X takes up 188 MB of memory

08-12-02, 05:19 AM

Hi

I'm using the nvidia driver (version 1.0.2314) on the student PC's running Red Hat 7.3 at the university where I work.

When X is started, before I even log in, X uses 188 MB of memory. That is a bit too much, when the machines have only 256 MB RAM... On my colleagues PC he runs the standard Red Hat driver, and his X uses "only" 120 MB of RAM. That means that the Nvidia driver uses 68 MB more than the standard one.

My question is, is the nvidia driver supposed to take up that much memory?

I know I'm not using the newest version, but I imagine a newer one will just take up even more memory?

The Graphichs card I use is some Nvidia GeForce 2 thing, 2-3 years old.

Best regards,

**Thunderbird** · 08-12-02, 05:42 AM

It is normal that X uses that much memory. In that amount of memory are X apps and so on included too. Try to use a newer driver version perhaps a newer version fixes some "problems".

r0gu3 · 08-12-02, 09:02 AM

Well i HAVE to speak to this point

First of all, if there is NOTHING running on a regular Xserver, there really should be not much memory being used, BUT if your using nvidia's drivers, than you will realize that it still uses a ton of resources. Nvidia's drivers aren't written to be extermely clean, as the xfree86 drivers are, Nvidia's drivers are written to be extremely fast.

- r0gu3

nutball · 08-12-02, 09:46 AM

I've seen this behaviour. Are you assessing size using 'top'? top was reporting my X server to be using 256MB of memory on a box with 128MB RAM + 128MB swap. The fact that the box worked quite normally suggested that the numbers weren't quite right somehow.

It's not clear to me that this is actual, used, physical memory. I wonder if it might be a shared-memory segment, or something.

08-12-02, 11:27 AM

Do not be frightened of big values in 'top' or 'ps'. These programs shows memory segment sizes associated with given process, neither the physical memory usage nor "common" virtual memory (physical + swap). Just nvidia driver uses device memory mapping (mmap("/dev/nvidia0",..) into process memory space) and uncommitted SysV IPC shared memory. Most of this memory are "pure virtual", i.e. does not associated with physical pages, but part of them are associated with AGP memory.

PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
2349 root 9 0 304M 15M 4824 S 0,0 6,2 8721m X

try to run cat /proc/2349/status (where 2349 is my PID of X, replace it by your own):

VmSize: 329020 kB
VmLck: 4294922036 kB
VmRSS: 15956 kB
VmData: 44248 kB
VmStk: 464 kB
VmExe: 1468 kB
VmLib: 4976 kB

We can see that X uses only 15MB of physical memory (RSS - Resident Set Size) and 44248 +464+1468+4976 (i.e. ~50MB) of "common" virtual memory (so ~35 MB in swap), where 5,5MB (VmExe and VmLib) can be shared (possibly) with another processes (another
copies of X for VmExe and another processes using the same shared libraries for VmLib).

08-12-02, 04:34 PM

As a couple have said it's not likely to be a big problem.

The mapped memory can be seen by doing

$ cat /proc/`pidof X`/maps

You'll see, amongst other things (like libc, libGL and other libraries), the driver will map the graphics card ram, agp space and registers.

08-12-02, 05:19 AM	#1
luke Guest Posts: n/a	X takes up 188 MB of memory Hi I'm using the nvidia driver (version 1.0.2314) on the student PC's running Red Hat 7.3 at the university where I work. When X is started, before I even log in, X uses 188 MB of memory. That is a bit too much, when the machines have only 256 MB RAM... On my colleagues PC he runs the standard Red Hat driver, and his X uses "only" 120 MB of RAM. That means that the Nvidia driver uses 68 MB more than the standard one. My question is, is the nvidia driver supposed to take up that much memory? I know I'm not using the newest version, but I imagine a newer one will just take up even more memory? The Graphichs card I use is some Nvidia GeForce 2 thing, 2-3 years old. Best regards,

08-12-02, 09:46 AM	#4
nutball International God of Sex Join Date: Jul 2002 Location: en.gb.uk Posts: 655	I've seen this behaviour. Are you assessing size using 'top'? top was reporting my X server to be using 256MB of memory on a box with 128MB RAM + 128MB swap. The fact that the box worked quite normally suggested that the numbers weren't quite right somehow. It's not clear to me that this is actual, used, physical memory. I wonder if it might be a shared-memory segment, or something. __________________ Got nuts?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
AgigaTech Patents Non-Volatile Memory DIMM Technology	News	Latest Tech And Game Headlines	0	09-03-12 04:10 PM
GPU memory consumption, back buffer, Quadro	tdsussman	NVIDIA Linux	5	08-14-12 09:14 PM
DOE Doles Out Cash to AMD, Whamcloud for Exascale Research	News	Latest Tech And Game Headlines	0	07-12-12 07:50 PM
Measuring card memory usage	peter_ga	NVIDIA Linux	0	05-24-12 08:07 PM
new memory tech QBM (Quad band memory)	sebazve	Rumor Mill	9	09-24-02 09:38 PM

08-12-02, 05:42 AM	#2
Thunderbird Join Date: Jul 2002 Location: Netherlands, Europe Posts: 2,105	It is normal that X uses that much memory. In that amount of memory are X apps and so on included too. Try to use a newer driver version perhaps a newer version fixes some "problems".

08-12-02, 09:02 AM	#3
r0gu3 Registered User Join Date: Jul 2002 Location: cleveland ohio Posts: 71	Well i HAVE to speak to this point First of all, if there is NOTHING running on a regular Xserver, there really should be not much memory being used, BUT if your using nvidia's drivers, than you will realize that it still uses a ton of resources. Nvidia's drivers aren't written to be extermely clean, as the xfree86 drivers are, Nvidia's drivers are written to be extremely fast. - r0gu3

08-12-02, 11:27 AM	#5
Sapa Guest Posts: n/a	Do not be frightened of big values in 'top' or 'ps'. These programs shows memory segment sizes associated with given process, neither the physical memory usage nor "common" virtual memory (physical + swap). Just nvidia driver uses device memory mapping (mmap("/dev/nvidia0",..) into process memory space) and uncommitted SysV IPC shared memory. Most of this memory are "pure virtual", i.e. does not associated with physical pages, but part of them are associated with AGP memory. PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND 2349 root 9 0 304M 15M 4824 S 0,0 6,2 8721m X try to run cat /proc/2349/status (where 2349 is my PID of X, replace it by your own): VmSize: 329020 kB VmLck: 4294922036 kB VmRSS: 15956 kB VmData: 44248 kB VmStk: 464 kB VmExe: 1468 kB VmLib: 4976 kB We can see that X uses only 15MB of physical memory (RSS - Resident Set Size) and 44248 +464+1468+4976 (i.e. ~50MB) of "common" virtual memory (so ~35 MB in swap), where 5,5MB (VmExe and VmLib) can be shared (possibly) with another processes (another copies of X for VmExe and another processes using the same shared libraries for VmLib).

08-12-02, 04:34 PM	#6
leahcim Guest Posts: n/a	As a couple have said it's not likely to be a big problem. The mapped memory can be seen by doing $ cat /proc/`pidof X`/maps You'll see, amongst other things (like libc, libGL and other libraries), the driver will map the graphics card ram, agp space and registers.