Silly request for netbook users

[Hulk]

I've been benchmarking my systems with CPUmark99 since my old Pentium 90 and was wondering if perhaps one or two of you wouldn't mind running this bench on your netbooks? I know this is an ancient benchmark, is only integer, and only uses one core among other things. But I am curious as to how these netbook CPU's (the atom in particular) compare to some of the desktop CPU's from a few years ago.

You can download it here and you don't even have to install anything, it just runs from the exe file. http://www.hyperactivemusic.com/cpumk99.zip

Here are the results from my systems (and a few from here) over the years.

Pentium 90 - 5.5, 16.4Mhz per CPUmark99
Celeron 450A (o/c) - 36.1, 12.5MHz per CPUmark99
PIII 850 - 76.1, 11.2MHz per CPUmark99
P4 2.4 - 157, 15.3MHz per CPUmark99
P4 3.06 - 194, 15.8MHz per CPUmark99
AMD X2 at 2.6GHz - 294, 8.8MHz per CPUmark99
E6400 o/c 3.2GHz - 445, 7.2MHz per CPUmark99
QX9650 at 3GHz - 432, 6.9MHz per CPUmark99
i7 920 at 2.66GHz - 415, 6.4MHz per CPUmark99

Of course I could run two instances of CPUmark99 with the E6400 and basically double the score but I'm really just i nterested in how a single core of the i7 performs. As you can see except for the P4 each successive CPU become more efficient at running this benchmark.

The strange thing is despite the age of this benchmark it still provides pretty accurate results. For example when comparing my overclocked E6400 to my old P4 I should get a little over 4 times the performance since the efficiency per clock is doubled and there are double the number of cores. And in Sony Vegas (video editing) running a script that is multithreaded I get a little better than four times the performance.

Thanks!

 

[n7]

Your question got me curious, so i found some results via google.

http://www.oc.com.tw/article/0810/readarticle.asp?id=6541
http://www.xtremesystems.org/forums/showthread.php?t=228660
http://overclockzone.com/tor_za/year_2009/06/atom_330_battle/index2.htm

From those three results, i see scores of 109-112 (different Atom versions).

Those scores are precisely why i'll never get a netbook, well, that & the fact that i work on netbooks pretty much every day, & hate them.

ULV mobile CPUs are far better performance, & you can find them in notebooks for only a bit more $$$.

 

[soccerballtux]

My dual threaded Atom feels far more responsive than my old Sempron 3100+ ever felt. I love these things.

Mine scored 134.3 (combined) running two instances. Because the Atom is In-order-execution only, I feel it is only fair to run 2 threads of this since the atom is dual threaded and single core.

Running once instance it gets 92.5.
So, 45% performance improvement with the dual threading. Hyperthreading doesn't help Out-of-order execution processors nearly as much.

I should note that I have the z520 overclocked at 1.4Ghz.

 

[IntelUser2000]

I've been benchmarking my systems with CPUmark99 since my old Pentium 90 and was wondering if perhaps one or two of you wouldn't mind running this bench on your netbooks?

Not a Netbook, but it uses an Atom CPU, so here it is....

Viliv S5(http://www.umpcportal.com/products/Viliv/S5/)
Z520 Atom 1.33GHz

Single instance - 90(14.8MHz per CPUMark99)
Two instances - 130.1(10.25MHz per CPUMark99)

Portable/Laptop power management/Windows XP Home Edition SP3

It's amazing how accurate the benchmark is even after all these years...

So, 45% performance improvement with the dual threading. Hyperthreading doesn't help Out-of-order execution processors nearly as much.

On i7, each instances of CPUMark99 with 8 instances running gets 440. With 4 instances its 650. Meaning Hyperthreading improves it by 35%.

 

[v8envy]

Was the benchmark compiled with optimizations for newer processors? I'm guessing: no.

I've found the i7 is far less sensitive to page alignment than a p4 or core2 when running integer heavy code. So running code tuned for a 386 on an i7 and core2 would show the i7 in a better light than running code tuned for core2 or p4.

It'd be fun to see the results from compiling the benchmark with emitted code tuned specifically for the processor being tested. My initial gut feel is the i7, PII and earlier wouldn't gain much, p3 and core2 would improve modestly, p4 quite a bit.

 

[soccerballtux]

Not a Netbook, but it uses an Atom CPU, so here it is....

Viliv S5(http://www.umpcportal.com/products/Viliv/S5/)
Z520 Atom 1.33GHz

Single instance - 90(14.8MHz per CPUMark99)
Two instances - 130.1(10.25MHz per CPUMark99)

Portable/Laptop power management/Windows XP Home Edition SP3

It's amazing how accurate the benchmark is even after all these years...



On i7, each instances of CPUMark99 with 8 instances running gets 440. With 4 instances its 650. Meaning Hyperthreading improves it by 35%.

I'd like to see benchmarks of that. We NEVER see that improvement in hyperthreading with the Core i7 in many applications where it matters (gaming). At MOST 20% usually 10-15% or so.
Encoding benchmarks are nice but at the end of the day we both have 4 cores and we both can encode that DVD in 20 minutes.

 

[Ben90]

I've found the i7 is far less sensitive to page alignment than a p4 or core2 when running integer heavy code. So running code tuned for a 386 on an i7 and core2 would show the i7 in a better light than running code tuned for core2 or p4.

It'd be fun to see the results from compiling the benchmark with emitted code tuned specifically for the processor being tested. My initial gut feel is the i7, PII and earlier wouldn't gain much, p3 and core2 would improve modestly, p4 quite a bit.

The phenom processors can get pretty beastly actually when the app isn't compiled on an intel compiler.

 

[IntelUser2000]

I'd like to see benchmarks of that. We NEVER see that improvement in hyperthreading with the Core i7 in many applications where it matters (gaming). At MOST 20% usually 10-15% or so.
Encoding benchmarks are nice but at the end of the day we both have 4 cores and we both can encode that DVD in 20 minutes.

4 instances each at 650: http://g.imagehost.org/view/0725/4gpcm99x4-2
8 instances each at 440: http://g.imagehost.org/view/0667/4gpcm99x8-2

The only reason Hyperthreading doesn't help in games because they don't go above 8 threads. A point to be made about why the dual core Clarkdale with HT will be awesome.

And Core i7 does have a much better SMT implementation than the Pentium 4: http://www.solidmuse.com/2008/12/core-i7-to-hyperthread-or-not.html

Based on the results I see above it would appear that hyper-threading does not hinder core SolidWorks performance. For those areas of SolidWorks that take advantage of all the threads available for processing, hyper-threading looks to offer a very nice performance advantage. It looks like Intel has done a very nice job in this new implementation of hyper-threading in the Core i7's.

I plan to keep hyper-threading enabled on my system.