key to performance. The thing is what I considered a big difference seems like a ?margin of error? compared to what Vee just described, that is not big that is inconceivable.

- Well... I just had to doublecheck that. So today, I went through the hazzle of installing the app on another P4. Indeed the first P4 seem to be ailing in some way. Probably hot and throttling, though I still don't know what is wrong (I didn't build that machine). I fiddled a bit with the bios, and made several timed runs, but it was all the same. I suppose temperatures don't check out. Even so, the A64 is still more than 2 times faster, or something like under 3 times per GHz.

 

[carlosd]

Really? FelixDeKat Just shows us at least one complete review of P4 EE beating a FX-53, JA JA JA. Good Luck finding it.:laugh:

 

[VIAN]

If you look at CPU benchmarks on Doom3, The P4 is the winner.

 

[carlosd]

Originally posted by: VIAN
If you look at CPU benchmarks on Doom3, The P4 is the winner.

which benchmarks?
You call benchmarks the vodoochilli's ones?
I won't take those benchmarks seriously until they use the latest nforce 3 boards.
Oh you have homework too: prove that those benchmarks are true. Very hard work to do since many inconsistencies have been found.