• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

GPU-Z incorrectly shows Dedicated Memory Usage

Joined
May 29, 2016
Messages
17 (0.01/day)
Processor Intel Xeon E5-1680 v2 @ Stock
Motherboard ASUS P9X79 WS/IPMI
Cooling Thermalright IFX-14
Memory 64GB DDR3L-1600 ECC Reg (8 x 8GB Hynix HMT41GR7BFR8A-PB)
Video Card(s) Gigabyte R9 290X MBA (modded to 8GB VRAM Elpida 4032BABG + modded pure Copper heatsink)
Storage Intel 750 400GB AIC + 2 x 1TB WD RE4 Intel RAID0+1
Display(s) 32" BenQ BL3200 (2560 x 1440 via DisplayPort)
Case Antec P182
Audio Device(s) Int. ALC1150
Power Supply Corsair AX760
Mouse Logitech MX Master 2S
Keyboard Microsoft Curve 2000
Software Windows 8.1 Professional 64-bit
I have quite a unique R9 290X card, it had passed the memory replacement procedure, from 4GB Hynix AFR to 8GB Elpida BABG.
After that, and after all proper vBIOS modifications (there's a long Hawaii BIOS Editing thread at overclock.net) the GPU-Z is the only program that incorrectly detects the memory buffer loading.

For example, I run MSI Kombustor GPU Memory burner 6GB test, and the reported memory load increases linearly from 2D value to ~4000MB (see the screenshot) and then, resets to 0MB again and continues counting on. So, according to GPU-Z I'm running the test with ~2300MB, while in fact the load equals ~6400MB.
The same problem is observed in different apps requiring more that 4096MB of VRAM.

I want to note that GPU-Z detects the memory properly (see the screenshot), 8GB Elpida GDDR5.

I tried some other monitoring tools (AIDA64, HWInfo64), they correctly report the VRAM loading.
It seems to me there's some hard-coding in the program code, where 290X is assumed to be a 4GB card, while it's not always true...
memory_overlap.gif memory_size.gif
 
Have u tried any of the previous versions of GPU-Z to see if they shows the correct amount of vram for ur R9 290X ?

and as said before plz fill out ur System Specs it helps also in the future
 
Well, have seen that bug since version 0.8.6 till the current 0.8.8.
I think it's an architectural bug, not the problem of any particular program build.

...Now specially tried version 0.7.4, as the first with official 290X support. The problem confirmed again, but the sensors behave a bit different, Memory Usage (Dedicted) stays at 0MB, while the sensor called Memory Usage (Dynamic) shows the behavior described above (rises to ~4000MB, falls to 0MB and rises again to ~2300MB).
So, nothing different in fact.
 
I assumed this to be a bug in the driver, it overflows when passing 4 GB, because it's internally represented as 32-bit.

But since you report that it works in other tools on same system with same drivers, I'll look at this again, must be a GPU-Z bug.
 
you seriously replaced the RAM on the card?

that's pretty hardcore
 
I don't think he did it himself, he had it done.

Aye it requires expensive tools and knowhow. The T-862 reballer can be had for pretty cheap (at least some years ago), but afaik it's pretty crappy and it's quite small. (I found one in Sweden some years ago for about €200-300 I think, but after reading up on it I decided it wasn't worth it) If you're serious you have to spend a lot more.
 
Pretty much you can ask any level3 mobile phone repar center to do it... daily job. Just a bit scary setting up the profiles and temperatures. As GPU's are heck of a good heat spreaders, and designed to be so.
 
For everyone who doesn't believe that memory chips have been replaced in fact, welcome here:
http://www.overclock.net/t/1561372/hawaii-bios-editing-290-290x-295x2-390-390x/2540#post_25123176

Yes, some specific equipment is needed, ERSA infra-red heater station for BGA solder/unsolder, and Lukey 852D+ for re-balling. And the balls, certainly, 0.5mm in diameter.
In fact, my card is not the only one that had been modified, the second one is 290 (non-X).

Everything is real. Hardcore only. :D
 
I assumed this to be a bug in the driver, it overflows when passing 4 GB, because it's internally represented as 32-bit.

But since you report that it works in other tools on same system with same drivers, I'll look at this again, must be a GPU-Z bug.
Is there any progress regarding the issue, I wonder? Has the bug been confirmed?
 
I assumed this to be a bug in the driver, it overflows when passing 4 GB, because it's internally represented as 32-bit.

But since you report that it works in other tools on same system with same drivers, I'll look at this again, must be a GPU-Z bug.
This does seem similar to the thread i posted quite a long time ago
http://www.techpowerup.com/forums/threads/memory-usage-counter-resets-above-4gb-r9-390.220697/
It does seem to be that GPU-Z does not report correctly on memory levels much above 4GB...
 
Same happens in Unigine Valley and Heaven benchmarks for 8Gb cards, they report as 4Gb, particularly the 980Ti's.
 
What does GPU Caps say?
 
For everyone who doesn't believe that memory chips have been replaced in fact, welcome here:
http://www.overclock.net/t/1561372/hawaii-bios-editing-290-290x-295x2-390-390x/2540#post_25123176

Yes, some specific equipment is needed, ERSA infra-red heater station for BGA solder/unsolder, and Lukey 852D+ for re-balling. And the balls, certainly, 0.5mm in diameter.
In fact, my card is not the only one that had been modified, the second one is 290 (non-X).

Everything is real. Hardcore only. :D

Confusing part about the links is that he swapped out Hynix for Elpedia.
 
It is called Elpida: https://en.wikipedia.org/wiki/Elpida_Memory

Why is this confusing? Do you think a videocard cannot handle different RAM types/brands?
I suppose this disappointment is from the mining point of view, but mining wasn't important in my case.

In fact it's not so easy to get 4Gbit GDDR5 chips for a reasonable price. I can say Micron AJR was nearly 3x times more expensive than Elpida 4032, and there's no sense in such an upgrade, it's easer to sell 290X and simply buy 390/390X instead.
 
I suppose this disappointment is from the mining point of view, but mining wasn't important in my case.

In fact it's not so easy to get 4Gbit GDDR5 chips for a reasonable price. I can say Micron AJR was nearly 3x times more expensive than Elpida 4032, and there's no sense in such an upgrade, it's easer to sell 290X and simply buy 390/390X instead.
Was not aware different brands/models of video RAM cause different mining speeds, thanks!
OK, what is your use case, then? Work (professional software) or play (gaming)?
 
Was not aware different brands/models of video RAM cause different mining speeds, thanks!
OK, what is your use case, then? Work (professional software) or play (gaming)?
The card is used at home, and the reason is different nVIDIA-optimized games that sometimes require more that 4GB of VRAM (in order to be well-played only on 980ti/titan x).

But after the upgrade, the card is still actual just because the 1080 is also equipped with 8GB :)

There was another point, being a HW and FW enginer it's quite interesting to make something unusial with the device/BIOS, make some tuning, etc.
 
It's common knowledge that Elpida memory does not overclock as well as Micron or Samsung chips.
OK - thanks!
I do not mine, but AFAIK there is not much use in overclocking the video card's RAM or even the core, if you are mining.
When mining, the best possible stability is desired and sought, not top speeds. :)
In fact sometimes a slight underclock and/or undervoltage of the videocard is done, in order to lower temps.
 
It is called Elpida: https://en.wikipedia.org/wiki/Elpida_Memory

Why is this confusing? Do you think a videocard cannot handle different RAM types/brands?

O no i spelt it wrong the Internets crashing down upon us, get over it.

OK - thanks!
I do not mine, but AFAIK there is not much use in overclocking the video card's RAM or even the core, if you are mining.
When mining, the best possible stability is desired and sought, not top speeds. :)
In fact sometimes a slight underclock and/or undervoltage of the videocard is done, in order to lower temps.

Aah now it makes sense why he got those chips.
 
Problem found, fix in next release
 
That's great! :)
Maybe, some testing/confirmation required before the release? I could do that.
 
Invited you to the beta testing private message thread, first post has the latest build with the changes for memory monitoring
 
Back
Top