A computer components & hardware forum. HardwareBanter

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

Go Back   Home » HardwareBanter forum » Video Cards » Nvidia Videocards
Site Map Home Register Authors List Search Today's Posts Mark Forums Read Web Partners

nvidia GTX 970 4 GiB stuffup



 
 
Thread Tools Display Modes
  #1  
Old April 12th 15, 01:48 PM posted to alt.comp.periphs.videocards.nvidia
[email protected]
external usenet poster
 
Posts: 220
Default nvidia GTX 970 4 GiB stuffup

So these cards have 3.5 GiB and 0.5 GiB of memory.
We went throught this about 10 years ago with Intel northbridge,
where a 32-bit Windows saw typically 3.25 GiB with 4 GiB installed.
If you had PAE linux, you got almost 4 GiB, but not continuous.
Didn't nvidia learn from history? Repeat!

  #2  
Old April 12th 15, 09:22 PM posted to alt.comp.periphs.videocards.nvidia
Paul
external usenet poster
 
Posts: 13,364
Default nvidia GTX 970 4 GiB stuffup

wrote:
So these cards have 3.5 GiB and 0.5 GiB of memory.
We went throught this about 10 years ago with Intel northbridge,
where a 32-bit Windows saw typically 3.25 GiB with 4 GiB installed.
If you had PAE linux, you got almost 4 GiB, but not continuous.
Didn't nvidia learn from history? Repeat!


That's not exactly true. The card in question has
one (or more) functional units turned off (for yield reasons),
preventing the memory interface unit for it from
working. There really is only 3.5GiB of memory.
None hidden. Not a PAE issue. The GPU ends up having
a weird number of memory interfaces on the side of
it, that leads to the 3.5GiB number.

The NVidia marketing department failed to get the
details right. And nobody technical caught the mistake
until it turned into a public relations disaster.

And obviously, for such a humongous card, you cannot
use a 32 bit Windows OS. There would simply be no
address space left over to allow the OS to run. And
this is caused by the Windows (address space limiting)
memory license for 32 bit Windows. PAE is enabled,
but only Ring0 code is allowed to use the entire space.
Applications running in Ring3 on a Windows 32 bit OS,
are not allowed outside the 4GB "box".

That large of a card, would mean practically none of
the system memory would be mapped, and Windows would
have no place to run. A 32 bit OS should really crash,
if that card had just been plugged in.

As an example, I run WinXP SP3 x32 for my "typing machine".
I have 8GB of RAM installed. Windows uses 4GB (minus
bus addresses of around 0.5GB). I have a RAMDisk which
uses the other 4GB. All 8GB are in usage right now.
(WinXP SP3 x32 has PAE enabled by default, for NX bit usage.)
The RAMDisk code is implemented as a "driver" and
lives in Ring0 with the kernel code, which is how
memory outside the licensed region can be accessed.
The code is not being used by a "program" (Ring3),
but is being used by a "driver" (Ring0), and that
is how the Windows memory license is partially
being bypassed.

I didn't buy the RAM solely for WinXP. Sometimes
I boot up the paid-for Windows 8.1 on this
machine, and then the entire memory can be
used by programs. So it's not totally lost.

And the RAMDisk idea is great. I use it for
unzipping large packages, ones that would really
fragment a hard drive. So if I unzip the source
for Thunderbird program, which is 125,000 files,
I keep them on the RAMDisk, which isn't slowed in
the least by fragmentation. In HDTune, the RAMDisk
has a bandwidth of slightly less than 4GB/sec.

(Look for "RAMDisk Lite", if your machine has more RAM
than a 32 bit OS can use. The free version can only use 4GB
of left-over RAM, while the paid version is likely to map
as much as 60GB or so. It's possible the PAE limit varies
with OS version. Something later than WinXP x32 might be
able to do more with PAE.)

http://memory.dataram.com/products-a...ftware/ramdisk

Paul
  #4  
Old April 24th 15, 03:54 AM posted to alt.comp.periphs.videocards.nvidia
Paul
external usenet poster
 
Posts: 13,364
Default nvidia GTX 970 4 GiB stuffup

Skybuck Flying wrote:


"Paul" wrote in message ...

wrote:
So these cards have 3.5 GiB and 0.5 GiB of memory.
We went throught this about 10 years ago with Intel northbridge,
where a 32-bit Windows saw typically 3.25 GiB with 4 GiB installed.
If you had PAE linux, you got almost 4 GiB, but not continuous.
Didn't nvidia learn from history? Repeat!


