Recalling Yonah/Sossaman Cinebench benchmarks

[IntelUser2000]

http://www.computerbase.de/news/hardwar.../august/idf_benchmarks_sossaman_yonah/

I heard people say that how Yonah sucks compared to X2 which is a previous gen processor, and how dual 2GHz Sossaman manages to barely beat single X2, even though the scales are linearly scaled according to clock speed. I have something to say about that.

First of all, Cinebench scales NEARLY LINEARLY according to clock speed. Take a look at X2 score.

4800+=638
3800+=536

638/536=19%, which is very close to 2.4GHz/2.0GHz of the processors. It has 95% scaling, which is NEARLY LINEAR.

Second, is that even though they say 2.0GHz Yonah score is scaled linearly, it is not if you calculate.

2.0GHz Yonah=521
1.5GHz Yonah=383

521/383=36%, but 2.0GHz/1.5GHz=33%

1.33x383=509, that's the actual linear score

It seems that 2.0GHz score that THE SITE SAID ITS LINEAR, is actually SUPER-LINEAR!!!


Yonah vs. Dothan advantage per clock.

1.5GHz Yonah=206
1.73GHz Dothan=228

228/206=10.7%

1.73GHz/1.5GHz=15.3%, on 95% scaling, performance advantage by higher clock is=14.6%

However, 1.73GHz Dothan is only 10.7% faster than 1.5GHz Yonah, not 14.6% faster. Per clock Yonah has---1.146/1.107=3.5% advantage per clock over Dothan

Going to dual core for ALL processors give approximately 85% advantage over single core at same clock speed. Cinebench isn't affected at ALL by memory bandwidth: http://www.pcper.com/article.php?aid=133&type=expert&pid=13

What makes Yonah faster per clock over Dothan?? I am not sure. Enhanced IDIV performance?? Better floating point?? Better cache architecture??(probably not as Cinebench is not affected by memory bandwidth) Better prefetch??

Whatever, I think this one particular benchmark doesn't show anything about Yonah's performance. They could have tested a benchmark that has Yonah own everything then people would be saying the site is Intel-biased, its cheating benchmark for Intel CPUs. If they show a benchmark that shows contrary, people would be saying, Yonah is dead, Intel is dead or whatever.

Yonah is January 6th. Looking at how long Dothan review from Anandtech took to surface, plus Sonoma benchmark never showed, I bet I won't be looking for Anandtech benchmarks, unfortunately.

 

[IntelUser2000]

It looks like Pentium M has 91.7% scaling per clock in Cinebench. That puts Yonah advantage at 2.96% over Dothan per clock.

 

[IntelUser2000]

I'll just say 3-5% in Cinebench for Yonah over Dothan per clock. I got the scaling using the benchmarks from TWO sites not one. Although looking at PD 3.2 score, it seems it SHOULDN'T matter.

One thing that seems weird is the CPUID. It shows 600MHz base clock for FSB. It says the FSB is 2400MHz, which is higher than the CPU's core clock. So what, the multiplier is 2.5?? It doesn't make sense as Yonah is supposed to have 166x4=667MHz FSB not 2.4GHz. Anyway, I would really like to see Yonah benchmarks.

 

[Pabster]

How about we just wait to see Yonah, and quit speculating and throwing out numbers.

As for Sossaman, we won't know for a LONG time.

 

[IntelUser2000]

I am not speculating, what makes you think that?? I am calculating from benchmark numbers available on the net. If you can't troll better than that, please GET OUT.

 

[Pabster]

Do you have a Yonah or Sossaman in YOUR POSSESSION?

I didn't think so.

Be careful calling someone a troll when you are using the name of a CPU manufacturer in your handle. LOL!

 

[IntelUser2000]

Oh, right my name invalidates the benchmark. I made up on quick since I didn't want to spend so much time making up one on some cool one that people won't insult about.

Now if you actually care to look at my numbers, you'll see I AM NOT WRONG. Since you are bashing me with no real sense, it seems you are the troll.

What does me having Yonah/Sossaman has to do with my calculation when I provided LINKS AND HOW I CALCULATED THEM!!?? How about you complain to Computerbase, not me eh??

 

[coomar]

what the hell is super linear?

its either exponential or linear

its still very close to a linear fit, linear just means that it can be fitted using y=mx+b

with just the 2 endpoints its pointless to do fits, but the reviewer is correct it is very close to a linear dependence

 

[Markfw]

What we are saying, it when you can buy any of them, and you have a production, or even a review sample for one that has been released, talk to us, until them quit speculating. Thats all you Intel fanboys do of late, since Intel has nothing to offer right now.

 

[Furen]

I don't even know why I'm responding, but here it goes:

This benchmarking was done at IDF so I'm sure they didn't chose the benchmark, Intel did. Now, knowing that Intel chose the benchmark, you can infer that they chose it because it shows something that they wanted the media to see, whether the better clock-for-clock performance compared to dothan, the nice dual-core scaling or something else I can't say. Anyway, I'm sure no one doubts that Yonah will probably be a good match for X2s, clock-for-clock, but I'll look forward to seeing the highest-end Yonah match the highest-clocked X2 (which will probably be 2.6GHz by then). Clock-for-clock comparisons between two completely different architectures are stupid because they are DESIGNED to have different clock-speed targets. Yonah will, most likely, follow up on Dothan's sucess and continue to thrive in low-power applications while the X2 will continue to rule the high-performance roost.

 

[IntelUser2000]

Exactly. But if you look at PC Perspective benchmark and see Smithfield 3.2 benchmark, you would know the benchmark is NOT wrong. I would like to see other benchmarks for Yonah.