* [Fwd: linus.linux.sgi.com]
@ 1999-01-19 23:28 ` dmanddmer
0 siblings, 0 replies; 22+ messages in thread
From: dmanddmer @ 1999-01-19 23:28 UTC (permalink / raw)
To: linux
[-- Attachment #1: Type: text/plain, Size: 435 bytes --]
Btw, I heard a rumor that the memory in the VPC is custom or unique
memory. Is this true? How much of the VPC, besides the obvious, is not
industry standard? If SGI is attempting solely to require customers to
be able to buy any upgrades ONLY from SGI, then SGI has shot itself in
the foot again. It did not work for Apple, it didn't work for Unisys,
it didn't work for Intergraph. In short, it hasn't worked for anyone.
David M.
[-- Attachment #2: Type: message/rfc822, Size: 3242 bytes --]
From: david@carrera.columbus.sgi.com (David Watters)
To: dmanddmer <dmerchan@hiwaay.net>
Cc: linux@cthulhu.engr.sgi.com
Subject: Re: linus.linux.sgi.com
Date: Tue, 19 Jan 1999 10:30:47 -0500 (EST)
Message-ID: <9901191030.ZM37783@carrera.columbus.sgi.com>
On Jan 19, 5:21am, dmanddmer wrote:
> Subject: Re: linus.linux.sgi.com
> Personally, since SGI made the questionable decision to port to NT, I
> wish they would commit to AMD's processors instead of Intels.
SGI couldn't. SGI had to replace so much of the IA/440xx chipset that there
had to be a prettty big agreement with Intel covering a lot of issues,
machines, processors, and markets to allow SGI to, for example, implement
the frontside bus in their own chipset. Don't forget all those Co'op
marketing dollars too. -dong ^dong -dong ^^dong
> Porting to NT alone is probably not enough to save the company. It
> didn't work for Intergraph.
Yeah, but Intergraph is lame so they had that working against them too. :)
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 1:29 ` Chad Carlin
0 siblings, 0 replies; 22+ messages in thread
From: Chad Carlin @ 1999-01-20 1:29 UTC (permalink / raw)
To: dmanddmer; +Cc: linux
[-- Attachment #1: Type: TEXT/PLAIN, Size: 811 bytes --]
The current mainstream memory offering is 70n/s and we are at 50n/s. The memory is not proprietary - any manufacturer could buy it if their system could drive it fast enough!
Chad Carlin
Silicon Graphics Inc.
chad@sgi.com
http://reality.sgi.com/chad
On Tue, 19 Jan 1999, dmanddmer wrote:
->Btw, I heard a rumor that the memory in the VPC is custom or unique
->memory. Is this true? How much of the VPC, besides the obvious, is not
->industry standard? If SGI is attempting solely to require customers to
->be able to buy any upgrades ONLY from SGI, then SGI has shot itself in
->the foot again. It did not work for Apple, it didn't work for Unisys,
->it didn't work for Intergraph. In short, it hasn't worked for anyone.
->
->David M.
[-- Attachment #2: Type: MESSAGE/RFC822, Size: 3250 bytes --]
From: david@carrera.columbus.sgi.com (David Watters)
To: dmanddmer <dmerchan@hiwaay.net>
Cc: linux@cthulhu.engr.sgi.com
Subject: Re: linus.linux.sgi.com
Date: Tue, 19 Jan 1999 10:30:47 -0500 (EST)
Message-ID: <9901191030.ZM37783@carrera.columbus.sgi.com>
On Jan 19, 5:21am, dmanddmer wrote:
> Subject: Re: linus.linux.sgi.com
> Personally, since SGI made the questionable decision to port to NT, I
> wish they would commit to AMD's processors instead of Intels.
SGI couldn't. SGI had to replace so much of the IA/440xx chipset that there
had to be a prettty big agreement with Intel covering a lot of issues,
machines, processors, and markets to allow SGI to, for example, implement
the frontside bus in their own chipset. Don't forget all those Co'op
marketing dollars too. -dong ^dong -dong ^^dong
> Porting to NT alone is probably not enough to save the company. It
> didn't work for Intergraph.
