Why ATI delayed the R600.....

[Roger]

Taken the "Red Pill" yet ?

All this writing and how many forgot the date? <)

John Lewis

Roger Halstead (K8RI & ARRL life member)
(N833R, S# CD-2 Worlds oldest Debonair)
www.rogerhalstead.com

 

[TheSmokingGnu]

The only way it could do that would be to not support an external bus,
which would make it damned difficult to attach peripherals.

You could conceivably use a PCIe bus without the x16 connections that a
video card needs; it wouldn't be very useful, as you say.

How about the ultra high bandwidth path between the CPU and GPU that is
made possible by having them both on the same die?

That's the other part of it, yes. eDRAM allows for some insanely wide
bus paths between processors, which is why you don't need very much of it.

 

[TheSmokingGnu]

Does that say the user can't buy a GeForce card?
No, it says the user would be more likely to buy a RenderX as it would
be cheaper than buying an entire card from the competition.

Buy != Use.

The article, for example, says nothing about how the Excellon deals with
external cards, or how the system could deal with having both external
and internal GPU chips.

One can easily take the implication that they were speaking of the
Excellon as a discrete system of units, and that competitor's system
would be more expensive.

Now if video cards still used AGP and none of those Excellion
motherboards came with an AGP slot, you might have had a point.

Sub PCIe x16 for AGP, and you've got it.

TheSmokingGnu

 

[J. Clarke]

Buy != Use.

The article, for example, says nothing about how the Excellon deals
with external cards, or how the system could deal with having both
external and internal GPU chips.

One can easily take the implication that they were speaking of the
Excellon as a discrete system of units, and that competitor's system
would be more expensive.


Sub PCIe x16 for AGP, and you've got it.

The thing is, the PCI Express bus isn't built into the chip, it's built
into the north bridge or other glue chip, which with AMD processors
attaches to one of the hypertransport channels that is also used for
inteprocessor communication in multiprocessor systems.

To remove PCI Express x16 they'd have to cripple the hypertransport
channels in a major way.

 

[John Lewis]

The thing is, the PCI Express bus isn't built into the chip, it's built
into the north bridge or other glue chip, which with AMD processors
attaches to one of the hypertransport channels that is also used for
inteprocessor communication in multiprocessor systems.

To remove PCI Express x16 they'd have to cripple the hypertransport
channels in a major way.

--

er, have you taken the "red pill" yet? See the article

John Lewis

 

[Ham Pastrami]

If, say, I purchase an Intel chip now, it's just an Intel chip. It only
does the CPU job, and when I buy a motherboard, I'm free to use
essentially any other competing product (even the motherboard!) with it.

Same thing with existing AMD chips. I can buy that chip and /only/ that
chip. Any other component is freely choosable by the consumer.

This is a contrived example, as how often do people upgrade their
motherboards while keeping the same CPU? The converse is far more likely --
people wanting to upgrade their CPUs without upsetting the rest of the
system. Thus, given any motherboard, you are locked into a vendor, be it
Intel or AMD.

Furthermore, the motherboard has to have the specific kind of socket that
your CPU uses, which can vary even within a given brand and model of CPU, so
even if you were to buy another board for the same brand, it still might not
work.

This architecture locks the consumer into a choice of video hardware,
and further subjugates them by forcing their upgrade path into, surprise
surprise, more of the same. If they wanted to switch mid-year from an
ATI setup to an nVidia one, they're SOL, thus proprietary.

The only reason why they couldn't switch is if nVidia chose not to make
products for that architecture. Up til now, ATI and nVidia were both simply
adhering to Intel's AGP and PCIe specs. Now there's competition between two
proprietary architectures, neither of which, as far as I can see, is more
proprietary than the other.

 

[Ham Pastrami]

*disclaimer against April fool's: even if this report is just a hoax, the
principles still apply.

 

[GT-Force]

For the ppl who missed it, at the REAL end of the article:

"Disclaimer
Please note that the "facts" and questions and answers in the previous pages
do not necessarily reflect the personal or professional opinions of anyone
at Team ARP. They were created with comic intent in mind."

GT