Non-intel benchmarks on Conroe vs AMD's AM2 FX62

[The little lost angel]

http://www.hexus.net/content/item.php?item=5692&page=3

Well, it doesn't look like smoke and mirrors anymore does it?

The bit I'm curious about is, is there any mistake in the Sciencemark
2.0 latency results? Without an onboard memory controller, the Conroe
E6600 is faster than the FX-62. Strangely enough, the faster E6700 has
higher latencies than either of the former.

 

[Jan Panteltje]

http://www.hexus.net/content/item.php?item=5692&page=3

Well, it doesn't look like smoke and mirrors anymore does it?

The bit I'm curious about is, is there any mistake in the Sciencemark
2.0 latency results? Without an onboard memory controller, the Conroe
E6600 is faster than the FX-62. Strangely enough, the faster E6700 has
higher latencies than either of the former.

Lowest speed grade AMD versus not yet available highest speed grade iNtel?,
no details on setup or board, maybe Intel payed them?

 

[Robert Redelmeier]

http://www.hexus.net/content/item.php?item=5692&page=3
Well, it doesn't look like smoke and mirrors anymore does it?

No, it looks like a phanboi sight[sic]

The bit I'm curious about is, is there any mistake in the Sciencemark
2.0 latency results? Without an onboard memory controller, the Conroe
E6600 is faster than the FX-62. Strangely enough, the faster E6700
has higher latencies than either of the former.

Yes, these _are_ curious.

For one thing, I never measure anything nearly as fast as 44 ns,
not with my round-trip pgm below. For another, Intel boasts
their Northbridges (MCH) have intelligence and will figure out
memory access patterns, so maybe it figures out the Sciencemark
latency stride and does prefetch.

FWIW, my current results:

Latency CPU@MHz mem.ctl RAM
ns

88 k8@2000 NForce3 DDR400
144 P3@1000 laptop SO-PC133?
148 2*P3@860 Serverworks ??
178 P4@1800 i850 RDRAM
184 K7@1667 SiS735 PC133
185 P3@600 440BX PC100
217 2*Cel@500 440BX PC90
234 P2@350 440BX PC100?
288 P2@333 440BX PC66

I do need to find & test some more modern systems, but I'm
underwhelmed by the slowness of latency improvement.



compile: $ gcc -O2 lat10m.c
run: $ time ./a.out [multiply user time by 100 to give ns]

/* lat10m.c - Measure latency of 10 million fresh memory reads
(C) Copyright 2005 Robert Redelmeier - GPL v2.0 licence granted */
int p[ 1<<21 ] ;
main (void) {
int i, j ;
for ( i=0 ; i < 1<<21 ; i++ ) p = 0x1FFFFF & (i-5000) ;
for ( j=i=0 ; i < 9600000 ; i++ ) j = p[j] ;
return j ; }


-- Robert

 

[The little lost angel]

Lowest speed grade AMD versus not yet available highest speed grade iNtel?,
no details on setup or board, maybe Intel payed them?

Huh? The FX62 they tested against will be the highest speed grade AMD
process available, to be launched today...

 

[Jan Panteltje]

Huh? The FX62 they tested against will be the highest speed grade AMD
process available, to be launched today...

OK, my error, but I am biased, (towards AMD), so they give no data on the board,
maybe used slow memory, wrong settings....
Intel cannot be better

 

[Yousuf Khan]

http://www.hexus.net/content/item.php?item=5692&page=3

Well, it doesn't look like smoke and mirrors anymore does it?

Nope, certainly looks like Intel might have a real contender here.

The bit I'm curious about is, is there any mistake in the Sciencemark
2.0 latency results? Without an onboard memory controller, the Conroe
E6600 is faster than the FX-62. Strangely enough, the faster E6700 has
higher latencies than either of the former.

Well, that's likely as much a result of the shared L2 cache as much as
the bus latency. It's likely that the faster E6700 is scraping up
against its bus latency a bit more than the E6600 -- it gets things done
a bit faster, therefore it's going to memory a bit more often.

It does look like a shared L2 is a secret weapon here. Even AMD is
contemplating it for K8L -- possibly even a shared ZRAM cache.

Yousuf Khan

 

[Mark]

Lowest speed grade AMD versus not yet available highest speed grade
iNtel?,
no details on setup or board, maybe Intel payed them?

What a stupid, b&*&^*%^ comment. The FX-62 is the fastest AMD chip, whereas
the E6600 and E6700 are middle of the road conroe chips (eg, check out the
price for the FX-62 - $1500?? - versus $315 for the E6600 in July). Why try
and mislead people through just speaking rubbish??