Yeah, but Intergraph is lame so they had that working against them too. :)
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 1:29 ` Chad Carlin
0 siblings, 0 replies; 22+ messages in thread
From: Chad Carlin @ 1999-01-20 1:29 UTC (permalink / raw)
To: dmanddmer; +Cc: linux
[-- Attachment #1: Type: TEXT/PLAIN, Size: 811 bytes --]
The current mainstream memory offering is 70n/s and we are at 50n/s. The memory is not proprietary - any manufacturer could buy it if their system could drive it fast enough!
Chad Carlin
Silicon Graphics Inc.
chad@sgi.com
http://reality.sgi.com/chad
On Tue, 19 Jan 1999, dmanddmer wrote:
->Btw, I heard a rumor that the memory in the VPC is custom or unique
->memory. Is this true? How much of the VPC, besides the obvious, is not
->industry standard? If SGI is attempting solely to require customers to
->be able to buy any upgrades ONLY from SGI, then SGI has shot itself in
->the foot again. It did not work for Apple, it didn't work for Unisys,
->it didn't work for Intergraph. In short, it hasn't worked for anyone.
->
->David M.
[-- Attachment #2: Type: MESSAGE/RFC822, Size: 3242 bytes --]
From: david@carrera.columbus.sgi.com (David Watters)
To: dmanddmer <dmerchan@hiwaay.net>
Cc: linux@cthulhu.engr.sgi.com
Subject: Re: linus.linux.sgi.com
Date: Tue, 19 Jan 1999 10:30:47 -0500 (EST)
Message-ID: <9901191030.ZM37783@carrera.columbus.sgi.com>
On Jan 19, 5:21am, dmanddmer wrote:
> Subject: Re: linus.linux.sgi.com
> Personally, since SGI made the questionable decision to port to NT, I
> wish they would commit to AMD's processors instead of Intels.
SGI couldn't. SGI had to replace so much of the IA/440xx chipset that there
had to be a prettty big agreement with Intel covering a lot of issues,
machines, processors, and markets to allow SGI to, for example, implement
the frontside bus in their own chipset. Don't forget all those Co'op
marketing dollars too. -dong ^dong -dong ^^dong
> Porting to NT alone is probably not enough to save the company. It
> didn't work for Intergraph.
Yeah, but Intergraph is lame so they had that working against them too. :)
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread* Re: [Fwd: linus.linux.sgi.com]
1999-01-20 1:29 ` Chad Carlin
(?)
@ 1999-01-20 2:04 ` job bogan
[not found] ` <job@piquin.uchicago.edu>
` (2 more replies)
-1 siblings, 3 replies; 22+ messages in thread
From: job bogan @ 1999-01-20 2:04 UTC (permalink / raw)
To: Chad Carlin; +Cc: linux, nldesai, ellidz
Your message dated: Tue, 19 Jan 1999 19:29:16 CST
>The current mainstream memory offering is 70n/s and we are at 50n/s. The memo
>ry is not proprietary - any manufacturer could buy it if their system could dr
>ive it fast enough!
pardon me but...
Cut the SGI - "We Are Better Than God, Bow Before Our I/O" crap.
Lots of PCs use 10ns SDRAM and 6ns SDRAM all of the time. And there is
plenty of memory saturation happening on those systems. Trust me. We
do it all the time w/ scientific matrix code.
Humm, a look at the glossies for the 320 and 540 make it look like all
of SGI is working off a typo. It lists:
* 100 MHz (50 ns) ECC synchronous dynamic RAM (SDRAM)
that should read 5ns. and beyond that, lots of people sell 100Mhz
capable SDRAM. (eg http://www.memoryx.com/generic.htm)
Anyhow, the only things that seem different from the std garage built PC
are:
Video expandable to 1/2gb of ram (though this looks like it's running
off the system ram, not dedicated ram. not a bad thing, but it
kills the system bus. and it's a bit misleading)
2 64-bit PCI busses
540 uses a WTX tower case. (320 uses ATX)
540 has SCA scsi disk bays. I wonder if i'll need $700 drive sleds...
Video Capture card (S-Video, RCA. in/out)
USB Keyboard & Mouse (this currently limits linux support - USB is
alpha'ish)
video connector for the SGI LCD Display
2 firewire ports. not hard to do, just pointless for the next 6months.
I might buy some if they are competitive w/ plain old white Linux boxes.
Esp. if they have good OpenGL performance under Linux. They would be
more interesting if they had a craylink type high speed connector that
could be used ala myrinet for making Beowulf clusters.
job
ps - 540 has 4 DIMM slots, the 320 2 slots.
