Le 04/09/2011 04:26, jhu nous fit lire :
> Alain <aze### [at] qwertyorg> wrote:
>> Le 2011/09/03 16:44, jhu a écrit :
>>> Some major tech sites use povray for their CPU benchmarks. However, they don't
>>> necessarily use the latest builds. For example:
>>> http://www.anandtech.com/bench/CPU/40
>>>
>>> This site is using beta 23, and their numbers seem unusually high. For example,
>>> an Athlon II x4 630 is reported on their site to be 2227 pixels/second. However,
>>> when I run RC3 on my computer (same CPU), I only get 706 pixels/second. I just
>>> want to see how accurate their numbers are.
>>>
>>>
>> The key question is:
>> Do they use the same scene as you are? With the same options?
> 
> As far as I remember they're using the benchmark menu option.
> 
> 
Benchmark was in a somehow-light-version in the previous beta (as some
features were not yet fully implemented), so comparing between beta is
not ok.

Just my 0.02€.

And for added tax-value, they are not specifying the compiler.
I got different result between intel and gnu for my 980X:

icpc around 2016 PPS, g++ around 1691 PPS
(these result are bullshit: with 12 threads, icpc take 2:10, g++ take
2:35, but not all threads are still running near the end)

At least with these data, you can adjust the 6424 (that's 40 seconds,
yes, I got to perform that with gnu on earlier beta, as far as I
remember) { with the 1691.)
So their 2227 should be 2227*1691/6424 == 586.
Which is not your result because *you cannot compare between beta*.

Post a reply to this message