 

[Mark]

http://www.hexus.net/content/item.php?item=5692&page=3

Well, it doesn't look like smoke and mirrors anymore does it?

The bit I'm curious about is, is there any mistake in the Sciencemark
2.0 latency results? Without an onboard memory controller, the Conroe
E6600 is faster than the FX-62. Strangely enough, the faster E6700 has
higher latencies than either of the former.

It's been obvious that it hasn't been smoke and mirrors for a while now, eg:

http://forumz.tomshardware.com/hard...Data-Core-Duo-Core-Extreme-ftopict183765.html

However, thanks for the new link. That Far Cry score at low resolution
(1024x768) is truly astonishing! 174.48 fps vs 125.1 fps for the FX-62!!
Can't wait to get my hands on one of these conroe chips!

 

[The little lost angel]

compile: $ gcc -O2 lat10m.c
run: $ time ./a.out [multiply user time by 100 to give ns]

/* lat10m.c - Measure latency of 10 million fresh memory reads
(C) Copyright 2005 Robert Redelmeier - GPL v2.0 licence granted */
int p[ 1<<21 ] ;
main (void) {
int i, j ;
for ( i=0 ; i < 1<<21 ; i++ ) p = 0x1FFFFF & (i-5000) ;
for ( j=i=0 ; i < 9600000 ; i++ ) j = p[j] ;
return j ; }

I tried this using gcc 4.0.2 20050808 on Ubuntu 5.1 (kernel
2.6.12-9-386), P4-M 1.7 (IBM T30). sorry anything newer and it's on my
main working machines, just not ready to risk linux on them yet :/

The results weren't consistent inside a terminal window, managed to
find a way to kill X and did it in a proper CLI after 20 minutes

first few runs 1.967s 1.957s 1.949s 1.942s
next bunch 2.025s 2.072s 2.020s 2.067s
last bunch 1.962s 2.020s 1.856s 1.857s

why isn't it consistent even in actual console or did I do something
wrong? :/

 

[Jan Panteltje]

However, thanks for the new link. That Far Cry score at low resolution
(1024x768) is truly astonishing! 174.48 fps vs 125.1 fps for the FX-62!!
Can't wait to get my hands on one of these conroe chips!

What you gona do with them, smash them?

 

[Mark]

What you gona do with them, smash them?

No, install it on a motherboard.

 

[Jan Panteltje]

No, install it on a motherboard.

Tell you what, if *I* was given the mobo, the conroe chips, and 1000$
to spend freely, I would refuse it, as I use no products from a company
that sues the hell out of each website that has 'inside' in its name.

And my experience with Intel and specs and rectification sheets is not
that good either.

 

[Robert Redelmeier]

compile: $ gcc -O2 lat10m.c
run: $ time ./a.out [multiply user time by 100 to give ns]

/* lat10m.c - Measure latency of 10 million fresh memory reads
(C) Copyright 2005 Robert Redelmeier - GPL v2.0 licence granted */
int p[ 1<<21 ] ;
main (void) {
int i, j ;
for ( i=0 ; i < 1<<21 ; i++ ) p = 0x1FFFFF & (i-5000) ;
for ( j=i=0 ; i < 9600000 ; i++ ) j = p[j] ;
return j ; }

I tried this using gcc 4.0.2 20050808 on Ubuntu 5.1 (kernel
2.6.12-9-386), P4-M 1.7 (IBM T30). sorry anything newer and it's on my
main working machines, just not ready to risk linux on them yet :/

Try a LiveCD Linux distro. Nothing is written to the HD,
and it boots from CD. I thought ubuntu did this.

The results weren't consistent inside a terminal window, managed to
find a way to kill X and did it in a proper CLI after 20 minutes

first few runs 1.967s 1.957s 1.949s 1.942s
next bunch 2.025s 2.072s 2.020s 2.067s
last bunch 1.962s 2.020s 1.856s 1.857s

why isn't it consistent even in actual console or did I do
something wrong? :/

No, you did it right. I would call these reasonably consistant,
195 ns +/-5%. since the system has overhead tasks and might
need to scrounge up RAM. To be more accurate, I should use RDTSC
after the pages have been filled.

-- Robert

 

[Mark]

Tell you what, if *I* was given the mobo, the conroe chips, and 1000$
to spend freely, I would refuse it, as I use no products from a company
that sues the hell out of each website that has 'inside' in its name.

Rather than refuse it, I would prefer if you would give them to me.