--
John Bogan
Director of Computing 773-702-2588
James Franck Institute 5640 South Ellis Ave
University of Chicago Chicago, Il 60637
^ permalink raw reply [flat|nested] 22+ messages in thread[parent not found: <job@piquin.uchicago.edu>]
* Re: [Fwd: linus.linux.sgi.com]
[not found] ` <job@piquin.uchicago.edu>
@ 1999-01-20 2:28 ` Richard Offer
1999-01-20 2:36 ` Ken Klingman
` (2 more replies)
0 siblings, 3 replies; 22+ messages in thread
From: Richard Offer @ 1999-01-20 2:28 UTC (permalink / raw)
To: linux
* $ from job@piquin.uchicago.edu at "19-Jan: 8:04pm" | sed "1,$s/^/* /"
*
*
* Your message dated: Tue, 19 Jan 1999 19:29:16 CST
*
* >The current mainstream memory offering is 70n/s and we are at 50n/s. The
memo
* >ry is not proprietary - any manufacturer could buy it if their system could
dr
* >ive it fast enough!
*
* pardon me but...
* Cut the SGI - "We Are Better Than God, Bow Before Our I/O" crap.
*
* Lots of PCs use 10ns SDRAM and 6ns SDRAM all of the time.
Well, I couldn't see anything faster than 60ns on Kingstons store.
http://www.ec.kingston.com/ecom/kepler/specs.asp
*
* Humm, a look at the glossies for the 320 and 540 make it look like all
* of SGI is working off a typo. It lists:
* * 100 MHz (50 ns) ECC synchronous dynamic RAM (SDRAM)
* that should read 5ns. and beyond that, lots of people sell 100Mhz
* capable SDRAM. (eg http://www.memoryx.com/generic.htm)
I'd guess that their 10ns is a misprint as its cheap and listed in the PC-66
section.
Isn't 100Mhz and 50ns a measure of two different things ? This is meant as a
serious question, as I don't know. But, I'd guess that one is the refresh and
one the time to access ?
*
* I might buy some if they are competitive w/ plain old white Linux boxes.
How do you measure 'competitive' ? For me, they aren't; for my next-cubicle
neighbour they would be. Luckily we don't live in a one size
(hardware/software) fits all world.
* job
*
richard.
_______________________________________________________________________
A guest signature by Dan Boyd (boyd at cs.buffalo.edu)
A competent sysadmin can turn any task into one resembling system
administration and requiring a Makefile.
^ permalink raw reply [flat|nested] 22+ messages in thread* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 2:36 ` Ken Klingman
0 siblings, 0 replies; 22+ messages in thread
From: Ken Klingman @ 1999-01-20 2:36 UTC (permalink / raw)
To: Richard Offer; +Cc: linux
The VW 320 and 540 use 288-bit wide synchronous DRAM, clocked
at 100MHz. 288 bits means it's not your everyday PC SDRAM.
That's 256 bits for data and 32 for ECC. 256bits = 32 bytes
at 100MHz = 3.2GB/sec theoretically possible between the
memory system and Cobalt, the memory controller/graphics
ASIC.
Ken
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 2:36 ` Ken Klingman
0 siblings, 0 replies; 22+ messages in thread
From: Ken Klingman @ 1999-01-20 2:36 UTC (permalink / raw)
To: Richard Offer; +Cc: linux
The VW 320 and 540 use 288-bit wide synchronous DRAM, clocked
at 100MHz. 288 bits means it's not your everyday PC SDRAM.
That's 256 bits for data and 32 for ECC. 256bits = 32 bytes
at 100MHz = 3.2GB/sec theoretically possible between the
memory system and Cobalt, the memory controller/graphics
ASIC.
Ken
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 21:18 ` David Watters
0 siblings, 0 replies; 22+ messages in thread
From: David Watters @ 1999-01-20 21:18 UTC (permalink / raw)
To: Ken Klingman, Richard Offer; +Cc: linux
On Jan 19, 6:36pm, Ken Klingman wrote:
> Subject: Re: [Fwd: linus.linux.sgi.com]
>
> The VW 320 and 540 use 288-bit wide synchronous DRAM, clocked
> at 100MHz. 288 bits means it's not your everyday PC SDRAM.
But it is!
It is just laid out differently.
Industry standard SDRAM chips but to get the wider memory path
they layed them out differently.
