Nvidia SLI, SLI's back with a vengeance

CapFusion wrote:


Smells like desperation to me. Seems like they can't keep up ATI's
technology so they're going for the brute force approach. Ironically,
NVidia criticized 3dfx for the same thing back in the late nineties.



Desperation or not, I do not see anything wrong with this. If they have
this leverage, why not use it? ATi will find some technology or else try
to trail as close it can until they come up with something. This the cruel
and brutal of business. Technology will advance as rival find new way to
be better than the other.

Or the market will look at it and snore.

CapFusion,...

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

"J. Clarke" wrote in message
...

Or the market will look at it and snore.


It may or may not. It really depend on the current user and how it deal with
it. If that technology produce the same image as Intel famous RDRAM [PC800
etc]. Then it may become a flop until user either totally reject it or very
slowly embrace it technology. During this time, maybe ATi will have
something new to combat this SLI from nVIDIA. Only time will tell.

CapFusion,...

Curious...I wonder why they decided to split the rendering horizonatally
as opposed to vertically...seems if they split the rendering vertically
they wouldn't have to bother w/ a separate algorithim to balance the rendering
load in realtime. If I understand the process correctly the rendering is
not split 50/50 but based on the rendering load of a scene/screen...
Furthermore...would this balancing act suckup GPU processing power??

....that could be used to render the scene perhaps???

Just seems to make more sense...perhaps someone can shed some light on
my ignorance here???

Thanks

Redbrick...who Loves his CLK



In article , says...

On Thu, 01 Jul 2004 14:50:21 GMT, "First of One"
wrote:

This is actually a very good technique. Dividing the screen into two
load-balanced halves means there's no redundant texture memory usage like
3dfx's scanline interleaving, and no mouse lag like ATi's alternate-frame
rendering. Wicked3D had something similar a couple of years ago, but the
immature drivers back then produced a black line between the two image
halves.

The real card to watch for is the 6800GT, which may actually be affordable
in SLI config.


Agreed !!

Once you have a PCI-Express chip-set that will support 2 or more
PCI-express sockets. The nforce4 chip-set, still in design at nVidia,
is very likely to do just that. And may for the first time make me
thinkvery seriously about leaving the Intel camp. Athlon 64 FX53
939-pin-- unlocked overclock, plus nForce4, plus dual 6800GT
PCI-express in SLI-configuration; the thought makes me really drool,
( and my pocket-book wilt ).

As far as the enthusiast community goes, Intel has really lost their
way in the past year. Besides power-hungry Prescott, the latest
Intel miss-step is to DELIBERATELY build-in a 10% overclock limit
into the 915/925 chip-sets. Intel has again become arrogant - they
periodically do that until the threat of real competition beats them
over the head.

John Lewis


--
"War is the continuation of politics by other means.
It can therefore be said that politics is war without
bloodshed while war is politics with bloodshed."


"SLIisBACK" wrote in message
...
http://media.hardwareanalysis.com/ar...arge/11206.jpg
http://media.hardwareanalysis.com/ar...arge/11208.jpg
http://media.hardwareanalysis.com/ar...arge/11207.jpg

http://www.hardwareanalysis.com/content/article/1728/

Folk wrote:

On Wed, 30 Jun 2004 19:16:52 -0400, "J. Clarke"
wrote:


Since a PCI Express video card uses a completely different design
(16X) and form factor than a 'regular' PCI Express slot, what makes
you think that having two video slots will block a 'regular' slot?

No, a PCI Express video card does _not_ use "a completely different design
and form factor than a 'regular' PCI Express slot". Any PCI Express board
is _supposed_ to work in any PCI Express slot with the same number of
lines
or higher. So a PCI Express 1X board is supposed to plug into that 16x
slot and work fine.

Further, Alienware says that they use all the available PCI Express lanes
on
the Intel 7525 chipset in their dual processor design. The 7525 has 24
PCI
Express lines arranged as 1 16x and 1 8x that can be split into 2 4x. So
they don't even have a full 16x port for the second video board, let alone
any lanes left over for other devices.

OK, I've been wrong before, but look he

http://www6.tomshardware.com/motherb...erwood-09.html

That shows the difference between a 1X and a 16X PCI slot.

What of it? Look closely and you'll see that there is nothing that prevents
a 1x board from being plugged into a 16x slot. The spacing of the contacts
is the same. There is one crosspiece in the slot and it is the same
distance from the left end on both. If you plug a 1x (or 4x or 8x) board
into a 16x slot, it will fit fine, it just won't use all the contacts,
anymore than a typical AGP board uses all the contacts in an AGP Pro slot.

Now most
mobos (discounting exotic designs like Alienware) are going to come
with a single 16X and one or more 1X slots. You're saying that a 1X
card will fit in a 16X slot, and that may be true, but I doubt anyone
will actually do that.

