By: Robert Myers (rbmyersusa.delete@this.gmail.com), October 19, 2012 7:24 am
Room: Moderated Discussions
anon (anon.delete@this.anon.com) on October 19, 2012 2:27 am wrote:
>
> Cough up the numbers,
> sir. Instead of handwaving, let's just see some of your numbers. Cite some
> studies or published results.
>
> For example, you claimed (without citing
> anything of course) that 3d FFTs don't scale well, but at least for some
> problems, they actually do, as this study shows. I can frame it as a reply to an
> exact quote of yours just a couple of posts up, if you would like.
Very near the beginning of this exchange, I attempted to beg off. I had no intention of ever going near this discussion again, because I am just tired of it.
A discussion very much like the one we just had and that I described is in the archives of comp.arch. It's been a long, long, long, tedious discussion, and you, despite your great importance to the world, haven't been a part of it. The discussion, with the numbers, of how Blue Gene scaled up to thousands of nodes and beat *all* of the competition happened, with citations. Go find it yourself. It went exactly the way I have described it. The shootout is only a few years back, and surely someone with your comprehensive knowledge of HPC would have been at the most important HPC conference of the year, and heard the results reported first hand.
Now you come along with another paper that appears to claim something on the level of "we have discovered perpetual motion." I'm quite sure the paper isn't claiming that, and I'm quite sure that, if I put enough time into it, I will discover the ghost in the machine. If I had knowledge only of one version of Blue Gene, or didn't have knowledge of how later versions of Blue Gene compared to how everything else that anyone could put forth performed (it beat them), I might be interested in your paper, but such is not the case.
If you're really interested in the issue and not in trolling, you have my email address, and I will undoubtedly find the time to figure out what's really going on in your magical paper. I don't have your email address, so, if and when I get around to it, you will never know.
Robert.
>
> Cough up the numbers,
> sir. Instead of handwaving, let's just see some of your numbers. Cite some
> studies or published results.
>
> For example, you claimed (without citing
> anything of course) that 3d FFTs don't scale well, but at least for some
> problems, they actually do, as this study shows. I can frame it as a reply to an
> exact quote of yours just a couple of posts up, if you would like.
Very near the beginning of this exchange, I attempted to beg off. I had no intention of ever going near this discussion again, because I am just tired of it.
A discussion very much like the one we just had and that I described is in the archives of comp.arch. It's been a long, long, long, tedious discussion, and you, despite your great importance to the world, haven't been a part of it. The discussion, with the numbers, of how Blue Gene scaled up to thousands of nodes and beat *all* of the competition happened, with citations. Go find it yourself. It went exactly the way I have described it. The shootout is only a few years back, and surely someone with your comprehensive knowledge of HPC would have been at the most important HPC conference of the year, and heard the results reported first hand.
Now you come along with another paper that appears to claim something on the level of "we have discovered perpetual motion." I'm quite sure the paper isn't claiming that, and I'm quite sure that, if I put enough time into it, I will discover the ghost in the machine. If I had knowledge only of one version of Blue Gene, or didn't have knowledge of how later versions of Blue Gene compared to how everything else that anyone could put forth performed (it beat them), I might be interested in your paper, but such is not the case.
If you're really interested in the issue and not in trolling, you have my email address, and I will undoubtedly find the time to figure out what's really going on in your magical paper. I don't have your email address, so, if and when I get around to it, you will never know.
Robert.