They used a different JDEC standard DIMM (the same as in some notebooks)
because the motherboard would have been a mess using the giant, more
commond desktop-PC JDEC standard modules.
The idea is a high-bandwith arch. at PC prices and to do that you have
to make some tough decisions. Otherwise the OCTANE is a far better (and $)
arch. that is just dying for faster processors and next-gen graphics.
Who cares. It is all in the application performance. We will soon see...
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 21:18 ` David Watters
0 siblings, 0 replies; 22+ messages in thread
From: David Watters @ 1999-01-20 21:18 UTC (permalink / raw)
To: Ken Klingman, Richard Offer; +Cc: linux
On Jan 19, 6:36pm, Ken Klingman wrote:
> Subject: Re: [Fwd: linus.linux.sgi.com]
>
> The VW 320 and 540 use 288-bit wide synchronous DRAM, clocked
> at 100MHz. 288 bits means it's not your everyday PC SDRAM.
But it is!
It is just laid out differently.
Industry standard SDRAM chips but to get the wider memory path
they layed them out differently.
They used a different JDEC standard DIMM (the same as in some notebooks)
because the motherboard would have been a mess using the giant, more
commond desktop-PC JDEC standard modules.
The idea is a high-bandwith arch. at PC prices and to do that you have
to make some tough decisions. Otherwise the OCTANE is a far better (and $)
arch. that is just dying for faster processors and next-gen graphics.
Who cares. It is all in the application performance. We will soon see...
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 3:01 ` job bogan
0 siblings, 0 replies; 22+ messages in thread
From: job bogan @ 1999-01-20 3:01 UTC (permalink / raw)
To: Richard Offer; +Cc: linux, ellidz
Your message dated: Tue, 19 Jan 1999 18:28:50 PST
>Well, I couldn't see anything faster than 60ns on Kingstons store.
>http://www.ec.kingston.com/ecom/kepler/specs.asp
100MHz is the bus speed. to meet that, you need ~5-6ns access times.
Kingston does list 100MHz memory.
>I'd guess that their 10ns is a misprint as its cheap and listed in the PC-66
>section.
look in the PC100 section. that's the Lintel world acronym for 100MHz
capable ram. And, yes, it's cheap. That's the point. =) If you are
using ECC, the worst i'll see is a hung machine. and judging by
previous SGI prices, i can buy twice as much 3rd party ram as SGI ram.
>Isn't 100Mhz and 50ns a measure of two different things ? This is meant as a
>serious question, as I don't know. But, I'd guess that one is the refresh and
>one the time to access ?
nope. see the stuff above. If you want more info, bug me off the list.
>* I might buy some if they are competitive w/ plain old white Linux boxes.
>
>How do you measure 'competitive' ? For me, they aren't; for my next-cubicle
>neighbour they would be. Luckily we don't live in a one size
>(hardware/software) fits all world.
I know, it's a tough question.
Competative, in my realm, is floating point performance based.
Specificly, Big (100's of MB) matrix manipulations. Lately, generic PCs
and sun E450's have won our $. Both well outperform SGIs for the
cost. (both O2000 or O200 based. we have some of each around...)
NT as the only OS for the SGI 540/320 means they are not competitive.
We cannot dedicate, and waste, that much CPU to one person's desktop.
If NT had a good remote user (ssh, telnet, hell a remote queuing system)
interface, we might look at them. The lack of true multi-user and
non-console interfaces in NT just takes these SGIs off the chart until
linux runs on them.
And once it's there, i see no reason to buy them unless i can get the
Video Performance out of them. (unless, perhaps, they clean up on raw
cpu due to a cleaned up system bus.)
i've never been the world's biggest SGI/Irix fan. But - for my env., and
most of the Academic world, i just don't see these as viable machines
until Linux or Irix runs on them. egh... you may be able to get the
CAD/CAM market back, but i don't see us buying more SGIs soon.
Anyhow, this is offtopic now.
job
--
John Bogan
Director of Computing 773-702-2588
James Franck Institute 5640 South Ellis Ave
University of Chicago Chicago, Il 60637
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 3:01 ` job bogan
0 siblings, 0 replies; 22+ messages in thread
From: job bogan @ 1999-01-20 3:01 UTC (permalink / raw)
To: Richard Offer; +Cc: linux, ellidz
Your message dated: Tue, 19 Jan 1999 18:28:50 PST
>Well, I couldn't see anything faster than 60ns on Kingstons store.
