Quote Dr. Pande as to why ATI vs Nvidia
"Yes. It's mostly about the number of shaders: 48 vs 12. However, there are a few other quirks of the ATI hardware which make it slow down if you use too many registers -- you don't see this much for graphics, but FAH gets hit hard (hopefully they will fix this soon)."
Correction by Dr. Pande
"I typed the above way too quickly (lots going on right now). For one thing, it's NVIDIA that has the register slow down problem, not ATI (that's why we're using ATI not NVIDIA). Secondly, in addition to this issue, ATI can execute much deeper loops than Nvidia can, forcing us to use a multi-pass approach on Nvidia. The other big one is the fetch latency hiding is better on ATI than NV.
(Thanks for Mike Houston for catching this!)"
Press Release
"MARKHAM, ONTARIO, Sep 29, 2006 (CCNMatthews via COMTEX News Network) -- Today ATI (TSX:ATY)(NASDAQ:ATYT) announced that it is pioneering a new technology known as Stream Computing, that has the potential to dramatically impact almost every sector of the market. Along with leading companies and academic institutions worldwide, ATI is working to build a stream computing ecosystem, one that delivers the performance, applications, software and tools necessary to turn ATI's vision into reality.
Stream computing allows ATI graphics processors (GPUs) to work in concert with today's high-performance, low-latency computer processors (CPUs) to solve complex computational problems. Using stream computing, in simulations today processing of risk assessment models similar to those used by financial institutions' were completed 16 times faster than traditional methods(1), oil and gas companies are seeing seismic model processing increased by more than 20 times(1), and Stanford University is seeing disease research accelerated by as much as 40 times, giving them the ability to process three years worth of research data in just one month(2).
Stream computing makes use of ATI's sophisticated graphics processors (GPUs) that have until now, been used solely to calculate and render millions of pixels onto computer monitors, hundreds of times each second. Stream computing harnesses that tremendous processing power for a wide range of scientific, business and consumer computing applications, providing organizations the ability to process incredible amounts of information in significantly less time. "
"Scientific research - Today ATI's stream computing efforts are helping to save lives by driving life sciences to produce results faster in areas such as disease research, giving organizations the option to do more granular studies in the same amount of time as in the past(4). ATI announced today that Stanford University will make available a new distributed computing application that takes advantage of ATI processors for disease research. "
Quote Dr. Pande
"It's mainly about the processor on the card. The card names are very misleading, as a high number card doesn't mean a powerful processor (as far as FAH is concerned). All the 1900 class cards have the R580 processor. None of the others do. That's the deciding factor right now."
Interesting tidbits on gpus for calculations
http://techreport.com/onearticle.x/10907
First kudos to ihrstdr for sticking it out in a rather curt thread. A personal observation. I am not as patient when it comes to what I percieve as rude people. I tip my hat to you sir.
Now we know that testing was done on 25 1900xt and 25 1900xtx's.
Please note the "up to" on the shaders.
http://www.ati.com/products/radeonx1900/specs.html
All of the x1900's (save for sli) I could find where of the 36 shader variety. From ATI's site the following at the bottom of the page:
http://www.ati.com/products/radeonx1900/radeonx1900gt.html
Pulled from legion hardware:
"However, where the X1800XT features 16 pixel shaders the newer 7900GT utilizes 24 pixel shaders. On the other hand the core is clocked significantly lower at 450MHz and the memory at 1.32GHz."
PLEASE not that Stanford has not indicated any support or hint of support for the x1800xt. I note this only as it appears that the shaders may be of great interest to us here at overclockers.
One more note. There appears to have been some talk of crediting the work done on a different standard rather than the same standard for the same work unit completed. One may want to keep an eye on this as it appears that all work units can not be done on gpus and cpus are still required to accomplish some work units. Further if the work unit points is adjusted to reflect the unit accomplishing the work rather than the work itself then we may not see the huge advantage expected. There of course would/should be bonuses in the beginning to encourage gpu usage. The advantage though may not be as large as anticipated after release. We shall all of course see. If the advantage is not as anticipated then my last 1900xt has been purchased. If the advantage is too large then those that are in it for the points as well as the medicine will have gpus working while the truly "in it for the medicine" would seemingly pick up the unwanted units with the cpu's. Touch of sarcasm noted of course. Dr. Pande seems to have recognized this and I am sure is in a bit of a quendry about it.