Northwood benchmarks

[Raystonn]

I have found some Northwood benchmarks run by a third party to which I can link. Since these are links to websites on the public internet and the content was not created by me, I can safely present them here.

What this person has is an engineering sample of a Northwood (0.13 micron) Pentium 4 1.8GHz running with a 141MHz external clock at 2.2GHz. He is running it on an i845 chipset motherboard (an Asus P4B Rev1.03 Bios:1004) because this was apparently the only Socket 478 motherboard he could find in his region that had a BIOS that would support this processor at this time. Other system components include the following: Tonicom PC-166 CL2 256MB TBGA SDRAM Module, Elsa Gladiac 920 GeForce3 64MB DDRRAM, Seagate Barracuda ATAIII 40GB UltraATA/100 7200RPM, Delta 350W PSU, Taisol CEP405092 w/Delta 7*7*1 3000RPM Fan.

Historically, the Pentium 4 (Willamette) coupled with the SDRAM i845 chipset has performed miserably, especially when compared to the Athlon MP/XP 1.4GHz. When coupled with the i850 chipset it has performed about 30% better, making the performance between the Athlon and Pentium 4 processors a bit more even, but still falling below the Athlon MP/XP 1.4GHz in many benchmarks.

I now present these Northwood benchmarks:

<A HREF="http://netcity2.web.hinet.net/UserData/adoli/P4N-2258CPU.JPG" target="_new">CPU Arithmetic Benchmark - SiSoftware Sandra</A>
<A HREF="http://netcity2.web.hinet.net/UserData/adoli/P4N-2258MM.JPG" target="_new">CPU Multi-Media Benchmark - SiSoftware Sandra</A>
<A HREF="http://netcity2.web.hinet.net/UserData/adoli/P4N-2258MEM.JPG" target="_new">Memory Bandwidth Benchmark - SiSoftware Sandra</A>

(These benchmark photos were taken from <A HREF="http://netcity2.web.hinet.net/UserData/adoli/test-5.htm" target="_new">here</A>. Note that <A HREF="http://babelfish.altavista.com" target="_new">http://babelfish.altavista.com</A> has better luck translating it using the Japanese to English filter rather than the Chinese to English filter for some reason.)

Mature comments are welcome.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

Hmm...memory bandwidth seems rather low, considering that the FSB is 141. Or is the memory bus different? I'm confused

Ok, hold on...141x4=564/3=188
If the memory is running at that (I think that's how you described it), then that seems low. But that's not the CPU, so I'll leave it alone.


As for the CPU...it doesn't seem very high. Less than 10% higher than the Willy 2.0? Am I missing something? (I haven't read the article yet, just wanted comments).

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

 

[Raystonn]

If you want more memory bandwidth I suggest going with the i850 platform with DRDRAM. This should boost benchmark scores by about 30% across the board as well judging by current Willamette w/i845 benchmark scores.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

(I edited my post after you replied, apparently)

Agreed, i850 is better for bandwidth. It's just that the numbers don't seem to match up well (compared to other SDRAM chipsets). Oh, and the benchmark numbers would be skewed, since the Sandra examples are for i850. Taking that into consideration, it's rather impressive. I'll have to have a look for some Willamette i845 scores.

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

 

[Raystonn]

Yes, do not forget that the Northwood is beating the Willamette by about 15-16%while very much handicapped by the i845 chipset with SDRAM. Add another 30% or so from a move to i850 and you get the approximate performance figures of a stock Northwood Pentium 4 2.2GHz. Toss in the fact that they all overclock extremely well (even with air cooling) and you get a processor that could easily leave everything else in the dust.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

Ah, you reminded me of a question I had. Are the Willamette socket 478's .13 micron, or .18 micron?

I couldn't find any reviews of i845+Willamette that showed the same CPU benches

BTW, thanks for the post, Raystonn.


<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

 

[Raystonn]

All Willamette Pentium 4 processors use 0.18 micron technology.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[Raystonn]

I found a page with a <A HREF="http://www.accelenation.com/?doc=70&page=6" target="_new">benchmark</A> comparing Sandra scores for the Pentium 4 coupled with various motherboards/chipsets. It looks like the i850 increases performance in these benchmarks by about 150-160%, not the 30% I had previously estimated. I will attempt to find more benchmarks to verify this.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

I found that, but it didn't have the Sandra CPU benchmarks (like the first review). That's what I'm interested in, to see exactly how much the i845 hampers the CPU itself.

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

 

[Kelledin]

D'oh...so much for a fixed FPU :frown: . This sample of the Northwood apparently still needs SSE2 to keep up.

The images appear to be dated Oct. 15, btw. The Northwood sample could be much older...

Kelledin

"/join #hackerz. See the Web. DoS interesting people."

 

[Raystonn]

What caused you to come to this conclusion? Remember that A) the CPU is castrated by the i845 chipset and B) this was not the final Northwood processor. Revisions have been made.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[zengeos]

True, it's using SSE2 to derive much of the FPU performance, but still, it shows a nice performance boost over it's siblings. It should be interesting to see how Thoroughbred and Northwood compare...rumor has it that Northwood won't be released until January now, which is cutting it close to rumored Thoroughbred release (within 1-2 months)

Just how much of the performance boost is from the die shrink and how much is due to increased on chip cache?

Mark-

When all else fails, throw your computer out the window!!!

 

[Raystonn]

Die shrinks themselves never increase performance. They simply make everything smaller and cooler and give you more room to add more features to the die.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[FatBurger]

Yeup, Raystonn is right.

