http://www.dailytech.com/article.aspx?newsid=5245
In the discrete mobile segment, ATI saw shipments decline dramatically on a quarterly basis while Nvidia saw shipments increase sequentially. Nvidia grew discrete mobile segment share from 37% in Q1’06 to 53% in Q3’06. ATI’s segment share fell from 63% in Q2’06 to 47% in Q3’06.
AMD just cant get a break :lol: A very unintellegent statement given the following: http://www.theregister.co.uk/2006/11/21/darpa_petascale/
http://www.ornl.gov/ornlhome/print/press_release_print.cfm?ReleaseNumber=mr20060825-00
http://www.tacc.utexas.edu/research/users/features/track2.php
http://www.hpcwire.com/hpc/1099300.html
http://www.supercomputingonline.com/article.php?sid=12402
http://www.hpcwire.com/hpc/701937.html
http://www.lbl.gov/CS/Archive/news081006.html
http://www.linuxhpc.org/stories.php?story=06/06/27/1618028
http://www.llnl.gov/pao/news/news_releases/2005/NR-05-11-04p.html
http://www.itjungle.com/tlb/tlb091206-story01.html I count 10 contracts for AMD for computers that will be in the Top15, 1 for IBM and none for Intel. Total is a little over $1.2 billion for AMD based supercomputers. That means Intel is left looking for scraps. Don't give the excuse that Core 2 wasn't around , it was. The most prestigeous benchmarks are awarded by IEEE/DARPA and the cutoff date was OCT 26.2006. Results , take a look:
http://www.hpcchallenge.org/ I see 8 placements for IBM and 4 for AMD out of twelve, 0 for Intel.
If you haven't figured it out yet in the HPC market which is where about 37% of profits are with 3% of sales , Intel is DOA. Gaming is less than .1% or 60% of the value of either the IBM P7 or Cray Cascade contracts by themselves. Ignoring a market segment that acounts for 37% of profits for the manufacturers(becuase they get the full sales price and don't have to split with wholesalers) is more than misleading.
AS for ATI, Stream is where the money is right now. http://www.hpcchallenge.org/ Considering that one of these cards sells for about $2500-$4000 depending on configuration and no middle men to take a cut. That means that each of these cards is worth as much profitwise as 10-15 of nVidia's 8800s. Those are on the market and in short supply. Stanford University will give you a concise description of the problem facing nVidia:
"Which cards are supported?
We now support serveral classes of GPU boards, including X1600, X1800, and X1900 class GPU's from ATI. At the launch, we supported X1900 cards only. X1800 cards do not provide the performance seen in X1900's and so we strongly recommend X1900 class cards (although we now officially support X1800). X1900 and X1800 cards are actually quite different -- they have different processors (R520, R530 vs. the R580 [in the X1900 series]). The R580 makes a huge difference in performance -- its 48 pixel shaders are key, as we use pixel shaders for our computations.
"What about video cards with other (non-ATI) chipsets?
The R580 (in the X1900XT, etc.) performs particularly well for molecular dynamics, due to its 48 pixel shaders. Currently, other cards (such as those from nVidia and other ATI cards) do not perform well enough for our calculations as they have fewer pixel shaders. Also, nVidia cards in general have some technical limitations beyond the number of pixel shaders which makes them perform poorly in our calculations. "
http://folding.stanford.edu/FAQ-ATI.html
So basicly Intel and nVidia face some tough engineering problems for uses in the real world outside of gaming. At SC06, Ken Kennedy(you don't know who he is look up his bio at the IEEE website) put it rather bluntly for Intel they have the memory on the wrong side of the CPU. With the northbridge in the way between the memory and the cpu Intel is not very successful in running GPU accleration and can't run the IBM Cell BE as an accelerator. A single core opteron with GPU acceleration is about 400GFLOPS or 10X Intel's Core 2. K8L with HTT 3.0 will allow the GPU to access the system memory and not be limited by onboard memory. K8L is the basis for the Cray XT4 and puts Intel further behind in the HPC market. Intel didn't even have a conceptual offering to compete with the IBM P7 or Cascade. Sun did have a new version of SPARC.
Mark Twain put it best : "there are lies , damned lies, and statistics."