D
David Kanter
George said:
Considering that hardware decimal support is considerably older than
PPC, I'd harder consider this any sort of novel integration.
DK
Considering that hardware decimal support is considerably older than
PPC, I'd harder consider this any sort of novel integration.
DK
R
Robert Redelmeier
In comp.sys.ibm.pc.hardware.chips David Kanter said:
Even x86 has some hardware decimal support (AA and FB
instructions), but this sounds like hardware decimal full
word multipliers. Maybe not a first, but certainly very rare.
Sure it's a collosal waste of transistors, but they're dirt
cheap and go begging for intelligent uses (dual?). The big
[small?] advantage of decimal FP is I/O. Converting decimal
FP to binary FP is a pain, and binFP to decimalFP is much worse.
-- Robert
K
krw
Decimal arithmetic is not rare at all, for IBM.
Financial institutions insist on decimal arithmetic. IBM's
customers are...
R
Robert Redelmeier
In comp.sys.ibm.pc.hardware.chips krw said:
Very true. Many IBM mainframes were designed for COBOL
which uses BCD.
Yep. Although AFAIK most financial clacs are done with
integers, in things like mils. Decimal floats have much
the same problems as binary floats -- round-off.
-- Robert
D
David Kanter
Robert said:
Yup. A lot of the high-end machines are used for financial stuff.
My understanding is that you obviously want to try and keep to integers
if you can, but even then, you can still have wrap around. IBM opted
to add this feature in on the basis that it can provide a nice speed up
over existing software libraries.
DK
G
George Macdonald
D'oh - read the article. This is the start of a new general vertical
approach to processor features.
Yes we all know that "commercial" mainframes have, and started with,
decimal arithmetic and that early binary FP machines were horrendously
expensive, and had limited business application, in comparison, C.F. 7074 &
7094. Even large corps used to buy "time" on a 7094... for ~$500./hour.
K
krw
(I ran the searches before but WinBlows hung when I was composing
the original response)
No I didn't remember, but evidently so:
INTC AMD
1996 431 184
1997 417 273
1998 737 557
1999 769 831 <==
2000 838 1056 <==
2001 873 1090 <==
2002 1132 1153 <==
2003 1634 910
2004 1631 807
2005 1579 533
Interesting. Seems AMD had a big push in the K7 days and has fallen
off since.
G
George Macdonald
Interesting indeed - November 2002 was when AMD coughed up a reputed $46M
to Big Blue in order to get SOI umm, working. I guess they err, let a few
folks go after that... or could it be because they have process engineers
working at IBM facilities now and the IP flow is one-way??
K
Keith
fammacd=! said:
No, I don't think AMD had any serious process technology before
2002 to offer another "way". The two companies have always been
quite close, even when IBM wasn't using AMD's stuff. Why AMD went
to Moto for SOI (who didn't have it themselves) is one of the
mysteries of the decade.
Also remember, there is at least a two year gap between a patent
being granted and its filing, not to mention the time to get a
patent written. No, it looks like they were driven in the K7 days.
Maybe they're going "on the cheap" now?
T
Tony Hill
ATI's main success seemed to come in the notebook chipset market. They
weren't beating out Centrino, but they did fill in most a lot of other
places. As for the low-end chipset production, I'm not sure that
Intel cares about trusting any company. They'll leave it to SiS or
VIA probably, no profit to be made bellow Intel's low-end chips (which
are pretty darn cheap these days... they are well beyond the
production shortages of the past while).
Niche products will continue to exist, just like the separate physics
co-processors mentioned above. However to make it in the PC world as
a mainstream thing it's a lot tougher. Graphics has proven to be the
one and only area that has really resisted integration for the reasons
mentioned previously.
ATI Radeon Xpress 200. The chipset was widely available, but the
"SidePort" (ATI's name for it) feature was very rarely used.
G
George Macdonald
I was under the impression that Intel was using those Ati chipsets on Intel
branded mbrds so a *certain* amount of "qualification" would be umm, in
order.
Via would surely have been a tough swallow there.
Y
Yousuf Khan
Tony said:
Yup, in the AMD notebook market, ATI chipsets rule, followed by SIS of
all people.
As for Intel and its chipset problems, I'm not sure why Intel didn't
just contract out its own chipsets to contract fabs? At the very least
contract out the very oldest chipsets to the fabs.
Yousuf Khan
Y
Yousuf Khan
krw said:
Also interesting is that Intel has stepped up its yearly patent
generation efforts. Maybe Intel is just trying to patent just about
anything? It might be a defensive thing, since there are a lot of
IP-only companies out there now (e.g. Rambus), which will try to make a
buck off of anything minor. This way Intel gets to prevent some IP
squatting.
As for AMD, it seems its biggest patent generating years where the years
it was first designing the K8 architecture. Right now it's in a somewhat
of a tweak-existing-stuff mode.
Yousuf Khan
K
krw
Perhaps. Defense is one of the best reasons for a large company to
have a large portfolio.
remember there is at *least* a two year lag between patent filing
and grant. Add another year or two before that. Without going
though each patent, it appears that it was the K7 that brought out
AMD's patent frenzy. Perhaps they knew they'd have to license
everything K8 to INTC anyway?