[Beowulf] Larrabee - Mark Hahn's personal attack

[Vincent Diepeveen]

On Jan 27, 2012, at 5:34 PM, Prentice Bisbal wrote:

> On 01/27/2012 11:12 AM, Vincent Diepeveen wrote:
>> And both are seem failures from user viewpoint, maybe not from intels
>> income viewpoint,
>> but from intels aim to replace and/or create a new long lasting
>> architecture
>> that can even *remotely* compete with other manufacturers,
>> not to mention far too high pricepoints for such cpu's.
>
> This argument is ridiculous. Just because two completely different
> technologies (architectures) both fail, doesn't make them similar.
>
> That's like saying a Ford Edsel and Pontiac Aztek are similar cars.
>
>> Assuming you're not completely born stupid, i assume you will realize
>> that IN ORDER to run
>
> Calling someone "completely born stupid" is unacceptable behavior.

Whereaas everyone knows the statements of intel on larrabee there and
that without cache coherency you can't multithread and everything  
also has
to be done blocked - so there is zero compatibility with x64 then and  
any compatibility
then cannot get garantueed.

You know this really well - yet you kept yourself dumb there trying  
to cheap score.

As without cache coherency of course it's easy to build big cpu's  
that scale well,
yet they don't work x64 then.

of course intel will be forced to design some kick butt design  
somewhere in future that's
not x64 compatible at all which isn't using things like cache coherency.

Which isn't remotely the idea of larrabee.

That's why you wrote it down as such.

>> most existing x64 codes, it needs to have cache coherency, and that
>> it always has been
>> presented as having exactly that.
>> Which is one of reasons why the architecture doesn't scale of course.
>
> Cache-coherent systems don't scale well? Really? SGI Origins were  
> ccNUMA
> systems, and they scaled well.
>

Indeed this didn't scale near lineair in price.

Each Origin3800 @ 64 processors @ 1.5Ghz was exactly 1 million dollar,
whereas a simple normal x64 cpu at the time had a price similar to  
the square root of that.

In GPU's it all scales very cheap, and when using cache coherency you  
start to lose that
scaling.

Yields will go down of course. Most manufacturers need a pretty high  
yield to sell a chip at
any decent price, so production costs of a larrabee chip in the same  
proces technology as a GPU,
having the same performance will be a huge factor higher. That also  
will cause intel to really sell few of them.

You would consider buying a larrabee at 1 million dollar a card?


>> Well you can forget about them running your x64 fortran codes on it
>> at any fast speed.
>>
>> You need to total rewrite your code to be able to use vectors of
>> doubles,
>> and in contradiction to GPU's where you can indirectly with arrays
>> see each PE or each 'compute core'
>> (which is 4 PE's of in case of AMD-ATI that can execute 1 double a
>
> This argument makes no sense in the context of this discussion.  You
> need to do a significant rewrite of your code to take advantage of  
> GPUs,
> too, so how are GPUs better?

If you need to rewrite it anyway, why not get a much faster  
performance at part of the price?

It's the same effort you have to do.



>
>> cycle),
>>
>> Such lookups are a disaster at larrabee - having a cost of 7 cycles
>> for indirect lookups,
>> so you really need to use vectors.
>>
>> Now i bet majority of your oldie x64 code doesn't use such huge  
>> vectors,
>> so to even get some remote performance out of it, a total rewrite of
>> most code is needed,
>> if it can work at all.
>>
>> We can then also see the insight that GPU's are total superior to
>> larrabee at most terrains and
>> most importantly at multiplicative codes.
>>
>> As you might know GPU's are worldchampion in doing multiplications
>> and CPU's are not.
>>
>> Multiplication happens to be something that is of major importance
>> for the majority of HPC codes.
>> Majority i really mean - approaching 90% at the public  
>> supercomputers.
>
> I'm at a loss for words...
>

http://www.nwo.nl/nwohome.nsf/pages/NWOP_8DEEKL_Eng

title:       "Overview of recent supercomputers 2010"
Author: Aad van der Steen

>
> Prentice
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin  
> Computing
> To change your subscription (digest mode or unsubscribe) visit  
> http://www.beowulf.org/mailman/listinfo/beowulf