>http://www.ec.kingston.com/ecom/kepler/specs.asp
100MHz is the bus speed. to meet that, you need ~5-6ns access times.
Kingston does list 100MHz memory.
>I'd guess that their 10ns is a misprint as its cheap and listed in the PC-66
>section.
look in the PC100 section. that's the Lintel world acronym for 100MHz
capable ram. And, yes, it's cheap. That's the point. =) If you are
using ECC, the worst i'll see is a hung machine. and judging by
previous SGI prices, i can buy twice as much 3rd party ram as SGI ram.
>Isn't 100Mhz and 50ns a measure of two different things ? This is meant as a
>serious question, as I don't know. But, I'd guess that one is the refresh and
>one the time to access ?
nope. see the stuff above. If you want more info, bug me off the list.
>* I might buy some if they are competitive w/ plain old white Linux boxes.
>
>How do you measure 'competitive' ? For me, they aren't; for my next-cubicle
>neighbour they would be. Luckily we don't live in a one size
>(hardware/software) fits all world.
I know, it's a tough question.
Competative, in my realm, is floating point performance based.
Specificly, Big (100's of MB) matrix manipulations. Lately, generic PCs
and sun E450's have won our $. Both well outperform SGIs for the
cost. (both O2000 or O200 based. we have some of each around...)
NT as the only OS for the SGI 540/320 means they are not competitive.
We cannot dedicate, and waste, that much CPU to one person's desktop.
If NT had a good remote user (ssh, telnet, hell a remote queuing system)
interface, we might look at them. The lack of true multi-user and
non-console interfaces in NT just takes these SGIs off the chart until
linux runs on them.
And once it's there, i see no reason to buy them unless i can get the
Video Performance out of them. (unless, perhaps, they clean up on raw
cpu due to a cleaned up system bus.)
i've never been the world's biggest SGI/Irix fan. But - for my env., and
most of the Academic world, i just don't see these as viable machines
until Linux or Irix runs on them. egh... you may be able to get the
CAD/CAM market back, but i don't see us buying more SGIs soon.
Anyhow, this is offtopic now.
job
--
John Bogan
Director of Computing 773-702-2588
James Franck Institute 5640 South Ellis Ave
University of Chicago Chicago, Il 60637
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 21:29 ` David Watters
0 siblings, 0 replies; 22+ messages in thread
From: David Watters @ 1999-01-20 21:29 UTC (permalink / raw)
To: job bogan, Richard Offer; +Cc: linux, ellidz
On Jan 19, 9:01pm, job bogan wrote:
> judging by
> previous SGI prices, i can buy twice as much 3rd party ram as SGI ram.
You will be glad to know that the memory will be available from 3rd parties.
> interface, we might look at them. The lack of true multi-user and
> non-console interfaces in NT just takes these SGIs off the chart
Some of what you need may be bundled. Take a look at the interoperability
kit that is bundled which includes a lot of what you mentioned.
> And once it's there, i see no reason to buy them unless i can get the
> Video Performance out of them. (unless, perhaps, they clean up on raw
> cpu due to a cleaned up system bus.)
What system bus? The FSB is the same, otherwise the off the shelf CPUs
(which I assume you will by 3rd party :) ) won't work. Our memory
controller is lower latency than the 440xx and has huge bandwidth from
memory to the controller. Both of these (mostly the latency) will show
on FP performance when you break cache.
> i've never been the world's biggest SGI/Irix fan.
Really? :)
> But - for my env., and
> most of the Academic world, i just don't see these as viable machines
> until Linux or Irix runs on them. egh... you may be able to get the
> CAD/CAM market back, but i don't see us buying more SGIs soon.
The small little CAD/CAM market. :)
If we get that back then we will build all the Linux/IRIX boxes you
need.
Thanks for your perspective though.
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 21:29 ` David Watters
0 siblings, 0 replies; 22+ messages in thread
From: David Watters @ 1999-01-20 21:29 UTC (permalink / raw)
To: job bogan, Richard Offer; +Cc: linux, ellidz
On Jan 19, 9:01pm, job bogan wrote:
> judging by
> previous SGI prices, i can buy twice as much 3rd party ram as SGI ram.
You will be glad to know that the memory will be available from 3rd parties.
> interface, we might look at them. The lack of true multi-user and
> non-console interfaces in NT just takes these SGIs off the chart
Some of what you need may be bundled. Take a look at the interoperability
kit that is bundled which includes a lot of what you mentioned.
