Incorrect GPU Memory report

[hexaae]

Hi, a picture is worth a thousand words:



As you can see the GPU memory reported for my GeForce 9600M GT is different from GPU-z...

BTW, can you add in the generic System Summary panel also the default GPU clock (e.g. 500/400/1250, clock/mem/shader) + computing infos (OpenCL, Cuda, PhysX, DirectCompute4.0...)?

[Martin]

I'm not sure who's right here, please try the nvidiaInspector tool to see what it displays.

[hexaae]

Martin Wrote:I'm not sure who's right here, please try the nvidiaInspector tool to see what it displays.

GPU-z and nVidia Inspector give the same result:

[Martin]

Could you please attach the HWiNFO32 Debug File (HWiNFO32.DBG) too?

[hexaae]

Martin Wrote:Could you please attach the HWiNFO32 Debug File (HWiNFO32.DBG) too?

Of course, attached.

[Martin]

Thanks! It's a bit strange, since the values from DBG do not give the results displayed.
Does HWiNFO32 report 100 MHz GPU Mem Clock too? If yes, what says GPU-Z or the inspector in this case?
Also, interesting would be to run HWiNFO32 without other tools running concurrently to see what memclk values are returned.

[hexaae]

Martin Wrote:Thanks! It's a bit strange, since the values from DBG do not give the results displayed.
Does HWiNFO32 report 100 MHz GPU Mem Clock too? If yes, what says GPU-Z or the inspector in this case?

Yes, in that case (almost idle) GPU-Z, nVidia Inspector and HWiNFO give the same results: 100MHz.

Quote:Also, interesting would be to run HWiNFO32 without other tools running concurrently to see what memclk values are returned.

I've tested this and is always incorrect... it's like there's a negative offset applied (real MHz - offset)...

[Martin]

I have implemented a small update and will release a new Beta in few hours. But I'm not sure if that will help..
In case it won't fix it, I'll need all possible GPU MemClk values it returns along with those which are actually valid.

[hexaae]

Martin Wrote:I have implemented a small update and will release a new Beta in few hours. But I'm not sure if that will help..
In case it won't fix it, I'll need all possible GPU MemClk values it returns along with those which are actually valid.

Tested latest beta but didn't solve:



Please, tell me exactly how to generate the trace/log for you...

[Martin]

One more idea.. Could you please try AIDA64 if that works properly ?

[hexaae]

Martin Wrote:One more idea.. Could you please try AIDA64 if that works properly ?

I'm afraid it would give the same results of GPU-z, SIW, nVidia Inspector... I don't like the idea to install the trial + remove of AIDA64....

[Martin]

How could you know it would give the same result as GPU-Z?
AFAIK, AIDA64 uses a similar method as HWiNFO32 where on the other hand the nVidia Inspector, SIW (and GPU-Z probably too) use NVAPI.
So I wanted to make sure if there's a bug in my implementation, or for this case I should rather use NVAPI.
You also don't need to install the AIDA64, just unpack the portable ZIP edition, run and then you can delete it.

hexaae Wrote:

Martin Wrote:One more idea.. Could you please try AIDA64 if that works properly ?

I'm afraid it would give the same results of GPU-z, SIW, nVidia Inspector... I don't like the idea to install the trial + remove of AIDA64....

[hexaae]

Martin Wrote:How could you know it would give the same result as GPU-Z?
AFAIK, AIDA64 uses a similar method as HWiNFO32 where on the other hand the nVidia Inspector, SIW (and GPU-Z probably too) use NVAPI.
So I wanted to make sure if there's a bug in my implementation, or for this case I should rather use NVAPI.
You also don't need to install the AIDA64, just unpack the portable ZIP edition, run and then you can delete it.

hexaae Wrote:

Martin Wrote:One more idea.. Could you please try AIDA64 if that works properly ?

I'm afraid it would give the same results of GPU-z, SIW, nVidia Inspector... I don't like the idea to install the trial + remove of AIDA64....

Ok, if you say it uses another method... here is a sshot:

[Martin]

I made more modifications to the method, please try this one: <!-- w --><a class="postlink" href="http://www.hwinfo.com/beta/hw32_366_1088.zip">www.hwinfo.com/beta/hw32_366_1088.zip</a><!-- w -->
If that still doesn't work, please create a new Debug File while running across different states (and note which values have been displayed and which ones should be correct).
BTW, when you run HWiNFO32 without Debug Mode (standard), does it report incorrect too?

[hexaae]

Martin Wrote:I made more modifications to the method, please try this one: <!-- m --><a class="postlink" href="http://www.hwinfo.com/beta/hw32_366_1088.zip">http://www.hwinfo.com/beta/hw32_366_1088.zip</a><!-- m -->
If that still doesn't work, please create a new Debug File while running across different states (and note which values have been displayed and which ones should be correct).

The values are those reported in the previous post in the sshot by AIDA64 (Level #1÷4) since also GPU-z reports the same things...



Attached the Debug file.

Quote:BTW, when you run HWiNFO32 without Debug Mode (standard), does it report incorrect too?

Yes always incorrect except when 100MHz.

[Martin]

Thanks for the extensive report. I'm analysing the data and it seems this card uses a custom Mem reference clock and that's the reason (27 MHz instead of 25 MHz default).
I'll let you know when I have a solution..

[Martin]

Please try this this build: <!-- w --><a class="postlink" href="http://www.hwinfo.com/beta/hw32_366_1089.zip">www.hwinfo.com/beta/hw32_366_1089.zip</a><!-- w -->
and let me know how it works.

[hexaae]

Martin Wrote:Please try this this build: <!-- m --><a class="postlink" href="http://www.hwinfo.com/beta/hw32_366_1089.zip">http://www.hwinfo.com/beta/hw32_366_1089.zip</a><!-- m -->
and let me know how it works.

Solved!
All clock levels are compliant to the other monitoring tools...

[Martin]

Thanks for the feedback