"
That's not exactly true. The card in question has
one (or more) functional units turned off (for yield reasons),
preventing the memory interface unit for it from
working. There really is only 3.5GiB of memory.
"

This would be news to me ?!

I think you might be miss-understanding.

There is 4 GB of RAM on it.

However only one section can be used at a time for reading.

So either 3.5 GB or 500 MB.

It is claimed to be possible to read from 3.5 GB and write to 500 MB at
the same time or vice versa.

Bye,
Skybuck.


You are correct.

http://www.anandtech.com/show/8935/g...ory-allocation

But you don't really want to do that.

"In the case of pure reads for example, GTX 970 can read the 3.5GB segment
at 196GB/sec (7GHz * 7 ports * 32-bits), or it can read the 512MB segment
at 28GB/sec, but it cannot read from both at once.

The same is also true for writes, as only one segment can be written to
at a time.

Unfortunately what this means is that accessing the weaker 512MB segment
blocks access to the stronger 3.5GB segment if both memory operations are
identical; or put another way, using the 512MB segment can harm the
performance of the 3.5GB segment. For example, if we want to issue reads
to both segments at once, reading the 512MB segment blocks any other reads
to the 3.5GB segment for that cycle. If the 3.5GB segment is blocked in
this fashion and doesn't have a non-blocking write to work on instead,
it would have to go idle for that cycle, which would reduce the effective
memory bandwidth of the 3.5GB segment. This means that taken over time
in our example, the larger the percentage of the time the crossbar is
reading the 512MB segment, the lower the effective read memory bandwidth
would be from the 3.5GB segment."

So it's best if the 512MB section is not used.

One of the tools listed in that article, shows this. And is
probably why I concluded 3.5GB was present when I glanced at
the article the first time.

http://images.anandtech.com/doci/8931/970_35GB.png

I wasn't really all that interested, as the chances of me
owning one of those is just about zero :-) I don't have
money for 4GB (or 3.5GB) video cards :-) And I still haven't
read the article to the very end. My eyes are glazing over.

Paul


  #5  
Old April 24th 15, 09:29 PM posted to alt.comp.periphs.videocards.nvidia
Skybuck Flying[_4_]
external usenet poster
 
Posts: 480
Default nvidia GTX 970 4 GiB stuffup

A better question is why that tool is only reporting 3.5 GB ?!

Cuda will report 4 GB ?!

So that's weird.

Bye,
Skybuck.
  #6  
Old April 24th 15, 09:30 PM posted to alt.comp.periphs.videocards.nvidia
Skybuck Flying[_4_]
external usenet poster
 
Posts: 480
Default nvidia GTX 970 4 GiB stuffup

Funny enough in same article this picture does show 4GB of RAM:

http://api.viglink.com/api/click?for...75px.png%22%3E

Very strange indeed.

Bye,
Skybuck.

 




Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

vB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump

Similar Threads
Thread Thread Starter Forum Replies Last Post
Nvidia's Ageia Purchase - Why Ageia matters - Next Generation SonyPlayStation GPU - Next Generation PC GPUs - Next Generation Console GPU -Nvidia Inside - Nvidia Everywhere sprite scaler AMD x86-64 Processors 0 February 8th 08 02:43 AM
Nvidia's Ageia Purchase - Why Ageia matters - Next Generation SonyPlayStation GPU - Next Generation PC GPUs - Next Generation Console GPU -Nvidia Inside - Nvidia Everywhere sprite scaler Ati Videocards 0 February 8th 08 02:43 AM
Nvidia's Ageia Purchase - Why Ageia matters - Next Generation SonyPlayStation GPU - Next Generation PC GPUs - Next Generation Console GPU -Nvidia Inside, Nvidia Everywhere sprite scaler Nvidia Videocards 0 February 8th 08 02:41 AM
nvidia forceware driver; Vista drivers; asus/nvidia vga card (?) JPK Nvidia Videocards 3 March 4th 07 05:39 AM
Guru3D nvidia boys and other tests comparing NVIDIA FX - ATI RADEON Juan Alvarez Nvidia Videocards 5 October 1st 03 01:29 AM


All times are GMT +1. The time now is 12:11 PM.


Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Copyright ©2004-2024 HardwareBanter.
The comments are property of their posters.