> And once it's there, i see no reason to buy them unless i can get the
> Video Performance out of them. (unless, perhaps, they clean up on raw
> cpu due to a cleaned up system bus.)
What system bus? The FSB is the same, otherwise the off the shelf CPUs
(which I assume you will by 3rd party :) ) won't work. Our memory
controller is lower latency than the 440xx and has huge bandwidth from
memory to the controller. Both of these (mostly the latency) will show
on FP performance when you break cache.
> i've never been the world's biggest SGI/Irix fan.
Really? :)
> But - for my env., and
> most of the Academic world, i just don't see these as viable machines
> until Linux or Irix runs on them. egh... you may be able to get the
> CAD/CAM market back, but i don't see us buying more SGIs soon.
The small little CAD/CAM market. :)
If we get that back then we will build all the Linux/IRIX boxes you
need.
Thanks for your perspective though.
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 3:08 ` Greg Chesson
0 siblings, 0 replies; 22+ messages in thread
From: Greg Chesson @ 1999-01-20 3:08 UTC (permalink / raw)
To: Richard Offer, linux
This is a case of rocket brains getting confused by misleading specs
in the commodity world: a 50ns dram or sdram is very fast - that refers
to the cycle time of the internal dram core whereas a PCI-100 compliant
memory must operate on a 100 Mhz (10ns) bus and needs a latch time
better than 10ns.
g
--
Greg Chesson
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
@ 1999-01-20 3:08 ` Greg Chesson
0 siblings, 0 replies; 22+ messages in thread
From: Greg Chesson @ 1999-01-20 3:08 UTC (permalink / raw)
To: Richard Offer, linux
This is a case of rocket brains getting confused by misleading specs
in the commodity world: a 50ns dram or sdram is very fast - that refers
to the cycle time of the internal dram core whereas a PCI-100 compliant
memory must operate on a 100 Mhz (10ns) bus and needs a latch time
better than 10ns.
g
--
Greg Chesson
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
1999-01-20 2:04 ` job bogan
[not found] ` <job@piquin.uchicago.edu>
@ 1999-01-20 3:32 ` Chad Carlin
1999-01-20 16:02 ` Eric Kimminau
2 siblings, 0 replies; 22+ messages in thread
From: Chad Carlin @ 1999-01-20 3:32 UTC (permalink / raw)
To: job bogan; +Cc: linux
[-- Attachment #1: Type: TEXT/PLAIN, Size: 4097 bytes --]
->Lots of PCs use 10ns SDRAM and 6ns SDRAM all of the time. And there is
A->plenty of memory saturation happening on those systems. Trust me. We
->do it all the time w/ scientific matrix code.
I appologize for the way I came accross. I meant no offense. Nor did I mean to suggest that your applications were unable to tax the memory subsystem of a variety of manufacturers PCs. To be candid, I have spent very little time with specs for PC memory.
I also believed that I said "mainstream" not record setting.
->pardon me but...
->Cut the SGI - "We Are Better Than God, Bow Before Our I/O" crap.
I would never suggest that you bow before our IO either. Some people need it, some don't. Not really my place to make that determination for you. Though, you have to admit that we do a pretty good job for those apps that really need it.
->Humm, a look at the glossies for the 320 and 540 make it look like all
->of SGI is working off a typo. It lists:
-> * 100 MHz (50 ns) ECC synchronous dynamic RAM (SDRAM)
->that should read 5ns. and beyond that, lots of people sell 100Mhz
->capable SDRAM. (eg http://www.memoryx.com/generic.htm)
That's the problem with the internet, no real way to verify your sources. That email could have come from a temp janitor on his last day. I really don't know any of the specifics or part numbers. That would be something to ask your sales rep.
->Anyhow, the only things that seem different from the std garage built PC
->are:
->Video expandable to 1/2gb of ram (though this looks like it's running
-> off the system ram, not dedicated ram. not a bad thing, but it
-> kills the system bus. and it's a bit misleading)
Much cheaper that raster managers. As more information comes out about the architecture of the new systems, I'll think you might change your mind. I've seen the some of the overviews and it looks impressive. Keep an eye out for something about io being gu
aranteed in the hardware.
->2 64-bit PCI busses
->540 uses a WTX tower case. (320 uses ATX)
->540 has SCA scsi disk bays. I wonder if i'll need $700 drive sleds...