Now, what's interesting is this:
Athlon was released to compete with the P3, right? It did so, very well (considering AMD's small history at that point). Now, Tualatin and Thunderbird compete pretty well. Palomino came out well after Willamette, but those two would line up. Yet Thunderbird holds its own quite well against Willamette. So that would place Northwood and Thoroughbred together, not Northwood and Palomino.
Did I miss anything?
Obviously calendar time is (to a certain extent) more important than CPU core generations, but I think it's remarkable that AMD has kept ahead of Intel generation-for-generation.

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

 

[Makaveli]

Those memory bandwidth numbers look correct for an P4 on a SDRAM platform! As for Northwood I speculate the increase in performance won't be more than 10% with RDRAM 800
not sure when @ 133 Fsb or 533 whichever you prefer.
Hmm so I think a Athlon PR2000+ will be about even with a Northwood @ 2.2 Ghz. All speculation on my part we will have to wait and see what the future holds.

Now if northwood does comeout on the 533Fsb with RDRam 1066 things would be even closer. But an XP on a 166fsb with DDR2700 would be able to keep up if not surpass it.

Once again all speculation on my part. Hmm wasn't expecting the Delay from November to now Q102. That's really close to the next athlon cpu. So if Intel does require the performance lead it won't be for long!

AMD still owns intel clock for clock!

<A HREF="http:////www.anandtech.com/mysystemrig.html?rigid=9802" target="_new">http:////www.anandtech.com/mysystemrig.html?rigid=9802</A>

 

[Yahiko81]

Trust Ray instead of speculation. When he says 30% or more he literally means it.

Nice <b><font color=green>Lizards</b></font color=green> <b>crunch</b> Trolls cookies....... Yummy!!

 

[zengeos]

Thank you for clarifying the die shrink question Ray. I had read on several tech sites that die shrinks tend to equate to improved performance due the the significantly shorter traces and the like..upwards of 5-10%.

That's why I was asking what can be attributed to what in the Northwood performance improvement.

Mark-

When all else fails, throw your computer out the window!!!

 

[MadCat]

I view the Thunderbirds as PII ISA compatiable with 3DNow! instructions (but SSE compatibility was more desirable for me). I view the Palomino as PIII ISA compatible with 3DNow! instructions (a bonus to run my old games that may have 3DNow! optimizations such as Quake II and run my current crops of games that are SSE enabled). I have been a fan of the PIII series (PIII 550e, PIII 700e) but today no viable upgrade path exists on the Intel side as far as I'm concerned (I don't like the current P4). With Intel processors, I overclocked on a sure bet to get a better price performance ratio. With AMD processors, I probably won't feel the need to overclock with the highest speed Athlon XPs I can obtain for the price around $220 that I was paying for the Celeron (c300a), and PIII series (PIII 550e, PIII 700e) at the time.

 

[Kelledin]

A) Sanda Whetstone/Dhrystone is probably not so bandwidth-dependent. Any sensible benchmark seeking to stress only the FPU/ALU processing power would feed the respective units a dataset small enough to fit in the on-die cache and (though this is more CPU-specific) conforms to best-case scenario for the CPU's caching algorithm.

B) The sample may indeed be old; I actually did point that out. I noted the mod date of the image 'cos that's all we have to go on for the sample's age.

Kelledin

"/join #hackerz. See the Web. DoS interesting people."

 

[Makaveli]

Since Raystonn works for intel, to me his opinon is always in slight favor for intel. Not saying that it always is don't get me wrong. But it's like being a cop and being personally involved with a Case! So until Northwood is released, I'll stick to my opinon! I might be right I might be wrong we'll just have to wait and see.

 

[Guest]

Hmmm... I hope the FPU on the released version of Northwood performs a lot better then that. My Sandra score is only 80 points lower and Im running my 1.8 @ 2.1 GHz. Im using RDRAM, but that shouldnt affect the CPU benchmarks. The interger scores are about 400 points higher then my system, no doubt do to Northwood's larger L2 cache. Too bad I cant get my CPU to run reliably at 2.2 GHz for a closer comparison.
Judging from that sample it appears that Northwood is nothing more then a .13 micron Willamette with 512KB of L2 cache.

 

[somerandomguy]

I have some questions.
How dependent are those theoretical <b>CPU</b> benchmarks (not the memory one) on the platform used (i.e. i845 or i850)? The Sandra test compares your results with other reference CPU's rather than CPU's and their platforms, so wouldn't this indicate that the test is largely independent of the platform?
If this is the case, those results aren't very impressive. All you have is a performance increase of approximately 13% with the 2.26Ghz Northwood, over the 2.0Ghz Willamette when it's clock speed is 13% faster.

"Ignorance is bliss, but I tend to get screwed over."

 

[Raystonn]

If you would like to test this, then please do a few tests on your machine using Sandra. I am going to assume you have an unlocked Athlon system. Correct me if I am wrong. What I would like you to do is to run the ALU and FPU tests with a 133MHz FSB and then with a 66MHz FSB. On the second test please raise your multiplier so you stay at the same CPU clockspeed as you used for the first test. We can compare the scores to see what effect a doubling of memory bandwidth has on the benchmark scores. Note that this might give a false reading of 'no significant effect' due to the Athlon not relying as much on bandwidth as the Pentium 4. However, it can safely prove that there is a significant effect if one shows up on these tests.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

 

[somerandomguy]

I would happily do this test, and would already be doing it, but unfortunately I can't adjust the multiplier as I have a Pentium 3.

"Ignorance is bliss, but I tend to get screwed over."

 

[Raystonn]

Alright then. Perhaps someone else with an unlocked Athlon will do this for us. The only unlocked processor I have is one from which I cannot post benchmark results.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =