Discussion in 'Videocards - NVIDIA GeForce' started by Milan, Jan 7, 2015.
Can u shrink that screenshot to 1280x1024 or use
No antialiasing is used on the second video. Plus it's a SLI.
You're not speaking for all nvidia customers.
OK I looked in systernals and there is nothing on GPU usage, I must have an old version, however I'm pretty sure I know the answer.
GPU memory is either Dedicated (onboard) or Dynamic (system).
The dedicated amount in the graph is the total amount of onboard vram in use.
Committed memory is a subset of vram in use which is unavailable for other purposes, so it can't be shared. Just like process memory usage.
It's a bit of red herring imho as I don't think it matters.
What matters is how much onboard vram is used, so that's the dedicated amount.
I hope that answers it.
You're on my ignore list until the banhammer smashes you out of here. Don't bother.
@reputable guru 3d ancient users.
Those with 970 can you replicate the "problem" that new members have as of January 2015?
big guns out now folks.
In most cases you will not gain anything or even notice anything. Many of the newer game engines manage memory efficiently, so there is hardly any noticeable difference between a 3GB and a 4GB video card in the same game even if the latter used up all its memory. With more vRAM available the game can already plan ahead and load data into the vRAM that is anticipated to be needed later. With lower amounts of vRAM the game has less headroom to plan ahead, but this does not necessarily mean the game will run bad on a card with less vRAM. But the chance increases of more stutters appearing the lower vRAM your video card has available. Also there is of course a minimum amount of data that is required to be in the vRAM at all times. If your video card cannot load that then your game might lock up or even crash.
So the issue with a lack of vRAM is about intermittent spikes in frame time which causes stuttering (low minimum fps). This happens when your video card needs data that is not currently loaded into the video RAM. I've explained it in more detail here
This is perfectly demonstrated in this Skyrim video posted by Headd: http://goo.gl/UFTdgk
A game like modded Skyrim needs to have the texture data available at all times! This means that when the data needed to display the rest of the scene is not available in the vRAM, another portion of the vRAM is unloaded, and the freed space is used to load the required data to continue rendering the scene. This time spent on loading and unloading the data causes the stuttering you see in the video.
In this case we wonder why the 970GTX is not using its remaining vRAM. If the card used its remaining ~500MB we would see less or possibly even no stuttering issues anymore!
I really think that the commit amount matters. It's the size of MB that the card should try to fit in it's dedicated memory. Since it can't fill over 3200mb dedicated the missing ~800 MB gets streamed from system ram.
The frames im kombuster drops about 2/3 cause of the failed allocation managing of commited gpu memory.
Please use the Process Explorer and look how much memory gets dedicated GPU memory.
As i mentioned before afterburner do not show the dedicated memory usage of your gpu.
3,5GB ram use, game crash due to lack of memory. PF off
It's Committed bytes in Dedicated memory, as I said.
I didn't say anything remotely close. Please quote me.
You were looking at WS and private bytes in TM to measure process memory use which is only a small part of the allocated amount.
What I said was look at process virtual size to know the real memory allocation of a given process.
As for your 10GB I don't know what you mean.....
Ok i was wrong with the statement about gpu-z.
It shows the full dedicated usage of memory.
That would mean that kombustor tries to allocate 64GB of memory.
Lol, yeah looks like it...but this is getting off topic. Your GPU is working fine.
Actually on second thoughts, who cares about the thread topic right?
Maxwell uses a completely different (more efficient) data compression method than Kepler. You can NOT compare the 2 based on VRAM usage. Maxwell can store more data in VRAM than Kepler could because Maxwell compresses the data considerably better. The exact same thing occured with the Tesla cards and Fermi cards, as well as the Fermi and Kepler cards. NVidia has improved data compression with each generation. They add more VRAM because they know it's going to become necessary down the road, even with more advanced data compression.
I've monitored memory usage across a few games and I'm not seeing where there's any issue with memory usage at all. Maxwell just appears to be doing a damn good job at compressing the data, just like it was designed to do.
And just like Furmark, Kombuster has "issues" with all post-Tesla cards.... Tesla being the GeForce 200 series cards.
Then explain why in 1920x1080 GTX980 use all 4GB Vram and GTX970 dont.
GTX980 reach 4GB Vram even before loading is finished...
It's just a driver issue that will be fixed soon.
Thank you for your answer sykozis, but this indeed doesn't explain the difference between the 980GTX and the 970GTX, which have exactly the same memory layout and architecture.