->Video Capture card (S-Video, RCA. in/out)
->USB Keyboard & Mouse (this currently limits linux support - USB is
-> alpha'ish)
Linux problem maybe. Late beta'ish maybe. My mom's iMac is working like a champ though. USB hub and all.
->video connector for the SGI LCD Display
->2 firewire ports. not hard to do, just pointless for the next 6months.
Imagine how mad you'd be in 6 months if we didn't include this originally but were preparing to sell it to you as an upgrade.
->I might buy some if they are competitive w/ plain old white Linux boxes.
->Esp. if they have good OpenGL performance under Linux.
Haven't gotten to play with hardware accellerated OpenGl on many "plain old white Linux boxes". How do they do? We'd be happy to forward the data to marketing.
->They would be
->more interesting if they had a craylink type high speed connector that
->could be used ala myrinet for making Beowulf clusters.
Did you know we're are already getting criticized for distributing a CD, with NT (service pak4) custom burned to include our GFX drivers. They are saying that we are being too proprietary. Can you imagine the beating we would take for custom building non-
industry standard PC hardware (that would probably need special RAM). There are choices to be made about what to make custom and what to make generic. I prefer custom but the number of folks that need it enough to pay for it, is not growing as fast as the
other segments of the computing markets.
Actually, there is a "craylink type high speed connector" coming in the near future. It's called GSN, Giga-Byte System Network. It is not slated to be part of the desktop family though. Once again, I think this technology, while useful by some, would be l
argely ignored by the desktop system buying public as being exotic or proprietary.
Once again, many appologies.
Chad
->Director of Computing 773-702-2588
->James Franck Institute 5640 South Ellis Ave
->University of Chicago Chicago, Il 60637
[-- Attachment #2: Type: APPLICATION/octet-stream, Size: 0 bytes --]
^ permalink raw reply [flat|nested] 22+ messages in thread* Re: [Fwd: linus.linux.sgi.com]
1999-01-20 2:04 ` job bogan
[not found] ` <job@piquin.uchicago.edu>
1999-01-20 3:32 ` Chad Carlin
@ 1999-01-20 16:02 ` Eric Kimminau
2 siblings, 0 replies; 22+ messages in thread
From: Eric Kimminau @ 1999-01-20 16:02 UTC (permalink / raw)
To: job bogan; +Cc: Chad Carlin, linux, nldesai, ellidz
job bogan wrote:
> ps - 540 has 4 DIMM slots, the 320 2 slots.
pssst. BANKS not SLOTS.
--
---------1---------2---------3---------4---------5---------6---------7
Eric Kimminau eak@detroit.sgi.com Field Technical Support
SGI 39001 W. 12 Mile Rd. Farmington, MI 48331-2903
Vox:(248)848-4455 Fax:(248)848-5600 VNET:6-327-4455
"I speak my mind and no one else's."
"I am a bomb technician. If you see me running, try to keep up..."
http://support.sgi.com
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
1999-01-19 23:28 ` dmanddmer
(?)
(?)
@ 1999-01-20 15:18 ` Eric Kimminau
1999-01-20 21:55 ` David Watters
1999-01-21 7:11 ` David S. Miller
-1 siblings, 2 replies; 22+ messages in thread
From: Eric Kimminau @ 1999-01-20 15:18 UTC (permalink / raw)
To: dmanddmer; +Cc: linux
Warning! Danger!
FUD. FUD. FUD.
dmanddmer wrote:
>
> Btw, I heard a rumor that the memory in the VPC is custom or unique
> memory. Is this true? How much of the VPC, besides the obvious, is not
> industry standard? If SGI is attempting solely to require customers to
> be able to buy any upgrades ONLY from SGI, then SGI has shot itself in
> the foot again. It did not work for Apple, it didn't work for Unisys,
> it didn't work for Intergraph. In short, it hasn't worked for anyone.
>
> David M.
It is not a rumor. It is fact. The ram being used in the VPC is like the
memory in no other PC (that I know of) being manufactured today.
You know that SGI has pioneered several mainstream simm/dimm technologies? In
fact we hold the patents on a couple I believe. I'll leave it as an excercise
to the curious to find out which ones. When you see the size of the 40 pin
dimms we use in the VPC and then realize the density they are capable of in
less than half the space of currently available dimms, perhaps you will
reconsider your position.