Of course they will. Consider some future machine that has all 16x slots.
Or somebody who gets the Alienware machine and then decides later that he
wants to use it as a server and pulls one of the video boards to use in
another machine, leaving its slot available for his gigabit board.

I haven't seen any board layouts yet, but it's certainly possible that
a board with two 16X slots to accommodate an SLI setup would have the
1X slots positioned far enough away from the dual 16X slots to make
the concept of "wasting a slot" disappear. Wouldn't that make sense
to you too?

I'm sorry, but if you have three slots and two of them are taken up by video
then there is only one additional slot available. The fact that it is
available does not mean that the slot with the second video board is also
available, thus the video board is tying up a PCI Express slot that could
be used for other purposes.

Regardless of any of this, the Alienware machine is _not_ going to have any
1X slots. They state that they are using the Intel 7525 chipset. The docs
for the 7525 are available on the Intel site (go to "workstation and
server" then "chipsets"). If you read them you will find that the 7525
does not support 1x slots, it supports a single 16x and either a single 8x
or two 4x slots. Alienware states that they are using all the available
lanes for video, which means that they are using the 8x slot for the second
video board and there will be no 4x slots, let alone 1x.

--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

OMG. *drooooooll* What a beast!!!

On Thu, 1 Jul 2004 12:10:40 -0400, "Tim"
wrote:


"SLIisBACK" wrote in message
...
http://media.hardwareanalysis.com/ar...arge/11206.jpg
http://media.hardwareanalysis.com/ar...arge/11208.jpg
http://media.hardwareanalysis.com/ar...arge/11207.jpg

http://www.hardwareanalysis.com/content/article/1728/


_________________________________________________ ___________________________
___
Nvidia SLI, SLI's back with a vengeance Jun 28, 2004, 07:30 AM


Smells like desperation to me. Seems like they can't keep up ATI's
technology

In what way ? Please explain ?

I though that it was nVidia that had overcome the significant
intricacies of a Dx9.0c implementation, but maybe I am
reading the wrong technical literature.

so they're going for the brute force approach.

Not quite. What nvIdia is doing is a simple microcosm for
desktop computers and graphic applications of the shared
processing approach used world-wide by number-
crunching super-computers. nVidia has had the foresight
to implemented the sharing mechanism in their current
silicon. Not exactly a new concept. In a similar domain a
few years ago, I was involved in the design of chips for
time-simultaneous processing of the 3 channels of
component-video (Y, Cr, Cb ) each with a link-port for
accurate synchronization and to coordinate task-
sharing with the other two chips.

John Lewis

Ironically,
NVidia criticized 3dfx for the same thing back in the late nineties.


Yes, solely for marketing reasons, never technical.

John Lewis

Unlike the 3dfx VSA-100, nowadays a single 6800 Ultra is competitive with
the X800XT, so this SLI thing is really just a matter of image and bragging
rights. Seriously, 0.5 GB video RAM, 32 textures in a single cycle, four
expansion slots...

The most thumpingly expensive setup is Quadro SLI. Total cost is at least
$5000.

--
"War is the continuation of politics by other means.
It can therefore be said that politics is war without
bloodshed while war is politics with bloodshed."


Smells like desperation to me. Seems like they can't keep up ATI's
technology so they're going for the brute force approach. Ironically,
NVidia criticized 3dfx for the same thing back in the late nineties.

On Thu, 01 Jul 2004 20:44:57 -0400, "J. Clarke"
wrote:


I'm sorry, but if you have three slots and two of them are taken up by video
then there is only one additional slot available. The fact that it is
available does not mean that the slot with the second video board is also
available, thus the video board is tying up a PCI Express slot that could
be used for other purposes.

I think you just like to argue.

Your initial complaint was about 'wasting' a PCI slot with an SLI
configuration. You have yet to see a mobo that will accommodate a
dual vid card setup, yet you seem convinced that when one does
surface, the board designers will stupidly put the other PCI slots so
close to those two that they will be unusable.

Regardless of any of this, the Alienware machine is _not_ going to have any
1X slots. They state that they are using the Intel 7525 chipset. The docs
for the 7525 are available on the Intel site (go to "workstation and
server" then "chipsets"). If you read them you will find that the 7525
does not support 1x slots, it supports a single 16x and either a single 8x
or two 4x slots. Alienware states that they are using all the available
lanes for video, which means that they are using the 8x slot for the second
video board and there will be no 4x slots, let alone 1x.

And why do you keep mentioning server chipsets and Alienware rigs?
That's not what the lion's share of this group's readers will be
interested in. Most enthusiasts will be using either the 925X or 915
chipsets, not some exotic Alienware or server class setup.

But whatever.... you seem to be one of those persons that is "always
right" and loves to argue about it. It just seemed odd to me that you
would immediately dismiss SLI as a blocker of PCI slots, when you
haven't even seen a real board design that accommodates SLI. Are you
psychic, or just a troll?

