| Author |
Message |
Anton Ertl
Guest
|
 Posted:
Wed Mar 02, 2005 1:32 pm Post subject:
Re: Free advice to AMD |
|
|
"kingzog" <kingzog_1@lycos.com> writes:
| Quote: | It seems to me that there might be a market for something like 8xK7
class on a chip with a modern bus to the outside world. In some ways
it's taking the Cell route of having a bunch of simpler cores, except
that these would just be older x86 designs. There some evidence that
individual core performance can be sacrificed in server applications to
get more cores on chip.
|
Looking at <e90782f7.0410071843.1032a498@posting.google.com>, it seems
that someone at AMD has looked at a simular idea; you'll also find the
reason why they don't pursue it for now.
- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html |
|
| Back to top |
|
 |
Terje Mathisen
Guest
|
| Posted:
Wed Mar 02, 2005 4:23 pm Post subject:
Re: Free advice to AMD (worth every penny) |
|
|
Bill Todd wrote:
| Quote: | del cecchi wrote:
....
Since the proposal is x86-64, what's the Microsoft windows policy?
Last I heard, I think they (unlike Oracle) plan to count chips, not cores.
|
At least for now, they are counting chips, making a dual-thread P4 a
single cpu, and a pair of them still just a pair, not four.
Terje
--
- <Terje.Mathisen@hda.hydro.com>
"almost all programming can be viewed as an exercise in caching" |
|
| Back to top |
|
|
Grumble
Guest
|
| Posted:
Wed Mar 02, 2005 5:15 pm Post subject:
Re: Software licensing for Niagra |
|
|
Anton Ertl wrote:
| Quote: | [...] the memory controller of the Niagra yet?
|
Is Pfizer now a competitor for Sun? |
|
| Back to top |
|
|
kingzog
Guest
|
| Posted:
Wed Mar 02, 2005 6:26 pm Post subject:
Re: Free advice to AMD |
|
|
Anton Ertl wrote:
| Quote: |
Looking at <e90782f7.0410071843.1032a498@posting.google.com>, it
seems
that someone at AMD has looked at a simular idea; you'll also find
the
reason why they don't pursue it for now.
|
Thanks, that's what I was looking for.
Zog. |
|
| Back to top |
|
|
Sander Vesik
Guest
|
| Posted:
Wed Mar 02, 2005 6:30 pm Post subject:
Re: Software licensing for Niagra |
|
|
Anton Ertl <anton@mips.complang.tuwien.ac.at> wrote:
| Quote: | Sander Vesik <sander@haldjas.folklore.ee> writes:
A lot of the obvious software to run on it doesn't have per-cpu licences
or comes from companies (which AFAIK includes Sun) that treat all multi-core
CPU-s as one CPU for the licence purposes. You probably wouldn't run
Oracle on it.
Why not? Supposedly it is database software like Oracle that runs on
a Niagra core almost as well as on more powerful CPUs, because this
software is supposedly always waiting for memory anyway. If that is
the case, why would Oracle not count cores or hardware threads when
determining the power of the system?
|
Unless its is going to be sold as a cheap, "kill-everything in entry
level" systems, it is going to be competing (for running oracle type
of things) with 490s/890s or follow-ups, by then probably also
with USPARC VI+. Which is 8-16 fast cores and lots of cache and
memory bandwidth.
I don't think not being a good set (IMHO anyways) for running big heavy
databases really substracts much from the usefulness of Niagra.
| Quote: |
Ok, a limiting factor for such a memory-limited application will be
the power of the memory subsystem, so maybe Oracle should license per
memory channel or somesuch:-)
BTW, is anything known about the memory controller of the Niagra yet?
|
I haven't seen anything new on Niagra details in quite a while.
--
Sander
+++ Out of cheese error +++ |
|
| Back to top |
|
|
Grumble
Guest
|
| Posted:
Thu Mar 03, 2005 1:47 pm Post subject:
Re: Software licensing for Niagra |
|
|
Sander Vesik wrote:
| Quote: | I haven't seen anything new on Niagra details in quite a while.
|
Rogntudjuuu!
Misspelling "Niagara" makes baby Jesus cry. |
|
| Back to top |
|
|
Anton Ertl
Guest
|
| Posted:
Thu Mar 03, 2005 3:00 pm Post subject:
Re: Software licensing for Niagra |
|
|
Grumble <devnull@kma.eu.org> writes:
| Quote: | Misspelling "Niagara" makes baby Jesus cry.
|
What makes you think that Niagra is a misspelling of Niagara? Is
Itanium a misspelling of Titanium? And why not Viagra?
