Article: Parallelism at HotPar 2010
By: AM (myname4rwt.delete@this.jee-male.com), August 10, 2010 11:03 pm
Room: Moderated Discussions
none (none@none.com) on 8/10/10 wrote:
---------------------------
>AM (myname4rwt@jee-male.com) on 8/10/10 wrote:
>---------------------------
>[...]
>>Besides, and closer to the topic, the claimed real advantage of 2.5x-5x of GPUs
>>over CPUs (by Mark Roulo) was not demonstrated so far even for such a small collection
>>of papers. Nor was it shown by David Kanter that these reports (*at least* these
>>10) are all misleading, bogus, erroneous or fishy in any other manner.
>
>The problem is that even if we "demonstrate" these 10
>articles are bogus, we won't have proved a 100x speedup
>isn't achievable (that's a simple logic problem), and you,
>or someone else, might come back with another pool of
>articles.
Right. And tbh, I don't believe that the "real" perf advantage of anywhere near to 2.5x-5x as claimed can be shown even for this small selection.
>I picked 2 of these 10 only to find they were not fair, I
>consider this as enough. I'd really like to have something
>>20x faster than my CPU, but given how biased these 2 papers
>are,
One report with unspecified CPU and another which not only provided some data about CPU used (more expensive, by the way, than the GPU used for comparison), but has sources available IIRC hardly counts as a biased or unfair job, let alone bogus, erroneous or misleading.
I don't want to waste more of my time. So why don't
>you go through these 10 papers and show us the one you
>think is the most convincing?
Nice bait, but no thanks. With research results widely available, I'd still like to see how the real perf gap can possibly be around 2,5x-5x as claimed by Mark Roulo. Ditto wrt David Kanter, but it seeems he hasn't done any research whatsoever.
---------------------------
>AM (myname4rwt@jee-male.com) on 8/10/10 wrote:
>---------------------------
>[...]
>>Besides, and closer to the topic, the claimed real advantage of 2.5x-5x of GPUs
>>over CPUs (by Mark Roulo) was not demonstrated so far even for such a small collection
>>of papers. Nor was it shown by David Kanter that these reports (*at least* these
>>10) are all misleading, bogus, erroneous or fishy in any other manner.
>
>The problem is that even if we "demonstrate" these 10
>articles are bogus, we won't have proved a 100x speedup
>isn't achievable (that's a simple logic problem), and you,
>or someone else, might come back with another pool of
>articles.
Right. And tbh, I don't believe that the "real" perf advantage of anywhere near to 2.5x-5x as claimed can be shown even for this small selection.
>I picked 2 of these 10 only to find they were not fair, I
>consider this as enough. I'd really like to have something
>>20x faster than my CPU, but given how biased these 2 papers
>are,
One report with unspecified CPU and another which not only provided some data about CPU used (more expensive, by the way, than the GPU used for comparison), but has sources available IIRC hardly counts as a biased or unfair job, let alone bogus, erroneous or misleading.
I don't want to waste more of my time. So why don't
>you go through these 10 papers and show us the one you
>think is the most convincing?
Nice bait, but no thanks. With research results widely available, I'd still like to see how the real perf gap can possibly be around 2,5x-5x as claimed by Mark Roulo. Ditto wrt David Kanter, but it seeems he hasn't done any research whatsoever.