driver bug still exists...97.44 and up....mem stuck.

T

TheRapture

Supreme [H]ardness
Joined
May 31, 2000
Messages
6,885
Yes, it does exist. Verified using BOTH atitool and rivatuner. Atitool will never find a max mem overclock because that's right, it is not actually changing ram speeds on the fly as it tests. Mine will max out hehe...yeah right, my GTS can run the memory at 1350 (2700 DDR) :p


I can run a a test in Doom3 at 1680x1050 with 4xaa and 16xaf (no vsync), with the card at stock speeds and get 108fps in timedemo1.

With the card set to only overclock the core to 625, I get 114fps, that's a gain of 6fps from the core overclock.

When I "overclock" the ram to 2000 along with the core, guess what, still only 114fps.

With the core at stock and the mem "overclocked" I still only get 108fps (same as completely stock).


The ram is NOT being overclocked with the 97.44 and 97.92 driver. I have not tested the 97.02's but I think I remember seeing that they are also affected. All benches bear this out.

So I bet LOTS of you are not actually seeing the overclocks you think you are....
 
GoldenTiger said:
Bleh... hope this is fixed soon!
Click to expand...


Here's a good tip straight from http://forums.guru3d.com//showthread.php?t=205416&page=3



Small trick for G80 owners:

You can use programmable GPUProbe plugin to monitor clocks of all 3 memory clock domains (2 of which are affected by bug of 97.28+ drivers and constantly run @ BIOS defined 400MHz clock). To add an ability of 3 memory domain clocks monitoring replace .\Plugins\Monitoring\GPUProbe.cfg file with the following text:


;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; NV1X compatible GPU presence checking routine
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[IsNV1X]

Cmd0 = 00000013 00001800
Cmd1 = 00000114 0000FFFF
Cmd2 = 00000A14 000010DE
Cmd3 = 00000B10

; MOV ACC , GPU_1800
; AND ACC , FFFFh
; CMP ACC , 10DEh
; SETZ ACC

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; NV4X compatible GPU presence checking routine
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[IsNV4X]

Cmd0 = 00000013 00001800
Cmd1 = 00000114 0000FFFF
Cmd2 = 00000A14 000010DE
Cmd3 = 00000B20 00000000
Cmd4 = 00000013 00000000
Cmd5 = 00000414 00000014
Cmd6 = 00000114 000000F0
Cmd7 = 00000A14 00000040
Cmd8 = 00000B10
Cmd9 = 00000112 00000000

; MOV ACC , GPU_1800
; AND ACC , FFFFh
; CMP ACC , 10DEh
; SETZ REG0
; MOV ACC , GPU_0
; SHR ACC , 20
; AND ACC , F0h
; CMP ACC , 40h
; SETZ ACC
; AND ACC , REG0

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; NV5X compatible GPU presence checking routine
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[IsNV5X]

Cmd0 = 00000013 00088000
Cmd1 = 00000114 0000FFFF
Cmd2 = 00000A14 000010DE
Cmd3 = 00000B20 00000000
Cmd4 = 00000013 00000000
Cmd5 = 00000414 00000014
Cmd6 = 00000114 000000F0
Cmd7 = 00000A14 00000050
Cmd8 = 00000B10
Cmd9 = 00000112 00000000

; MOV ACC , GPU_88000
; AND ACC , FFFFh
; CMP ACC , 10DEh
; SETZ REG0
; MOV ACC , GPU_0
; SHR ACC , 20
; AND ACC , F0h
; CMP ACC , 50h
; SETZ ACC
; AND ACC , REG0

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 0 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source0]

Name = GPIO 0
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 00600818
GetDataCmd1 = 00000114 00000100
GetDataCmd2 = 00000414 00000008

; MOV ACC , GPU_600818
; AND ACC , 100h
; SHR ACC , 8h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 1 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source1]

Name = GPIO 1
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 00600818
GetDataCmd1 = 00000114 01000000
GetDataCmd2 = 00000414 00000018

; MOV ACC , GPU_600818
; AND ACC , 1000000h
; SHR ACC , 18h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 2 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source2]

Name = GPIO 2
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00000004
GetDataCmd2 = 00000414 00000002

; MOV ACC , GPU_60081C
; AND ACC , 4h
; SHR ACC , 2h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 3 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source3]

Name = GPIO 3
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00000040
GetDataCmd2 = 00000414 00000006

; MOV ACC , GPU_60081C
; AND ACC , 40h
; SHR ACC , 6h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 4 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source4]

Name = GPIO 4
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00000400
GetDataCmd2 = 00000414 0000000A

; MOV ACC , GPU_60081C
; AND ACC , 400h
; SHR ACC , Ah

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 5 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source5]

Name = GPIO 5
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00004000
GetDataCmd2 = 00000414 0000000E

; MOV ACC , GPU_60081C
; AND ACC , 4000h
; SHR ACC , Eh

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 6 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source6]

Name = GPIO 6
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00040000
GetDataCmd2 = 00000414 00000012

; MOV ACC , GPU_60081C
; AND ACC , 40000h
; SHR ACC , 12h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 7 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source7]

Name = GPIO 7
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 00400000
GetDataCmd2 = 00000414 00000016

; MOV ACC , GPU_60081C
; AND ACC , 400000h
; SHR ACC , 16h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 8 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source8]

Name = GPIO 8
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 04000000
GetDataCmd2 = 00000414 0000001A

; MOV ACC , GPU_60081C
; AND ACC , 4000000h
; SHR ACC , 1Ah

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 9 pin monitoring for NV1X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source9]

Name = GPIO 9
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV1X
GetDataCmd0 = 00000013 0060081C
GetDataCmd1 = 00000114 40000000
GetDataCmd2 = 00000414 0000001E

; MOV ACC , GPU_60081C
; AND ACC , 40000000h
; SHR ACC , 1Eh

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 10 pin monitoring for NV4X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source10]

Name = GPIO 10
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV4X
GetDataCmd0 = 00000013 00600850
GetDataCmd1 = 00000114 00000004
GetDataCmd2 = 00000414 00000002

; MOV ACC , GPU_600850
; AND ACC , 4h
; SHR ACC , 2h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 11 pin monitoring for NV4X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source11]

Name = GPIO 11
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV4X
GetDataCmd0 = 00000013 00600850
GetDataCmd1 = 00000114 00000040
GetDataCmd2 = 00000414 00000006

; MOV ACC , GPU_600850
; AND ACC , 40h
; SHR ACC , 6h

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; GPIO 12 pin monitoring for NV4X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source12]

Name = GPIO 12
MaxLimit = 1
MinLimit = 0
GridDim = 2
GetDescCmd = IsNV4X
GetDataCmd0 = 00000013 00600850
GetDataCmd1 = 00000114 00000400
GetDataCmd2 = 00000414 0000000A

; MOV ACC , GPU_600850
; AND ACC , 400h
; SHR ACC , Ah

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; MPLL0 clock monitoring for NV5X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source13]

Name = Memory clock \ domain 0
Dim = MHz
MaxLimit = 1500
MinLimit = 0
GridDim = 10
GetDescCmd = IsNV5X
GetDataCmd0 = 00000023 00000000 0000400C
GetDataCmd1 = 00000023 00000001 00004008
GetDataCmd2 = 00000022 00000002 00000000
GetDataCmd3 = 00000124 00000002 000000FF
GetDataCmd4 = 00000022 00000003 00000000
GetDataCmd5 = 00000424 00000003 00000008
GetDataCmd6 = 00000124 00000003 000000FF
GetDataCmd7 = 00000022 00000004 00000001
GetDataCmd8 = 00000424 00000004 00000010
GetDataCmd9 = 00000124 00000004 00000007
GetDataCmd10 = 00000014 00006978
GetDataCmd11 = 00000812 00000003
GetDataCmd12 = 00000912 00000002
GetDataCmd13 = 00000412 00000004
GetDataCmd14 = 00000914 000000FA

; MOV REG0 , GPU_400C
; MOV REG1 , GPU_4008
; MOV REG2 , REG0
; AND REG2 , FFh
; MOV REG3 , REG0
; SHR REG3 , 8
; AND REG3 , FFh
; MOV REG4 , REG1
; SHR REG4 , 16
; AND REG4 , 7
; MOV ACC , 27000
; MUL ACC , REG3
; DIV ACC , REG2
; SHR ACC , REG4
; DIV ACC , 250

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; MPLL1 clock monitoring for NV5X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source14]