Every time SGI introduces new ram technology, very shortly Kingston or some
other chip vendor starts making them and they are always cheaper than what you
can get from SGI directly. Then they start appearing in all the "standard"
generic clones.
Every new platform we have announced since the Indy has used ram technology
that noone else was really using. We use the best technology to gain the
highest levels of performance. You can buy Octane and O2 ram at WalMart now,
for chrissakes.
A quick search for "SGI O2 memory" on any search engine will show you a
hundred or so hits. VPC will take a little bit, but you can be sure there will
be 3rd party memory sources.
How much of the VPC is standard today? large portions. dual PCI bus (get that
from a clone) (and what does it cost to get it on a Sun?), the CPU's, serial
port, fire wire interfaces, parallel port, video in and out, digital audio
with all the standard ports, etc. The important parts, the parts that blow the
doors off of every other same-ol, same-ol generic clone PC isn't industry
standard, YET.
How about you take this elsewhere, please.
EAK.
--
---------1---------2---------3---------4---------5---------6---------7
Eric Kimminau eak@detroit.sgi.com Field Technical Support
SGI 39001 W. 12 Mile Rd. Farmington, MI 48331-2903
Vox:(248)848-4455 Fax:(248)848-5600 VNET:6-327-4455
"I speak my mind and no one else's."
"I am a bomb technician. If you see me running, try to keep up..."
http://support.sgi.com
^ permalink raw reply [flat|nested] 22+ messages in thread* Re: [Fwd: linus.linux.sgi.com]
1999-01-20 15:18 ` Eric Kimminau
@ 1999-01-20 21:55 ` David Watters
1999-01-21 7:11 ` David S. Miller
1 sibling, 0 replies; 22+ messages in thread
From: David Watters @ 1999-01-20 21:55 UTC (permalink / raw)
To: eak; +Cc: linux
On Jan 20, 10:18am, Eric Kimminau wrote:
> It is not a rumor. It is fact. The ram being used in the VPC is like the
> memory in no other PC (that I know of) being manufactured today.
Yes it is. Same chips, less common DIMM.
Calm down Eric... let me bring you back. :)
Linux... Linux... Linux...
--
David Watters | Silicon*Graphics
Systems Engineer | http://www.sgi.com/
Silicon Graphics, Inc.| http://reality.sgi.com/davester/ (6/13/97)
david.watters@sgi.com | 1.800.800.SGI1 (Sales) 1.800.800.4SGI (Support)
DID 1.614.844.3820 | http://www.nintendo.com/ (N64, the $130 SGI!)
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
1999-01-20 15:18 ` Eric Kimminau
1999-01-20 21:55 ` David Watters
@ 1999-01-21 7:11 ` David S. Miller
1999-01-22 19:14 ` Jeffrey Watts
1 sibling, 1 reply; 22+ messages in thread
From: David S. Miller @ 1999-01-21 7:11 UTC (permalink / raw)
To: eak; +Cc: dmerchan, linux
Date: Wed, 20 Jan 1999 10:18:54 -0500
From: Eric Kimminau <eak@detroit.sgi.com>
dual PCI bus (get that from a clone) (and what does it cost to get
it on a Sun?)
~$2,000 with monitor and keyboard
Later,
David S. Miller
davem@dm.cobaltmicro.com
^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [Fwd: linus.linux.sgi.com]
1999-01-21 7:11 ` David S. Miller
@ 1999-01-22 19:14 ` Jeffrey Watts
0 siblings, 0 replies; 22+ messages in thread
From: Jeffrey Watts @ 1999-01-22 19:14 UTC (permalink / raw)
To: David S. Miller; +Cc: eak, dmerchan, linux
On Wed, 20 Jan 1999, David S. Miller wrote:
> dual PCI bus (get that from a clone) (and what does it cost to get
> it on a Sun?)
>
> ~$2,000 with monitor and keyboard
Please take SGI/Sun advocacy off this list. I know it is hard to not
respond to flamebait like that, but please try.
Thanks,
Jeffrey.
o-----------------------------------o
| Jeffrey Watts |
| watts@sunflower.com o-------------------------------------o
| Systems Analyst | "Outside of the killings, |
| Sprint - Systems Management | Washington has one of the lowest |
o-------------------------------| crime rates in the country." |
| -- Mayor Marion Barry |
o-------------------------------------o
^ permalink raw reply [flat|nested] 22+ messages in thread