Generic CPU Bench

[RadeonX2]

ei guys I have suggestion why not separate Intel processors from AMD processors column so that AMD will have its own top 1 in benchmarking

 

[btarunr]

From the results so far, it can be deduced that:

• The results get very inconsistent sometimes
• The test is multi-threaded, quad-core chips have a genuine advantage over dual-core ones
• The test is cache-intensive
• Having a broad system bus helps (effective FSB for Intel, HyperTransport link speed for AMD)
• The test flows large amounts of data through the available cores/threads

 

[ascstinger]

2.0Ghz (200x10) AMD Sempron (aka neutered athlon xp)

 

[Boneface]

Well heres mine

 

[Zehnsucht]

How about a scoring system, throughput/time = LOL.OMG generic cpu bench points?

 

[Binge]

lol wow a xenon and q6600 are the reigning champs

 

[Zehnsucht]

lol wow a xenon and q6600 are the reigning champs

X3350 = Q9450

 

[btarunr]

Let's not have an AMD-Intel demarcation, but how about a dual-quad demarcation? A lot of these C2D scores are mighty impressive.

 

[DOM]

 

[Fitseries3]

e7200 at 3.18ghz...

gonna try the e8600 at 5ghz now.

 

[Fitseries3]

q6600 at 3.9ghz

 

[NastyHabits]

Still won't run on my P-35, but here's my score on my lowly (and ancient) Prescott P4 @ 3.0 GHz.

 

[Fitseries3]

e8600 @ 4.6ghz fastest time yet i believe

 

[btarunr]

omfg!

I'll split the dual and quad core lists.

 

[Fitseries3]

omfg!

more coming....

 

[DOM]

e8600 @ 4.6ghz fastest time yet i believe

damn you

i need a 8600 or 8500 but gots no money

still got the Highest Throughput

 

[wolf2009]

Q9450 @ stock

 

[roofsniper]

cpu clock isn't everything for this program. the memory plays a big role as well. so you could have better ram but have a lower cpu clock and still beat a a cpu with a better clock but slower ram.

 

[Solaris17]

well i took a look at the program for you guys i havent dove into it a whole lot...but i may beable to explain the diffirences in scores.


the program reserves 64mb of data for each core present on the system. It then forces all this data at one time through the cache of the CPU. however through the 64mb that it allocates it allocates this data in 3 diffirent types

16bit data
32bit data
64bit data

and allocates 16mb for each bit type. the diffireces in scores will probably var because it is mapping 16mb of 16bit info in a matrix fashion and so on and so forth since in programs like sciencemark it takes longer time to process higher matrix levels of data the same assumption can be made here. the scores may differ by alot or a little because since were dealing with small its of data in the cache any diffirence can affect the scores greately at that speed. For example opening the start manu and letting it run as apposed to having it closed are things that could affect its score much like it chaning from 3:59 to 4:00 could.

 

[Particle]

Hey guys. I was surprised to see this thing spreading around here this morning. I wrote it yesterday for a few of us over at XS to play around with. I see there are a lot of questions about how this thing ticks. I'd be happy to answer some questions if you'd like. I'm just thrilled to see people having fun with it.

 

[btarunr]

Tell us more about it. Keep up the good work!

 

[DOM]

btarunr my last run was at 3.8GHz

well vista 64bit its slower @ 3.6GHz
8.550ms
30,660 KCU/s

 

[Particle]

Over on another forum, we were having a discussion about SuperPI and if it was a relevant indication of CPU speed (stemming from the AMD vs Intel performance it demonstrates), and I decided to make a humorously themed benchmark ("Generic CPU Benchmark") to throw into the discussion for us to play with. At first, I was going to implement a Pi-solving algorithm, but after ten minutes or so into the project I thought--"Don't we have enough of these already?" So, I decided to go a different direction. Instead of finding Pi, I decided to test the sorts of commonly used mathematical operations that make up real programs. This includes a battery of operations such as multiplication, division, modulo, square roots, etc. Since real programs pull data from somewhere, manipulate it, and put it back, I also decided to it would be more realistic to manipulate a large data set than just play exclusively in the CPU's cache. As such, this program benefits from both raw CPU speed and also memory speed--just like real apps do. Despite the humorous interface, the core of the app is a serious crunching routine.

The indicated throughput (KCU/s) is a real reflection on the amount of work done by your system. It may vary between runs a bit, but that is reflective of what's actually going on on your computer at the time. Time elapsed doesn't necessarily mean everything. For example, a single-core processor with hyper-threading for instance will end up running two worker threads. While each would take longer to complete than a single one would on the same CPU, it is crunching a dataset that is twice as large. As such, throughput will likely be a little higher despite the longer completion time.

 

[Tatty_Two]

Looks good, I'll give this one a try sometime!

 

[btarunr]

@ Particle: How would you explain the somewhat significant spread in scores between runs? Do build on this idea. Teh interwebs needs multi-threaded benches.