Name = Memory clock \ domain 1
Dim = MHz
MaxLimit = 1500
MinLimit = 0
GridDim = 10
GetDescCmd = IsNV5X
GetDataCmd0 = 00000023 00000000 00004014
GetDataCmd1 = 00000023 00000001 00004010
GetDataCmd2 = 00000022 00000002 00000000
GetDataCmd3 = 00000124 00000002 000000FF
GetDataCmd4 = 00000022 00000003 00000000
GetDataCmd5 = 00000424 00000003 00000008
GetDataCmd6 = 00000124 00000003 000000FF
GetDataCmd7 = 00000022 00000004 00000001
GetDataCmd8 = 00000424 00000004 00000010
GetDataCmd9 = 00000124 00000004 00000007
GetDataCmd10 = 00000014 00006978
GetDataCmd11 = 00000812 00000003
GetDataCmd12 = 00000912 00000002
GetDataCmd13 = 00000412 00000004
GetDataCmd14 = 00000914 000000FA

; MOV REG0 , GPU_4014
; MOV REG1 , GPU_4010
; MOV REG2 , REG0
; AND REG2 , FFh
; MOV REG3 , REG0
; SHR REG3 , 8
; AND REG3 , FFh
; MOV REG4 , REG1
; SHR REG4 , 16
; AND REG4 , 7
; MOV ACC , 27000
; MUL ACC , REG3
; DIV ACC , REG2
; SHR ACC , REG4
; DIV ACC , 250

;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;
;
; MPLL2 clock monitoring for NV5X compatible GPUs
;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;

[Source15]

Name = Memory clock \ domain 2
Dim = MHz
MaxLimit = 1500
MinLimit = 0
GridDim = 10
GetDescCmd = IsNV5X
GetDataCmd0 = 00000023 00000000 0000401C
GetDataCmd1 = 00000023 00000001 00004018
GetDataCmd2 = 00000022 00000002 00000000
GetDataCmd3 = 00000124 00000002 000000FF
GetDataCmd4 = 00000022 00000003 00000000
GetDataCmd5 = 00000424 00000003 00000008
GetDataCmd6 = 00000124 00000003 000000FF
GetDataCmd7 = 00000022 00000004 00000001
GetDataCmd8 = 00000424 00000004 00000010
GetDataCmd9 = 00000124 00000004 00000007
GetDataCmd10 = 00000014 00006978
GetDataCmd11 = 00000812 00000003
GetDataCmd12 = 00000912 00000002
GetDataCmd13 = 00000412 00000004
GetDataCmd14 = 00000914 000000FA

; MOV REG0 , GPU_401C
; MOV REG1 , GPU_4018
; MOV REG2 , REG0
; AND REG2 , FFh
; MOV REG3 , REG0
; SHR REG3 , 8
; AND REG3 , FFh
; MOV REG4 , REG1
; SHR REG4 , 16
; AND REG4 , 7
; MOV ACC , 27000
; MUL ACC , REG3
; DIV ACC , REG2
; SHR ACC , REG4
; DIV ACC , 250


Then go to monitoring -> Setup -> Plugins and enable GPUProbe.dll plugin. After doing that, you'll see Memory clock \ doman 0, Memory clock \ doman 1 and Memory clock \ doman 2 graphs reflecting state of each of 3 memory clock generators.

 
Imagine how much better these cards will be when we get some drivers that will actually run the three memory banks at full speed, right now mine are stuck at 792/396/396.

Imagine the card running all three banks at 950/950/950, should be a nice speed boost!
 
TheRapture said:
Imagine how much better these cards will be when we get some drivers that will actually run the three memory banks at full speed, right now mine are stuck at 792/396/396.

Imagine the card running all three banks at 950/950/950, should be a nice speed boost!
Click to expand...

Wait, our cards aren't running at full speed? Shiiii....
 
This bug didn't exists in the drivers before 97.44 did it? Pre 97.44 is the only time I oc'ed my mem, and it would artifact at anything about 1700
 
MrFace said:
This bug didn't exists in the drivers before 97.44 did it? Pre 97.44 is the only time I oc'ed my mem, and it would artifact at anything about 1700
Click to expand...

I think the official 97.02's were OK, according to the guy over at the rivatuner forums.

I am going to re-bench and overclock tonight with the 96.89's, I just reinstalled them at lunch today. Then I am going to compare numbers between the 96.89 and the 97.92's I just got rid of....



