Nvidia's G92 GPU does not get a 512-bit memory bus interface - they stay with 384-bit
 

http://theinquirer.net/?article=42015


G92 is a high-end chip, sports 384-bit controller

INQ moves to clear the confusion

By Theo Valich
RUMOUR HAS IT that upcoming G92 is a mainstream chip, not the high-end
refresh that the net has been wibbling about.

Well, we can tell you that this rumour is nothing else but a bit of
good old FUD, since either G92 is a high-end refresh or the company
decided that a replacement for 8800GTS will be significantly faster
than GeForce 8800 Ultra, leaving its high-end offering in shambles.

This baby sports some serious shading power, but more important is the
fact that the memory installed on board is now GDDR4. 768MB of GDDR4,
to be more precise. The amount of memory discards rumours of 256-bit
bus, since it is obvious that Nvidia will keep 384-bit memory
controller for the high-end series of products. We would welcome this
memory controller in mainstream arena, though.

The bandwidth has now jumped over the 105 GB/s barrier, and it remains
to be seen what will be the final clock of the memory - our estimate
is between 1.0 and 1.2 GHz, or 2.0-2.4 GHz, but final clocks are far,
far from being decided. The company needs to get the revised chip
first, in order to have DisplayPort working nice and cleanly. Display
Port is required for this company to get the Dell XPS contract - a new
machine will be launched for Winter 2007/08, probably during CES 2008
in Las Vegas.

The company has already had some revisions of the board sent to their
favourite game developers and other partners, and we can now tell you
that the board is almost identical to old 8800 GTX/Ultra ones.

We would like to see 1.5GB of memory on consumer boards as well - the
Quadro FX already spots 1.5GB of GDDR3 memory- but it is all a matter
of price. Qimonda's high-capacity GDDR3 sounds very tasty for Nvidia
plans, but for now. The reference boards sport Samsung GDDR4 chips.

Nvidia's G92 GPU does not get a 512-bit memory bus interface - they stay with 384-bit
 

On Mon, 03 Sep 2007 16:45:18 -0700, Air Raid
wrote:

http://theinquirer.net/?article=42015


G92 is a high-end chip, sports 384-bit controller

INQ moves to clear the confusion

By Theo Valich
RUMOUR HAS IT that upcoming G92 is a mainstream chip, not the high-end
refresh that the net has been wibbling about.

Well, we can tell you that this rumour is nothing else but a bit of
good old FUD, since either G92 is a high-end refresh or the company
decided that a replacement for 8800GTS will be significantly faster
than GeForce 8800 Ultra, leaving its high-end offering in shambles.

This baby sports some serious shading power, but more important is the
fact that the memory installed on board is now GDDR4. 768MB of GDDR4,
to be more precise. The amount of memory discards rumours of 256-bit
bus, since it is obvious that Nvidia will keep 384-bit memory
controller for the high-end series of products. We would welcome this
memory controller in mainstream arena, though.

The bandwidth has now jumped over the 105 GB/s barrier, and it remains
to be seen what will be the final clock of the memory - our estimate
is between 1.0 and 1.2 GHz, or 2.0-2.4 GHz, but final clocks are far,
far from being decided. The company needs to get the revised chip
first, in order to have DisplayPort working nice and cleanly. Display
Port is required for this company to get the Dell XPS contract - a new
machine will be launched for Winter 2007/08, probably during CES 2008
in Las Vegas.

The company has already had some revisions of the board sent to their
favourite game developers and other partners, and we can now tell you
that the board is almost identical to old 8800 GTX/Ultra ones.

We would like to see 1.5GB of memory on consumer boards as well - the
Quadro FX already spots 1.5GB of GDDR3 memory- but it is all a matter
of price. Qimonda's high-capacity GDDR3 sounds very tasty for Nvidia
plans, but for now. The reference boards sport Samsung GDDR4 chips.


Digitimes is far closer to the truth.

http://www.digitimes.com/news/a20070829PD216.html

nV has not been spending their design time since releasing the 8800
series just spinning their wheels. The Dx10 refresh
(second-generation) is a complete redesign, not just a 65nm shrink and
a significant performance boost for all members of the new series.

Do not expect the high-end GPUs to be first out of the chute. nVidia
is well aware of the lucrative $$/performance-gap between the 8600GTS
and the 8800GTS. They are also very anxious indeed to phase out the
huge, hot, low-yield G80 (8800-series) ASAP. The profit margin is
nowhere near what they would like it to be. A replacement for the
8800GTS with the performance of the 8800GTX initially at the same
price as the 8800GTS (and with a far bigger profit margin on the
smaller-geometry chip) would be just what the nV financial doctor
ordered. Volume would be huge and the ATi 2900 competition would be
totally wiped out. The current 8800 series would phase out rapidly.
And, of course, the genuine high-end version of the new series still
waiting in the wings.......

The (National) Inquirer lives up to its name....again...

Regardless of how the scenario plays out, anybody actually purchasing
the current 8800-family (any flavor) within the next few months may be
a very unhappy camper indeed by the turn of the year.

