First off. They said;
"Belgian researchers have assembled a relatively simple enthusiast PC
with an emphasis on graphics processing capability, which beats a
multi-million dollar supercomputer in its target application."
http://www.tomshardware.com...

"... in its target application." This is wrong.
If it were true, then they (the guys in Antwerp) would have made a
major mistake back in 2005 in purchasing the $4.6M supercomputer.

The so-called CalcUA supercomputer is used for a large varied field
of applications;
http://www.hoise.com/primeu... .

Secondly. A supercomputer can't simply be replaced by a PC running on
GPU-steroids. It really depends on the application as well as on its size.

Almost all technical complex problems today are much more
computationally complex than "simply" computing derived information
from a CT-scan! CT is usually the first application to be consider on a new
architecture, because it is directly related to human (internal) diseases
and can be directly presented to the public without any difficulty to
gain more support for future projects. Don't get me wrong here, CT is
very important! There are a lot of open problem related to CT.

Since a couple of years, the HPC market is increasing with a rapid pace
(with a two-digit grow per year). A lot of companies are now trying to
enter into this market by any means necessary. Nvidia is one of those
companies. They want to establish themself within this market by
producing such systems like the Telsa C870 graphics card to convince
people that the raw computational power of a set of GPUs can be put
to good use. Yes, but it really depends on the application under
consideration. Computing derived information from a CT-scan is a fine
task for a GPU, since the CT data is usually given in the form of a
three-dimensional uniform orthogonal structured grid which translate
to a unit-stride memory block perfect suited for a streaming architecture.
However, more complex problems don't have such an ideal grid
(data-structure), they rely in unstructured grids; sparse data-structures;
not being contiguous layout out in memory; and with an indirect
memory addressing scheme. Such things don't map well to a streaming
architecture (GPUs).

However, what bothers me is the mindless comparison of GPUs vs CPUs
and whether a PC stacked up fully of GPUs can in general stand upon a
general CPU-based supercomputer. This is ridiculous!

Tell me again how the multiplatform strategy was supposed to fix everything. Look at your results and compare them to Nintendo, which keeps thriving with exclusives. Your loyal fans are gone. And let’s not forget the pile of IPs you purchased only to let them rot.

The worst thing is this is down from last years numbers/expectations which was down from the previous years numbers/expectations. You think about how long “down” has been a consistent part of the communication and you realize those subscriptions are what’s holding that boat together. The price cut of gamespass is still a price raise in a string of price raises. They are trying to use players to offset the missing revenue. Now call of duty is out of gamespass, which is literally WHY they bought the company. To draw people to the subscription. They realized they missed a boatload of money and that wasn’t sustainable. Now you pay for it like you would have if they never bought the company… so why buy the company? Mind you those billions still need to be paid back. Yeah it gives you a chance to say we make the biggest FPS… but the internal conversations must be brutal at high level on topics that really matter. Like revenue. Profitability. And ROI for the shopping spree.

Is this like the 10th quarter in a row they're down 33% on hardware how is that even mathematically possible. You would think their mediocrity would Plateau