Z
ZombieFred
Unconfirmed Member
Any good?
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
GeForce GTX 970s seem to have an issue using all 4GB of VRAM, Nvidia looking into it
- Thread starter Naked Snake
- Start date
- Status
There are two sides to this.
The card should definitely not create slowdown problems at certain times or other weird behavior. It's indeed a highly valid topic, and Nvidia should absolutely answer the question.
With that said, no need to panic, your 970 is just as fast as it was yesterday, and keep in mind that the 970 is not supposed to be as fast as the 980.
..And if you are into VR and are going to buy the Oculus Rift later this year, you probably need a new generation GPU to drive it anyway
The card should definitely not create slowdown problems at certain times or other weird behavior. It's indeed a highly valid topic, and Nvidia should absolutely answer the question.
With that said, no need to panic, your 970 is just as fast as it was yesterday, and keep in mind that the 970 is not supposed to be as fast as the 980.
..And if you are into VR and are going to buy the Oculus Rift later this year, you probably need a new generation GPU to drive it anyway
vandalhearts
Member
Gigabyte G1 Gaming 970
I'm running it into Windows and, ye, the gap seems legitimately taken by OS usage, as stated by MSI Afterbuner in my [updated] picture.
RurouniZel
Asks questions so Ezalc doesn't have to
Samsung memory. In the clear?
Course drivers may crash when windows is rendered in the VRAM...
Too much misinformation I'll have to run my own tests I think.
Just trying to share information to see if it can help solving this mystery. I don't know how you're supposed to run this test without windows.
michaelius
Banned
Connect pc to IGP in your i3/i5/i7 cpu
Not without windows but running off the integrated GPU, whilst testing the GTX 970 with 0mb VRAM allocated.
Objectively Bad Opinion
Member
You can use GPUs headless, and tunnel into the machine from another. Or have Windows boot into another GPU in the system and leave the 970 idle.
Pie and Beans
Look for me on the local news, I'll be the guy arrested for trying to burn down a Nintendo exec's house.
Was a week or so away from ordering one and a new rig to prepare for Oculus sometime this year (longterm plan being getting another to SLI down the road) so I guess... NOW WE WAIT?
Horsemama1956
Member
Uhh, it wouldn't be considered a problem if this was the case.
Doesn't have to be a VRAM issue, but something else on the card.
Dictator93
Member
It could still be hardware related (memory bus, controller, etc.)
Or it could be by secret design so Nvidia can force people into thinking the need to upgrade.
tinfoil hat engage!
TAJ
Darkness cannot drive out darkness; only light can do that. Hate cannot drive out hate; only love can do that.
Xdrive05
Member
The bigger news sites need to report this.
If this ends up being an unfixable hardware design issue, then #1) I don't for one second believe Nvidia didn't already know about this and #2) that may be partially why it was priced so much lower than its slightly bigger brother. Scummy if true.
And very interesting that none of the big review sites reported this. I wonder why.
If there's enough backlash, then hopefully they do something about it. Drop the 980 price plz.
If this ends up being an unfixable hardware design issue, then #1) I don't for one second believe Nvidia didn't already know about this and #2) that may be partially why it was priced so much lower than its slightly bigger brother. Scummy if true.
And very interesting that none of the big review sites reported this. I wonder why.
If there's enough backlash, then hopefully they do something about it. Drop the 980 price plz.
Dreams-Visions
Member
So there's no reason to run the benchmark if you can't stop Windows from rendering on the GPU? I don't have an integrated GPU to switch to for testing.
I guess I'll keep an eye on this and maybe use my Step Up option to get a 980 or something. Or, I'll certainly consider the option.
I guess I'll keep an eye on this and maybe use my Step Up option to get a 980 or something. Or, I'll certainly consider the option.
AHA-Lambda
Member
Just heard about this, anyway to test if my 970 is affected?
Oh crap, my friend is building a gaming rig for the first time and I knew he was getting a 970.
I just text him about this issue hoping to stop him before he purchased one, but his 970 just happened to have arrived at his door an hour before I sent the text.
He isn't too happy lol.
I just text him about this issue hoping to stop him before he purchased one, but his 970 just happened to have arrived at his door an hour before I sent the text.
He isn't too happy lol.
Baron_Calamity
Member
I almost ditched my 3gb Radeon 7950 for a 4gb GTX970. Glad I didn't. Yes the GTX970 is faster, but man, to spend the money and have this issue showed up would have really ticked me off. See, I went from an Nvidia gtx 660 to the AMD card because of a similar issue. The GTX660 has 2 gigs but only 1.5 was useable and the bus was weird also that caused it to struggle over 1 gig of memory used. I kinda understand Nvidia making this mistake once, but twice? It sounds like they think if you aren't buying their top of the line card, you must be some sort of idiot that only cares about specs and not real world performance.
I have ran the bench on my MSI Gamer GTX 970 4GB. I have only had time to kill DWM.exe and have not run off IGP yet. MSI AB showed 290MB VRAM in use after which I exited the program and ran Nai's bench from cmd. Last 3 chunks showed degraded performance. When I get time I'll try run off the IGP for proper headless mode.
SliChillax
Member
So should I try and get a refund or just wait? Any word from Nvidia? Sorry if my questions have already been answered but I'm at school at the moment.
Terbinator
Member
They're all affected.
This can't take anything other than a BIOS upgrade/flash though, I'd imagine.
ashecitism
Member
No word from Nvidia yet.
Xdrive05
Member
If you already own one, you should probably wait and see what they say or do about it. They may be able to fix it somehow, or do something else to make it right.
Or they may release a PR-speak nightmare claiming that "it is the way it is, and the card is still a great value for the money, and thanks fuckers for that money, and have you even SEEN fur shading!?, and by the way fuck you."
Best to wait and see before making decisions right now.
Dreams-Visions
Member
contributing. not connecting to IGP or whatever.
stock 970 FTW
It won't run, requires CUDA.
Oh I see thanks. Over in the other thread, there's a bench of the GTX980 without Windows OS overhead getting maximum performance up to the last chunk at 4GB. It doesn't look good, that's for sure.
Yeah, so far I haven't seen anyone get the 970 with proper results like a 980, even those who are using headless mode. The 980s who fail to get the right result seem to not be using headless mode.
It could just be the benchmark so we have to wait.
From a trouble shooting perspective, we have to consider that the benchmark could be wrong. Perhaps previous tests before getting to the final chunks are having a knock on effect that makes this phenomenon visible.
If the source code for the benchmark is available, I would like to see if someone could edit it so it only to test Chunk no 29. As in every case this chunk always seems to suffer. Anyone here who can help with this?
If the source code for the benchmark is available, I would like to see if someone could edit it so it only to test Chunk no 29. As in every case this chunk always seems to suffer. Anyone here who can help with this?
Naked Snake
Member
Everyone should at the very least
Also, I second the request for source code of the benchmark, I'd like to see how it decides where an allocated chunk is in physical memory, because I don't know a reliable way to do that.
- disable Shadowplay and
- make note of their prior windows/application VRAM usage levels
Also, I second the request for source code of the benchmark, I'd like to see how it decides where an allocated chunk is in physical memory, because I don't know a reliable way to do that.
Pretty sure it was known fairly early on. I remember it being mentioned in the reviews.
- Status