- anton
--
M. Anton Ertl Some things have to be seen to be believed
anton@mips.complang.tuwien.ac.at Most things have to be believed to be seen
http://www.complang.tuwien.ac.at/anton/home.html |
|
| Back to top |
|
|
Grumble
Guest
|
|
| Back to top |
|
|
Sander Vesik
Guest
|
| Posted:
Fri Mar 04, 2005 7:10 am Post subject:
Re: Software licensing for Niagra |
|
|
Grumble <devnull@kma.eu.org> wrote:
| Quote: | Sander Vesik wrote:
I haven't seen anything new on Niagra details in quite a while.
Rogntudjuuu!
Misspelling "Niagara" makes baby Jesus cry.
|
I never got the baby jesus reference so let him cry :P But yeah, it
was a silly misspelling.
--
Sander
+++ Out of cheese error +++ |
|
| Back to top |
|
|
Douglas Siebert
Guest
|
| Posted:
Sat Mar 05, 2005 8:41 pm Post subject:
Re: Free advice to AMD |
|
|
anton@mips.complang.tuwien.ac.at (Anton Ertl) writes:
| Quote: | "kingzog" <kingzog_1@lycos.com> writes:
It seems to me that there might be a market for something like 8xK7
class on a chip with a modern bus to the outside world. In some ways
it's taking the Cell route of having a bunch of simpler cores, except
that these would just be older x86 designs. There some evidence that
individual core performance can be sacrificed in server applications to
get more cores on chip.
Looking at <e90782f7.0410071843.1032a498@posting.google.com>, it seems
that someone at AMD has looked at a simular idea; you'll also find the
reason why they don't pursue it for now.
|
Which makes it seem even more useful, instead of going from dual core at
90nm to quad core at 65nm, to go to dual K8 cores + a bunch of little
486 (but supporting everything through SSE3) cores instead. What Mitch
says in that referenced post makes sense, current benchmarks don't show
nearly enough parallelism to make something as radical as a 8x486 CPU
seem reasonable in comparison with a single core Opteron. But once you
get beyond simple dual core, making them all big and complex sounds self
defeating. If Intel and AMD are assuming that everyone is going to go
out and thread their applications as much as possible in preparation for
the dual core CPUs coming out soon, the quad cores that will arrive by
the end of next year, and the "multicore" CPUs after that, if one vendor
moved first with a complex + simple combination to support many more
threads up front, it might be a winner. Maybe not in SPECint (though it
would sure show up nicely in SPECintrate) and perhaps not at first in
many benchmarks. But a few overly threaded games and applications might
show this new CPU several times better and that would open the eyes of
some people. And it sounds tailor made for servers, especially ones that
run a lot of lightweight threads like say Java.
Yeah yeah, I know the objections, "hard job for the OS to schedule tasks
when there are two CPU types with very different characteristics". All I
can say there is that if you think its a hard job, the OS could devote one
of those little 486 cores to run a scheduler thread full time to monitor
performance info like cache misses etc. from various CPUs and migrate tasks
as necessary. I don't think that kind of a blunt instrument approach would
be necessary, I don't think it would be that hard of a job. But the power
is there to do it if someone believed that was what needed to be done.
--
Douglas Siebert dsiebert@excisethis.khamsin.net
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety" -- Thomas Jefferson |
|
| Back to top |
|
|
Mark Smotherman
Guest
|
| Posted:
Sun Mar 06, 2005 1:34 am Post subject:
Re: Free advice to AMD |
|
|
Douglas Siebert <dsiebert@excisethis.khamsin.net> writes:
| Quote: | ... a complex + simple combination to support many more
threads up front, it might be a winner. ...
|
Such a combination also has dynamic power and temperature management
benefits. E.g.,
Lim, et al., "A Thermal-Aware Superscalar Microprocessor,"
ISQED, 2002
Kumar, et al., "Single-ISA Heterogeneous Multi-Core Architectures:
The Potential for Processor Power Reduction," Micro-36, 2003 |
|
| Back to top |
|
|
|
|
|
|
FAQ
Memberlist
Register
Profile
Log in to check your private messages
Log in