Dedicated GPU Memory Usage Discrepancy

[Aquinus]

Howdy @W1zzard . I've been doing a ton of debugging of Elite: Dangerous to help FD figure out why there is a slow down in super cruise mode and I came across this little gem. Both MSI AfterBurner and ProcessExplorer are reporting the same kinds of numbers but, GPU-Z seems to think that a lot more GPU memory is being used. I think that both numbers here may be correct but reading different sources. I was wondering if you have any idea why these two numbers would be so vastly different if they're supposedly measuring the same thing. Please ignore that blue circle, that's for the guys over at FD. The red circles indicate what I'm describing. I know that ProcessExplorer is measuring just the application but at idle I'm using ~500MB according to GPU-Z, so if we just factor that out, there is still a pretty big difference.

 

[Aquinus]

@W1zzard : Ping?

 

[Aquinus]

@W1zzard : Are you going to give me an answer or continue ignoring me?

 

[Toothless]

@W1zzard : Are you going to give me an answer or continue ignoring me?

That's rude. He's a busy man running the site and has other things like life to work with. Maybe PM him and be a decent person about it?

 

[Aquinus]

That's rude. He's a busy man running the site and has other things like life to work with. Maybe PM him and be a decent person about it?

I'm not looking for a dissertation, I'm looking for a quick answer at minimum. In all seriousness, it's a legitimate question. Also, others have asked GPU-Z questions between when I posted the first post and now and they got responses and I haven't. In fact there is more than just one. It's not even a bug, just a question on how GPU-Z operates. I just want an answer and it's not like I'm looking for him to spend more than 5 minutes to give a reply unless there is really something wrong, which I doubt.

...and @Toothless , keep your self-righteous comments to yourself. It's not like your contributing to the thread by commenting on my attitude. I'm not in the mood for this kind of BS.

 

[R-T-B]

I'll post this in the GPU-Z beta thread where Wiz may read it.

 

[W1zzard]

I have no idea what's going on. Wanted to look into it before going on vacation, but didn't find time for it. Back on Dec 13

 

[Toothless]

I have no idea what's going on. Wanted to look into it before going on vacation, but didn't find time for it. Back on Dec 13

Don't worry about us and enjoy your vacation! Everyone knows you earned it.

 

[Aquinus]

I have no idea what's going on. Wanted to look into it before going on vacation, but didn't find time for it. Back on Dec 13

Thanks W1zz. That's all I was looking for. Have a good vacation. Eat, drink, and relax.

 

[AsRock]

What does task manager say the usage is ?, maybe one is a combination of system ram and graphic card ram combined.

 

[Aquinus]

What does task manager say the usage is ?, maybe one is a combination of system ram and graphic card ram combined.

I was thinking that as well. Task Manager doesn't display GPU memory usage which is why I was using both Afterburner and ProcessExplorer to monitor it, as well as GPU-Z. Afterburner seems to be more in line with what ProcessExplorer is saying however, there are occasions where Elite: Dangerous spikes GPU memory usage on Afterburner and ProcessExplorer, then slowly working their way back down to say, 700-1000MB as the Super Cruise bug starts to happen whereas GPU-Z shows a steady increase in GPU memory usage.

I'm thinking that the difference may be active GPU memory being used. My hunch is that GPU-Z is indicating the full pool of all resources allocated to VRAM whereas Afterburner and ProcessExplorer might be only showing memory that's actually loaded into the GPU. It seems to be the only thing other than excessive CPU usage on a single thread that I could narrow down for this super cruise bug.

For what it's worth, I do want to apologize for my attitude, I'm getting beyond frustrated with this entire issue between FD and AMD that it's making me go out of my way to figure out what is going on because I want to play the game smoothly again. I did after all pay for it and I also want to get Horizons but, that's not going to happen until this damn bug is fixed.

I think that the explanation of the difference between the two may shed some light on what's going on. For what it's worth, someone showed some screenshots how this bug doesn't really seem to exist in Crossfire with the 290 but does when CFX is disabled which adds a weird dynamic to the problem.

 

[AsRock]

You tried their forums to see what they had to say about it ?.

 

[Aquinus]

You tried their forums to see what they had to say about it ?.

I'm waiting on a reply from AMD on this matter. I've been trying to communicate with Frontier about their game but their moderators have a real stick up their ass. I did a run around them and went straight to Brett C. who oversees the forum community and he acknowledged that this issue is annoying. He was hoping to get an official response out to the public in a couple weeks (it has almost been a couple weeks,) so needless to say, Frontier has been 100% useless.

This is the chaos thread: https://forums.frontier.co.uk/showthread.php?t=170946
This is my bug report: https://forums.frontier.co.uk/showthread.php?t=203508
This was my post to AMD: https://community.amd.com/thread/189159?q=elite#2689872

Edit: For what it's worth, a game like Cities: Skylines GPU memory usage seems to almost match (minus 200MB or so for the difference that I stated in the OP.) I have to test more games but, I suspect that Elite: Dangerous probably will be the only game with the behavior I described above. I asked W1zz what the difference could be because it could indicate an issue with something like streaming textures or offloading unused GPU memory.

 

[The N]

It seems a serious bug in memory usage monitoring. both gpuz and task manager contrary to results. Let us know if you received any worthy statement response from other sites.

 

[Aquinus]

It seems a serious bug in memory usage monitoring. both gpuz and task manager contrary to results. Let us know if you received any worthy statement response from other sites.

I actually think I can explain... partially. It seems that Afterburner and ProcessExplorer are losing track of some memory. There is another (albeit hard to find,) section in ProcessExplorer for system wide information which perfectly matches GPU-Z. So I don't think it's a bug, as I said earlier. It does seem though that a chunk of GPU memory still being used, is being disassociated with the application. Check this screenshot out.

I'm not sure why this is happening but, it could be AMD's driver offloading unused VRAM from the GPU so memory doesn't get wasted. It's rather interesting because having 8GB of VRAM means I have that much more that can be wasted.

Edit: I'm going to do more investigation later but I have to run out and have lunch with the family. W1zz probably doesn't need to get back to me because at this point, I think we can say that GPU-Z is keeping track of system wide memory and that it is AB and PE that are losing track of this extra memory.

 

[The N]

more of software. could be AMD driver, process explorer and GPUz at above perfectly allied. you cannot exclude window 10 factor as well.

 

[Aquinus]

more of software. could be AMD driver, process explorer and GPUz at above perfectly allied. you cannot exclude window 10 factor as well.

Very true. WDDM 2.0 has much more intelligent VRAM management which will keep only what's being used in active VRAM from what I've read. That very well could explain what is going on as well.