But yeah, most users of the 97.44 and up drivers, your mem is not running at full speed! Check to be sure!
 
wow.. I went back to 97.02 and benched it out and didn't see a diff.. maybe a point or two.. hrm.. maybe I should try the 96.89s?
 
revenant said:
wow.. I went back to 97.02 and benched it out and didn't see a diff.. maybe a point or two.. hrm.. maybe I should try the 96.89s?
Click to expand...



I am going to try them out tonight. As soon as I get home. The bug may not affect all of us, it might have something to do with the BIOS version the particular card uses...or it could be worse! :p

Have you checked the mem clocks using Rivatuner as specified in the above post?
 
From reading posts on guru3d, it sounds like you need to bench a game that uses a LOT of video memory. http://forums.guru3d.com//showthread.php?t=205416&page=4 Something like 3DMark06 normally uses less than 1/3 the video ram, so supposedly the fact that 2 out of the 3 memory clocks are wrong isn't noticable in that case.

I don't pretend to understand this. The 384 bit wide 768mb video memory must be organized as 3 independent banks of 128 bit wide 256mb memory, each with separate clocks. Remember that twelve ram chips are used on the board, each with a 32 bit data bus. It's hard to imagine that a game that used 256mb or less of video memory would only be using 1 bank of 128 wide 256mb memory, and the other 256 bits of the memory width were unused. If that were the case we'd essentially have the same performance as a 128 bit video card. Remember the nice performance boost we got when video cards moved from 128 bit to 256 bit. It just makes no sense to me that the 8800gtx is essentially operating in 128 bit mode when 256mb or less of video ram are used. We don't have 3 256mb banks of 384 bit wide ram, we have 3 banks of 128 bit ram working in parallel to get 384 bits.
 
TheRapture said:
I am going to try them out tonight. As soon as I get home. The bug may not affect all of us, it might have something to do with the BIOS version the particular card uses...or it could be worse! :p

Have you checked the mem clocks using Rivatuner as specified in the above post?
Click to expand...

yes - I saw the "bug" with 97.44 vs 97.02, but some were reporting this to be cosmetic only.. I think when I looked at the clocks for the 97.92s they looked normal again.. but I could be wrong.. I will have to check tonight!
 
Just got time to start some testing....with 96.89, my mem clocks do show properly and all three of them respond in unison. However, in my first run of Doom3, I get only 87fps at stock clocks. Overclocking the core to 625 and ram to 999 get me 98fps, which is still less than the stock clocks using the 97.92 drivers (108fps stock and 114fps overclocked core, respectively).

So at this point, at least for Doom3, the 97.92 drivers at STOCK CLOCKS get me MORE speed at the same settings, as the 96.89 drivers do when OVERCLOCKED. LARGE difference. Even if the 97.92's have the mem bug, they are faster. Of course this might just be better OpenGL drivers doing the work...

I have some more testing to do with other games like CoD2 and Source engines, more to follow later.
 
Keep us posted... I tried this out and mine is also showing only the first domain being changed on the memory clocks on my eVGA 8800GTS with the 97.92 driver set.
 
Signing in with memory bug with eVGA 8800gts running 97.44 drivers :(
 
How can you tell if you have the bug? I am overclocking mine and it reports what i have it set to in ntune. Is this related to the gtx or gts?
 
Astrodave said:
How can you tell if you have the bug? I am overclocking mine and it reports what i have it set to in ntune. Is this related to the gtx or gts?
Click to expand...

Both... you have to use RivaTuner to see all three clock domains with the info posted above. The software THINKS it's changing all three, but in reality, it's only changing the first of three, and thus giving little effect.
 
Is there any way to OC all three with the newest driver? I see mine is only increasing one of three after using the rivatuner plugin posted above. Clocking the core to 600 and leaving memory stock to see if that does anything :D
 
liveload said:
Is there any way to OC all three with the newest driver? I see mine is only increasing one of three after using the rivatuner plugin posted above. Clocking the core to 600 and leaving memory stock to see if that does anything :D
Click to expand...


You guys are lucky. With 97.44 or 97.92, I can't overclock even a single mem controller. At all. It stays at 792mhz, and the other two stuck at 396mhz.


With 96.89 I can overclock normally, but to be honest, I get slightly better benches even in REAL games ,with the newer drivers even with the mem clocks running slow, than I do with the 96.89's running way overclocked.
 