 

[Jan Panteltje]

compile: $ gcc -O2 lat10m.c
run: $ time ./a.out [multiply user time by 100 to give ns]

/* lat10m.c - Measure latency of 10 million fresh memory reads
(C) Copyright 2005 Robert Redelmeier - GPL v2.0 licence granted */
int p[ 1<<21 ] ;
main (void) {
int i, j ;
for ( i=0 ; i < 1<<21 ; i++ ) p = 0x1FFFFF & (i-5000) ;
for ( j=i=0 ; i < 9600000 ; i++ ) j = p[j] ;
return j ; }

I tried this using gcc 4.0.2 20050808 on Ubuntu 5.1 (kernel
2.6.12-9-386), P4-M 1.7 (IBM T30). sorry anything newer and it's on my
main working machines, just not ready to risk linux on them yet :/

Try a LiveCD Linux distro. Nothing is written to the HD,
and it boots from CD. I thought ubuntu did this.

The results weren't consistent inside a terminal window, managed to
find a way to kill X and did it in a proper CLI after 20 minutes

first few runs 1.967s 1.957s 1.949s 1.942s
next bunch 2.025s 2.072s 2.020s 2.067s
last bunch 1.962s 2.020s 1.856s 1.857s

why isn't it consistent even in actual console or did I do
something wrong? :/

No, you did it right. I would call these reasonably consistant,
195 ns +/-5%. since the system has overhead tasks and might
need to scrounge up RAM. To be more accurate, I should use RDTSC
after the pages have been filled.

Run that as:
nice -n -19 .....
to get max prioriy?

 

[Jan Panteltje]

On a sunny day (Tue, 23 May 2006 10:51:28 -0700) it happened "Mark"

Tell you what, if *I* was given the mobo, the conroe chips, and 1000$

Rather than refuse it, I would prefer if you would give them to me.

Sorry have not got it yet

 

[Robert Redelmeier]

Run that as:
nice -n -19 .....
to get max prioriy?

Only if the system is non-idle. Even then, the Linux
scheduler is very "fair" and a CPU hog lit lat10m
will get downrated.

-- Robert

 

[The little lost angel]

Try a LiveCD Linux distro. Nothing is written to the HD,
and it boots from CD. I thought ubuntu did this.

I got the Installer version. I'll try it with a LiveCD version when it
finish downloading.

<CTL><ALT><BS> takes 20 minutes to do?

finding it was the problem. I remember vaguely there was a menu
shortcut to shutdown into CLI mode which I used previously (i think I
misremembered and that method was only available on the previous
Gentoo X on the laptop).

So wasted about 5 minutes looking through all the various menus but
couldn't find it. Tried to see if I could shutdown into terminal and
that took a while, didn't work. So finally went googling for it, read
a few before I found a page that described the actual short cut.
Although the same page claims that it's an ungraceful way to exit and
most wm/desktops provide a better way... the problem is finding it I
guess!

No, you did it right. I would call these reasonably consistant,
195 ns +/-5%. since the system has overhead tasks and might
need to scrounge up RAM. To be more accurate, I should use RDTSC
after the pages have been filled.

How's that to be done? Sorry, I'm practically a noob with C and last
touched any kind of assembly before I first came into the NG :/

 

[Jan Panteltje]

Only if the system is non-idle.

The systemm, in Linux, is nion-idle very often,
you will note a zillion modules and deamons running.
ps avx
lsmod
So the chance that some process runs during one of
the test runs is not zero, making the results different.

Even then, the Linux
scheduler is very "fair" and a CPU hog lit lat10m
will get downrated.

Not on my system!
Things will get VERY slow if you type
nice -n -19 yes
in an xterm

without 'yes'
hdparm -T /dev/hdc
/dev/hdc:
Timing cached reads: 740 MB in 2.00 seconds = 369.98 MB/sec


with 'yes'
hdparm -T /dev/hdc
/dev/hdc:
Timing cached reads: 416 MB in 2.05 seconds = 203.11 MB/sec

-T is cache read.

But benchmarks.... I'd rather run some applictations, run my ffmpeg
h264 encoding (it runs 24/7 on this box for video camera).
DivX is old stuff.

 

[Jan Panteltje]

On a sunny day (Tue, 23 May 2006 19:12:26 GMT) it happened
[email protected] (The little lost angel) wrote in
<[email protected]>:

There is a nice, mainly command line oriented (but it has X) Linux
CD image you can download www.grml.org (it is running my system).
And that one you can also install to hard disk.

I use this because the rest seems to get or pick up bloat.