Re: [eigen] Eigen Benchmark (number of FLOPS) |

[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]

*To*: eigen@xxxxxxxxxxxxxxxxxxx
*Subject*: Re: [eigen] Eigen Benchmark (number of FLOPS)
*From*: Benoit Jacob <jacob.benoit.1@xxxxxxxxx>
*Date*: Mon, 19 Dec 2011 08:04:17 -0500
*Dkim-signature*: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :content-type; bh=O6CGHHkdG6Q0oOiqbaJdMXBvNfWqRfdhU4vBEiH4U6w=; b=aI2bMdV3l+d94o/UoCwGzkpPHb/yfpaeOsFnniBqFl8eZlyzzR+siHMYfMTuR80FB9 bkkypeczSrTTFZaRyzkzcUETqD+0ATiqKKpKC1CKcHXu65bO+PoEZhjoGXxL8g5YUt9n NunWu6ScSeUx8pQTKzLM7IDeRke2o7YezEM1M=

2011/12/18 Not Sure <kuerzn@xxxxxxxxxxxxxx>:
> Hi!
> On the website http://eigen.tuxfamily.org/index.php?title=Benchmark you
> show some pretty impressive comparison of Eigen's performance.
> I have two consecutive questions:
>
> 1. Do the different implementations/library of BLAS functions differ in
> the number of floating point operations (FLOP) they need to complete?
>
> 2. If yes: Did you calculate the FLOP per second using some reference
> FLOP number or the number of actual performed FLOP of each
> implementation/library?
My understanding is that the FLOPs in this benchmark are based on a
theoretical estimate of the number of operations required by a given
algorithm. For example: 2*n^3 for GEMM on nxn matrices, etc. This is
sometimes inaccurate for small sizes, but at least the scale is the
same for all contenders.
Benoit
>
> Thank you for your answer,
>
> Johannes
>
>