Prev: Will Barcelona surpass Core Duo, Woodcrest in CPU speed, or just multi-chip performance?
Next: Computer science conference ranking reference at www.conference-ranking.net
From: Nick Maclaren on 30 Jan 2007 09:36

Has anyone done any STREAMS, lmbench or similar testing of the
various Intel chipsets? A colleague would be interested to know
some real comparisons.

The Web seems very short of real timings, though full of hype.


Regards,
Nick Maclaren.
From: David Kanter on 31 Jan 2007 00:24
On Jan 30, 6:36 am, n...(a)cus.cam.ac.uk (Nick Maclaren) wrote:
> Has anyone done any STREAMS, lmbench or similar testing of the
> various Intel chipsets? A colleague would be interested to know
> some real comparisons.
>
> The Web seems very short of real timings, though full of hype.
>
> Regards,
> Nick Maclaren.

So you basically want to see the impact of the snoop filter, right?

DK

From: Nick Maclaren on 31 Jan 2007 05:30

In article <1170221064.326756.247750(a)h3g2000cwc.googlegroups.com>,
"David Kanter" <dkanter(a)gmail.com> writes:
|> On Jan 30, 6:36 am, n...(a)cus.cam.ac.uk (Nick Maclaren) wrote:
|> > Has anyone done any STREAMS, lmbench or similar testing of the
|> > various Intel chipsets? A colleague would be interested to know
|> > some real comparisons.
|> >
|> > The Web seems very short of real timings, though full of hype.
|>
|> So you basically want to see the impact of the snoop filter, right?

No.

That is ONE of the things that I should like to see, but Intel themselves
have said that there are other performance differences. The only memory
benchmarks I have seen show Blackford performing very poorly (6-7 GB/sec,
with either 2 or 4 cores in use on dual socket systems). Someone may
have run ones that are less dire, but I haven't seen them. I haven't
seen any post-production Greencreek ones.

Whether Blackford has a snoop filter or not is still very unclear; it
certainly has a bit to select its use, but I have seen no evidence that
the bit is actually connected to anything! As my colleague points out,
the Blackford data sheet contains 4 almost throw-away references to
"snoop filter" and the Greencreek one contains pages and pages of
detailed specifications. Go figure.


Regards,
Nick Maclaren.
From: David Kanter on 1 Feb 2007 15:49
On Jan 31, 2:30 am, n...(a)cus.cam.ac.uk (Nick Maclaren) wrote:
> In article <1170221064.326756.247...(a)h3g2000cwc.googlegroups.com>,"David Kanter" <dkan...(a)gmail.com> writes:
>
> |> On Jan 30, 6:36 am, n...(a)cus.cam.ac.uk (Nick Maclaren) wrote:
> |> > Has anyone done any STREAMS, lmbench or similar testing of the
> |> > various Intel chipsets? A colleague would be interested to know
> |> > some real comparisons.
> |> >
> |> > The Web seems very short of real timings, though full of hype.
> |>
> |> So you basically want to see the impact of the snoop filter, right?
>
> No.
>
> That is ONE of the things that I should like to see, but Intel themselves
> have said that there are other performance differences. The only memory
> benchmarks I have seen show Blackford performing very poorly (6-7 GB/sec,
> with either 2 or 4 cores in use on dual socket systems). Someone may
> have run ones that are less dire, but I haven't seen them. I haven't
> seen any post-production Greencreek ones.
>
> Whether Blackford has a snoop filter or not is still very unclear; it
> certainly has a bit to select its use, but I have seen no evidence that
> the bit is actually connected to anything! As my colleague points out,
> the Blackford data sheet contains 4 almost throw-away references to
> "snoop filter" and the Greencreek one contains pages and pages of
> detailed specifications. Go figure.

It's actually quite clear, a quick application of google for the
appropriate three key words will clear up the mystery entirely.

DK

From: Nick Maclaren on 1 Feb 2007 16:08

In article <1170362973.355761.135910(a)s48g2000cws.googlegroups.com>,
"David Kanter" <dkanter(a)gmail.com> writes:
|> >
|> > Whether Blackford has a snoop filter or not is still very unclear; it
|> > certainly has a bit to select its use, but I have seen no evidence that
|> > the bit is actually connected to anything! As my colleague points out,
|> > the Blackford data sheet contains 4 almost throw-away references to
|> > "snoop filter" and the Greencreek one contains pages and pages of
|> > detailed specifications. Go figure.
|>
|> It's actually quite clear, a quick application of google for the
|> appropriate three key words will clear up the mystery entirely.

You're trolling AGAIN! It isn't known as the Net of a Million Lies
for nothing, you know.

Why don't you (a) state unequivocally what YOU believe the current
situation is and (b) provide some references as to why you believe it?
Come on - don't be shy.


Regards,
Nick Maclaren.
 |  Next  |  Last
Pages: 1 2
Prev: Will Barcelona surpass Core Duo, Woodcrest in CPU speed, or just multi-chip performance?
Next: Computer science conference ranking reference at www.conference-ranking.net