TheRapture said:
You guys are lucky. With 97.44 or 97.92, I can't overclock even a single mem controller. At all. It stays at 792mhz, and the other two stuck at 396mhz.


With 96.89 I can overclock normally, but to be honest, I get slightly better benches even in REAL games ,with the newer drivers even with the mem clocks running slow, than I do with the 96.89's running way overclocked.
Click to expand...

Odd, mine overclocks just the first domain, does that mean it's working properly (97.92 drivers)? I assumed you meant all three should be increasing... if so, woohoo ;)!
 
GoldenTiger said:
Odd, mine overclocks just the first domain, does that mean it's working properly (97.92 drivers)? I assumed you meant all three should be increasing... if so, woohoo ;)!
Click to expand...

No, with 96.89, when I overclock, all three domains overclock. With anything else higher, NONE of them will.
 
TheRapture said:
...It stays at 792mhz, and the other two stuck at 396mhz...
Click to expand...

Rivatuner reports the same thing to me ; however I can overclock one of the controllers. Last night I clocked the core to 600, left mem at defaults, and set the fan to 80%. Then I left the ATITool scanning for artifacts. It passed ~8 hours and was at 59ºc this morning. Runs ok in oblivion, but faster video memory makes a difference with respect to consistent performance (from what I saw, anyway).

Does clocking one out of three hurt the card, its performance, or its stability? I just don't know. If it doesn't then I don't see why we shouldn't, but I'd like to know more about this.;)
 
liveload said:
Rivatuner reports the same thing to me ; however I can overclock one of the controllers. Last night I clocked the core to 600, left mem at defaults, and set the fan to 80%. Then I left the ATITool scanning for artifacts. It passed ~8 hours and was at 59ºc this morning. Runs ok in oblivion, but faster video memory makes a difference with respect to consistent performance (from what I saw, anyway).

Does clocking one out of three hurt the card, its performance, or its stability? I just don't know. If it doesn't then I don't see why we shouldn't, but I'd like to know more about this.;)
Click to expand...

I don't see how it could hurt anything, but it would be good to have nVidia fix it so all three are overclocked :D !
 
Does anyone know if the new x100.xx drivers have fixed this? I don't think I will be DLing them untill the end of this month (Vista comes out on the 30th).
 
None of the overclocking software such as RivaTuner/nTune works in Vista as far as I know at this point.
 
okay, i am a total noob. does this mean if i'm using rivatuner to oc my 8800gts with 97.92, then my mem is really getting oc'd?
 
ericeash said:
okay, i am a total noob. does this mean if i'm using rivatuner to oc my 8800gts with 97.92, then my mem is really getting oc'd?
Click to expand...

Only one of the three memory domains is being overclocked, some people have NO mem overclock with them, it varies. All three are *supposed* to be changed, but aren't.
 
liveload said:
I wonder if this is something a registry hack could fix
Click to expand...

Wondering as well, although the driver seems to be the culprit. The card BIOS runs the mem at 400 during the boot process. Once windows loads, the driver takes over and is supposed to tell the card BIOS to boot the mem to spec. The newer drivers don't do that, and so the card sits at it's default boot speeds. I am unlucky enough that not even a single mem controller will overclock, so for now I am sticking with the 96.89 drivers. So far they are working fine,and hoping the next release will fix the issue. We are going to need the additional mem bandwidth once a couple high end games (like Crysis) hit.....
 
Hello All,
First time posting here

i read this thread with interest, after stumbling across the one at guru3d.

there's another thread at nvidia's forum. It has a response from nvidia that, in short, states that all's okay
http://forums.nvidia.com/index.php?showtopic=23958

from the post
"....We do not use RivaTuner but we have tried NVIDIA nTune which shows correct clocks for a eVGA GeForce 8800GTX card. Video and game performance is as expected from a product with 900Mhz memory. So maybe there is a problem with interop with 3rd party non NVIDIA utilities or something else is going on. However, if there is some problem with performance or functionality, please contact the appropriate manufacturer / vendor of your product... "


>>they dont say which driver version they were using<<

Maybe nvidia has changed the way memory clock is controlled, or this could be a bug after all.

I personally started on this driver (97.92), so cant comment on any performance difference there might be.
But reading this thread tells me that this is the driver to use for the g80 series of cards.

