By: anon (anon.delete@this.anon.com), October 19, 2012 4:47 am
Room: Moderated Discussions
anon (anon.delete@this.anon.com) on October 19, 2012 2:34 am wrote:
> Robert Myers (rbmyersusa.delete@this.gmail.com) on October 18, 2012 9:47 am
> wrote:
> > anon (anon.delete@this.anon.com) on October 17, 2012 8:36 pm
> wrote:
> >
> > >
> >
> > > You have thin skin for someone who
> is
> > > happy to talk about everyone
> > else being wrong. I would have
> thought you'd be
> > > quite accustomed to being
> > called to answer
> for your claims.
> > >
> >
> > I have a thin skin for people who
>
> > adopt a presumption of superior knowledge and who post
> anonymously.
>
> You made claims, I'm asking you to provide evidence. That you
> equate that with my presuming superior knowledge (for daring to question you
> anonymously?) or even care about such a thing, is weird.
Seeing as you were getting upset about me not directly quoting you, I'll give some (from here and comp.arch thread you linked).
"As to the running faster to stay in the same place analogy, that's
what I have long suspected is going on, and it's one of the most
disheartening things you have said to me.
Computers have fulfilled their promise only in the area of computer
graphics and animation."
"As to computational physics, I have long had serious doubts as to
whether bigger computers at this point are producing anything more
than more visually-engaging graphics. They may well be an obstacle to
insight and real progress--especially as the vested interests that are
more interested in the computers than in the science become ever more
vested, visible, and powerful."
"If your answer to every query about why you are not building better computers is that you can't, then my response to every claim made by everyone in the business is that you should stop building *bigger* computers until you know how to build *better* computers. "
"Aside from [flops/watt], the business has essentially turned into a marketing claims factory."
From this, it certainly seems like you're saying HPC field is not achieving anything other than linpack. You accuse me of misrepresenting your position, but from what you have posted and linked to, that seemed to be pretty close to your position, if slightly exaggerated.
And you've said something like this:
"Since you seem to think that caches and changed software assumptions can address all problems of importance, you may have to be told explicitly that the transform in question is the Fourier transform. The last time I was paying close attention, Blue Gene could use all of 512 of its tens of thousands of processors effectively in doing a volumetric FFT."
And you still have not clarified the problem you are talking about when faced with the code and paper I cited, so we'd have to conclude that you have not in fact being paying close attention for a long time. The turbulence simulation using 3d FFTs scales 98% (i.e., extremely well) when going from 16K to 32K CPUs on BG system. So in fact a great deal of progress has been made in software and hardware scalability on these large systems since you were last paying attention.
You now tried to say scalability is probably achieved by sacrificing accuracy, again without actually citing anything. There was no indication that the simulations being done were not producing useful results.
Again I'm not trying to prove myself, or demonstrate superior intellect or knowledge. I'm posting anonymously, have nothing to gain. I'm hardly claiming any knowledge beyond simply asking you questions, and linking things which are easily found. In fact I freely admit my knowledge of HPC is very limited; an expert in the field could easily best me if I attempted to debate them on it.
> Robert Myers (rbmyersusa.delete@this.gmail.com) on October 18, 2012 9:47 am
> wrote:
> > anon (anon.delete@this.anon.com) on October 17, 2012 8:36 pm
> wrote:
> >
> > >
> >
> > > You have thin skin for someone who
> is
> > > happy to talk about everyone
> > else being wrong. I would have
> thought you'd be
> > > quite accustomed to being
> > called to answer
> for your claims.
> > >
> >
> > I have a thin skin for people who
>
> > adopt a presumption of superior knowledge and who post
> anonymously.
>
> You made claims, I'm asking you to provide evidence. That you
> equate that with my presuming superior knowledge (for daring to question you
> anonymously?) or even care about such a thing, is weird.
Seeing as you were getting upset about me not directly quoting you, I'll give some (from here and comp.arch thread you linked).
"As to the running faster to stay in the same place analogy, that's
what I have long suspected is going on, and it's one of the most
disheartening things you have said to me.
Computers have fulfilled their promise only in the area of computer
graphics and animation."
"As to computational physics, I have long had serious doubts as to
whether bigger computers at this point are producing anything more
than more visually-engaging graphics. They may well be an obstacle to
insight and real progress--especially as the vested interests that are
more interested in the computers than in the science become ever more
vested, visible, and powerful."
"If your answer to every query about why you are not building better computers is that you can't, then my response to every claim made by everyone in the business is that you should stop building *bigger* computers until you know how to build *better* computers. "
"Aside from [flops/watt], the business has essentially turned into a marketing claims factory."
From this, it certainly seems like you're saying HPC field is not achieving anything other than linpack. You accuse me of misrepresenting your position, but from what you have posted and linked to, that seemed to be pretty close to your position, if slightly exaggerated.
And you've said something like this:
"Since you seem to think that caches and changed software assumptions can address all problems of importance, you may have to be told explicitly that the transform in question is the Fourier transform. The last time I was paying close attention, Blue Gene could use all of 512 of its tens of thousands of processors effectively in doing a volumetric FFT."
And you still have not clarified the problem you are talking about when faced with the code and paper I cited, so we'd have to conclude that you have not in fact being paying close attention for a long time. The turbulence simulation using 3d FFTs scales 98% (i.e., extremely well) when going from 16K to 32K CPUs on BG system. So in fact a great deal of progress has been made in software and hardware scalability on these large systems since you were last paying attention.
You now tried to say scalability is probably achieved by sacrificing accuracy, again without actually citing anything. There was no indication that the simulations being done were not producing useful results.
Again I'm not trying to prove myself, or demonstrate superior intellect or knowledge. I'm posting anonymously, have nothing to gain. I'm hardly claiming any knowledge beyond simply asking you questions, and linking things which are easily found. In fact I freely admit my knowledge of HPC is very limited; an expert in the field could easily best me if I attempted to debate them on it.
