Originally posted by: apoppin
but there is no point in arguing with you at all
:brokenheart: =*(
Originally posted by: BFG10K
You tell me. On what basis was the original claim?.any chip company is capable of making a over sized chip with 1.4bn transistors on it that can do 2 tflops. As long as the architecture is bloated- nothing can be done about the performance. Let's see how Larrabee performs... made?
If you asking me then the answer is "it's not about transistor count but what you do with it".
nVidia has a 1.4 billion transistor 65 nm GPU that has the single card performance crown.
nVidia has a product that that has an economically justification! the fact that it can sell a chip of that size is the important metric here. The fact that that product has the single card performance crown can be argued (not enough competition) but I'm not interested in playing the devil advocate.
Right now due to the nature of task at hand (rendering) and it's characteristics (extremely easy to do in parallel) there is a justification for a single massive piece of silicon.
If that massive piece of silicon was one monolithic extraordinary complex chip I'm not sure nVidia would have been able to pull it off. The fact that that chip is made up from
many identical units (SP)
IS one of the more important metric here, it is one of the reasons that fabricating a 1.4 billion transistor chip is economically justified, designing 1.4 billion transistor worth of logic in a single (or only a few) big, complex processing unit(s) would have been an order of magnitude (if not a few) harder.
If Intel would have the economical justification to design and build such products I personally would be hard pressed to believe that they would have not done so, with their know-how of chip fabrication.
The fact that nVidia did it first (making a > billion transistor chip) or at all for that matter, is because nVidia had better reason to do it. it really is that simple.
Classical CPUs manufacturer and designer just don't have that much to gain from creaming a few dozens of 'cores' into a single piece of silicon, most of the code they execute is mostly sequential. and this is exactly where Larrabee is different from Classical CPUs, Larrabee is a CPU that specialize in executing parallel code.
A vision.
Again if all transistors are ?equal? as was claimed then by that metric if I a ship a core that has nothing but caches can I clam this core is equivalent to any CPU or GPU? I think not.
What?!
IF TRANSISTOR
COUNT IS A METRIC, THEN ALL TRANSISTOR
ARE EQUAL!
Again, if transistors
count is a metric, stating that the job/function of the transistor is influential toward that count, it is exactly the point when transistors
count as a metric is a no longer relevant. the metric becomes "the count of the transistors that does X".
And if you want to go
that route... From a design perspective, if you are compering 'transistors count that does X' you should do it properly, apples to apples, and that would be processing
unit to a processing
unit (
SM to a core2duo
core)
Or maybe they can't compete with nVidia and ATi directly with rasterization so they're trying to use marketing ploys and market share muscle to force developers into their back yard?
Only one problem with that, at the end of the day these developers (or more precisely those who pay their salary)
also need an economical justification to take that path, because that path will cost, a lot, at least at first. if it (moving to software rendering) is not economically justified nobody will take that route. Intel doesn't have enough "marketing ploys and market share muscle
to force" other to take a path that will make them lose money.
Sure, TSMC played a part but I think you?re vastly downplaying nVidia's achievement here. The fact is Intel can't even do what they have pulled off with the GT200 (again cache transistors don?t count) and nVidia don?t even have their own fab like Intel does.
...
How is it nonsense? Is designing a cache transistor the same as designing a computation one? Of course not. The cache is just dumb storage; it doesn?t do any work.
...
Right, and? Again I?ll ask if I shipped a 1 4 billion transistor die with nothing but cache (i.e. it is incapable of doing any kind of computation), is that the same as a CPU or GPU that does actual work and also has 1.4 billion transistors?
I mean let?s take it a step further. Can I count the VRAM transistors used in the 1 GB RAM that is included with the GTX280? Using your argument I can because if I removed the VRAM performance would go down. Right?
See above.
ActiveX? I think you mean DirectX.
*scratches head* Arr.. yeah.
/many edits for many typos and bad grammar