As for ppl who cant change any mem clock, the reason might be that they are using multi-monitor mode (with a single card).
This is mentioned in the release notes for 97.92.

regards
ali
 
lovely

nvidia says it's rivatuner's problem. rivatuner says it's nvidia's problem

*grumbles*

Would someone just own the damn thing already...it needs to be fixed. I would drop the whole thing and go away if I didn't have issues with all previous revisions.
 
alitayyab said:
Hello All,
First time posting here

there's another thread at nvidia's forum. It has a response from nvidia that, in short, states that all's okay
http://forums.nvidia.com/index.php?showtopic=23958

from the post
"....We do not use RivaTuner but we have tried NVIDIA nTune which shows correct clocks for a eVGA GeForce 8800GTX card. Video and game performance is as expected from a product with 900Mhz memory. So maybe there is a problem with interop with 3rd party non NVIDIA utilities or something else is going on. However, if there is some problem with performance or functionality, please contact the appropriate manufacturer / vendor of your product... "


>>they dont say which driver version they were using<<

Maybe nvidia has changed the way memory clock is controlled, or this could be a bug after all.

I personally started on this driver (97.92), so cant comment on any performance difference there might be.
But reading this thread tells me that this is the driver to use for the g80 series of cards.

[bold]As for ppl who cant change any mem clock, the reason might be that they are using multi-monitor mode (with a single card).
This is mentioned in the release notes for 97.92.[/bold]

regards
ali
Click to expand...



I am running a multi monitor setup off of my single GTS, as NV advertises. Should be NO issue there as the card is DESIGNED to run dual displays. Then comes the issue that 96.89 does NOT exhibit the problem, while 97.92 as well as 97.28 DO have the problem. That points BACK to a driver issue.

So by NV's reasoning, if I turn off my dual display, then install the 97.92 driver, and check it with Ntune, I SHOULD be able to see the correct clocks AND overclock the card. However, Ntune does not show ALL THREE memory clocks like Rivatuner can, thus, NV's method of verifying all three mem clocks is not very accurate.

The gist of my issue now is, I CANNOT overclock even a single mem controller from default speeds with anything above driver 96.89, while the other two are indeed stuck at half speed. Maybe my XFX card is having an issue, but I doubt since otherwise it is flawless, it does not seem to be brand specific which makes sense since all the initial cards were all made by the same company regardless of the sticker on the card.

I will be testing the multiple display thing tonight to see if turning off one of them affects the issue.
 
i also have a xfx 8800gts, and can atleast oc one of the mem domains.
i'm pretty sure its not card specific.

if this were really a problem then benchmark for atleast some ppl would certainly reflect so.

i mean playing at wide screen res is bound to use most of the available memory for frame buffer.

i have yet to come across anyone (nvnews, rage3d, hardforums) complaining of a decreased performance with the drivers that cause this to happen.

i ran 3dmark06 every time i raised my core/mem speeds. unfortunately i raised them at the same time so cant say for sure that raising memory speedhad no impact. i went from 80xx to about 8700 from stock to 585/900(?) at 3dmark's default settings.

@TheRapture: If u can oc one domain using a single monitor setup as mentioned in the release notes, then i guess nvidia did really change the way the ram clock are set or the previous set (of drivers) were buggy
 
On my card, using the 96.89 drivers, I get the largest increase in REAL gaming benchmarks by far, by raising the core speed. Memory makes small differences on the order of 2-4 fps.

These cards have LOTS of mem bandwidth, which might explain why the synthetic benches are not really showing any loss, if the mem is running less than it should be.
 
OK, with a single display enabled, and a fresh install of the 97.92 driver, here is my graph of the clock speeds...

stock_9792.jpg



Now, I can overclock the card using NTune, you can see that a single mem controller is indeed being overclocked, but the other two are stuck at half speed, which is what the card is SUPPOSED to do during the initial boot process....after Windows loads the driver is supposed to take over and set the clocks to 513/792 (stock clocks) but is not happening...

OC_9792.jpg



For me, 97.92 is still "broken" in that it only overclocks 1 of the 3 controllers, I am going back to my save restore point now and the 96.89 drivers, they are thus far working great for the games I play and I will continue to follow this issue or wait for the newer X drivers.....
 
Very interesting, could this be an end to this "problem" ??? I guess I really need to get nutty and do some serious in depth benching........that should show if those other two "un-used" clock generators really matter.....now I need a hard core bench that would suck up lots of vid card ram....
 
You must log in or register to reply here.
Back
Top