John Lewis

Nvidia's G92 GPU does not get a 512-bit memory bus interface- they stay with 384-bit
 

John Lewis wrote:
On Mon, 03 Sep 2007 16:45:18 -0700, Air Raid
wrote:

http://theinquirer.net/?article=42015


G92 is a high-end chip, sports 384-bit controller

INQ moves to clear the confusion

By Theo Valich
RUMOUR HAS IT that upcoming G92 is a mainstream chip, not the high-end
refresh that the net has been wibbling about.

Well, we can tell you that this rumour is nothing else but a bit of
good old FUD, since either G92 is a high-end refresh or the company
decided that a replacement for 8800GTS will be significantly faster
than GeForce 8800 Ultra, leaving its high-end offering in shambles.

This baby sports some serious shading power, but more important is the
fact that the memory installed on board is now GDDR4. 768MB of GDDR4,
to be more precise. The amount of memory discards rumours of 256-bit
bus, since it is obvious that Nvidia will keep 384-bit memory
controller for the high-end series of products. We would welcome this
memory controller in mainstream arena, though.

The bandwidth has now jumped over the 105 GB/s barrier, and it remains
to be seen what will be the final clock of the memory - our estimate
is between 1.0 and 1.2 GHz, or 2.0-2.4 GHz, but final clocks are far,
far from being decided. The company needs to get the revised chip
first, in order to have DisplayPort working nice and cleanly. Display
Port is required for this company to get the Dell XPS contract - a new
machine will be launched for Winter 2007/08, probably during CES 2008
in Las Vegas.

The company has already had some revisions of the board sent to their
favourite game developers and other partners, and we can now tell you
that the board is almost identical to old 8800 GTX/Ultra ones.

We would like to see 1.5GB of memory on consumer boards as well - the
Quadro FX already spots 1.5GB of GDDR3 memory- but it is all a matter
of price. Qimonda's high-capacity GDDR3 sounds very tasty for Nvidia
plans, but for now. The reference boards sport Samsung GDDR4 chips.


Digitimes is far closer to the truth.

http://www.digitimes.com/news/a20070829PD216.html

nV has not been spending their design time since releasing the 8800
series just spinning their wheels. The Dx10 refresh
(second-generation) is a complete redesign, not just a 65nm shrink and
a significant performance boost for all members of the new series.

Do not expect the high-end GPUs to be first out of the chute. nVidia
is well aware of the lucrative $$/performance-gap between the 8600GTS
and the 8800GTS. They are also very anxious indeed to phase out the
huge, hot, low-yield G80 (8800-series) ASAP. The profit margin is
nowhere near what they would like it to be. A replacement for the
8800GTS with the performance of the 8800GTX initially at the same
price as the 8800GTS (and with a far bigger profit margin on the
smaller-geometry chip) would be just what the nV financial doctor
ordered. Volume would be huge and the ATi 2900 competition would be
totally wiped out. The current 8800 series would phase out rapidly.
And, of course, the genuine high-end version of the new series still
waiting in the wings.......

The (National) Inquirer lives up to its name....again...

Regardless of how the scenario plays out, anybody actually purchasing
the current 8800-family (any flavor) within the next few months may be
a very unhappy camper indeed by the turn of the year.

John Lewis

I recently bought a new PC after 3-4 years (old one P4 3Gig with nVidia
7300 GPU - New one Core 2 Duo 6600 with 8800GTS). I have gone back to a
lot of my old games (eg Return to Castle Wolfenstein, etc) and cranked
up the graphics - lovely! I purchased FEAR and love running it with
everything cranked to the max.
I am a happy camper with the 8800 I have, but will be looking with
interest when the new gear comes out!

cheers
Gorby

Nvidia's G92 GPU does not get a 512-bit memory bus interface - they stay with 384-bit
 

Rem: The more memory a Graphic card has the more that memory has to use
upper
addresses & get in the way of main Mem ..a problem issue with 32 bit O/S 's
....with a surprising number of people using SLI 2 lots of 1 gig or more
would
cut of the main mem to around 2 Gig's .
mouse
@@

Nvidia's G92 GPU does not get a 512-bit memory bus interface - they stay with 384-bit
 

"Trimble Bracegirdle" writes:

Rem: The more memory a Graphic card has the more that memory has to use
upper
addresses & get in the way of main Mem ..a problem issue with 32 bit O/S 's
...with a surprising number of people using SLI 2 lots of 1 gig or more
would
cut of the main mem to around 2 Gig's .
mouse
@@

You have no clue what you are talking about.

Nvidia's G92 GPU does not get a 512-bit memory bus interface - they stay with 384-bit
 

"Walter Mitty" wrote in message ...
"Trimble Bracegirdle" writes:

Rem: The more memory a Graphic card has the more that memory has to use
upper
addresses & get in the way of main Mem ..a problem issue with 32 bit O/S 's
...with a surprising number of people using SLI 2 lots of 1 gig or more
would
cut of the main mem to around 2 Gig's .
mouse
@@

You have no clue what you are talking about.

You chipset disables RAM when the memory used by PCI devices conflicts.