Folk wrote:

On Thu, 01 Jul 2004 20:44:57 -0400, "J. Clarke"
wrote:


I'm sorry, but if you have three slots and two of them are taken up by
video
then there is only one additional slot available. The fact that it is
available does not mean that the slot with the second video board is also
available, thus the video board is tying up a PCI Express slot that could
be used for other purposes.

I think you just like to argue.

Pot, kettle.

Your initial complaint was about 'wasting' a PCI slot with an SLI
configuration. You have yet to see a mobo that will accommodate a
dual vid card setup, yet you seem convinced that when one does
surface, the board designers will stupidly put the other PCI slots so
close to those two that they will be unusable.

I'm sorry, but I have at no point expressed concern that any slot will be
any particular distance from any other slot. The Alienware machine has two
and only two PCI Express slots. With the dual video there is nowhere to
plug in a third PCI device. It is not a matter of the third slot being "so
close to those two", it is that THERE IS NO THIRD SLOT.

Regardless of any of this, the Alienware machine is _not_ going to have
any
1X slots. They state that they are using the Intel 7525 chipset. The
docs for the 7525 are available on the Intel site (go to "workstation and
server" then "chipsets"). If you read them you will find that the 7525
does not support 1x slots, it supports a single 16x and either a single 8x
or two 4x slots. Alienware states that they are using all the available
lanes for video, which means that they are using the 8x slot for the
second video board and there will be no 4x slots, let alone 1x.

And why do you keep mentioning server chipsets and Alienware rigs?

Because that looks to be the _only_ machine coming in the near future that
will have a provision for PCI Express dual video?

That's not what the lion's share of this group's readers will be
interested in. Most enthusiasts will be using either the 925X or 915
chipsets, not some exotic Alienware or server class setup.

So let's see, you're going to plug your super hotrod PCI Express 6800 board
into a 4x slot to use SLI? And how well do you think that that is going to
work? Assuming of course that someone actually _makes_ a 915 or 925X
machine that has the 4 PCI Express lanes on the ICH6 configured as a single
4X slot instead of 4 1x slots. And if the machine _is_ configured with the
4x slot then it will have two and only two PCI Express slots both of which
have video boards plugged into them and there will be no way to attach a
third PCI Express device.

Now why, precisely, do you think that Alienware chose to use a workstation
chipset instead of a desktop chipset on their machine? Hmmm? Maybe a 4x
second slot that is the largest possible on the 925X and 915 doesn't have
enough bandwidth to make the dual video worth the effort?

But whatever.... you seem to be one of those persons that is "always
right" and loves to argue about it.

I'm quite happy to admit that I am wrong when I am. But I haven't seen you
provide any information that would lead me to that belief.

It just seemed odd to me that you
would immediately dismiss SLI as a blocker of PCI slots, when you
haven't even seen a real board design that accommodates SLI. Are you
psychic, or just a troll?

Who said anything about "a blocker of PCI slots"? First, nobody said
anything about PCI slots except YOU. It was PCI EXPRESS (got that,
EXPRESS, not the same as PCI) that was the concern, and the concern was not
that some slot would be "blocked" because it was "too close" to some other
slot, it was that the machines have damned few PCI Express slots to begin
with and the second video board is physically inserted into one of them.

As to being "psychic or just a troll", I'm neither. Just someone who has
read the spec sheets for the chipsets and understands their implications.

You seem determined to misunderstand the issue.

To simplify, the chipsets you favor are not capable of providing two 16x
slots. The _best_ they can do is a 16x and a 4x, and in that configuration
those are the _only_ PCI Express slots. If you accept that the utility of
an SLI machine with one video board in a 16x slot and one in a 1x slot is
virtually nonexistent, then you have to agree that if SLI is going to be
used with a 925X or a 915X then there will be exactly two PCI Express
slots, a 16x and a 4x, with a video board plugged into each slot, and with
NO other PCI Express slots anywhere in the machine. Do you begin to
understand the issue?


--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)

On Fri, 02 Jul 2004 05:40:38 GMT, "First of One"
wrote:

Unlike the 3dfx VSA-100, nowadays a single 6800 Ultra is competitive with
the X800XT, so this SLI thing is really just a matter of image and bragging
rights. Seriously, 0.5 GB video RAM, 32 textures in a single cycle, four
expansion slots...

The most thumpingly expensive setup is Quadro SLI. Total cost is at least
$5000.


Yep.

Pros pay $1000 where consumers pay $100 for almost the same
thing nowadays in the technology markets.

I do freelance video work and ensure maximum quality for
my capital-equipment-buck by very judiciously mixing pro
and 'high-end-domestic" tools and hardware.

John Lewis

--
"War is the continuation of politics by other means.
It can therefore be said that politics is war without
bloodshed while war is politics with bloodshed."


Smells like desperation to me. Seems like they can't keep up ATI's
technology so they're going for the brute force approach. Ironically,
NVidia criticized 3dfx for the same thing back in the late nineties.