NVidia GTX970 - Memory Allocation Bug Found

[lvh1]

Ah, so this is why performance starts tanking at around 3200 MB vram in use... I guess they can fix it with a driver update some day in the future. 

[Uphorics]

Just tested mine, starts to drop around 3400MB...

 

Here's the thing though, this really shouldn't affect a lot of people. In order to max out VRAM you need a pretty high resolution and a ton of AA enabled to hit that last 1GB - 500MB.

[Astorrin]

Just tested mine, starts to drop around 3400MB...

 

Here's the thing though, this really shouldn't affect a lot of people. In order to max out VRAM you need a pretty high resolution and a ton of AA enabled to hit that last 1GB - 500MB.

Try Middle Earth Shadow of Mordor or new consol portet games, in 1080 that most people use, and if windows use 400mb vram, not much is left for games, on higher graphics.

[Uphorics]

Try Middle Earth Shadow of Mordor or new consol portet games, in 1080 that most people use.

 

That's just some games (like I said it SHOULDN'T affect A LOT of people) Other AAA titles like BF4 don't use that much VRAM at 1080p.

[Astorrin]

Still a problem, if you want to play far cry 4, hope it will be fixed soon.

[AmbellinaThePrise]

Had a chat with Nvidia support earlier and they said they are aware of the issue, and they claim it to be an issue with the drivers/firmware. They said they are going to be working on a fix for it. Feel free to take this with a huge grain of salt, because I am an idiot and didn't think to nab a screenshot of the chat.

 

Quick edit: Doesn't matter that it may not affect every person with the card, the issue is that it affects every card, and the owners aren't getting what they paid for.

[Uphorics]

One more thing, we also have to think how many people actually HAVE a GTX 970  Spoiler: (not very many)

[AmbellinaThePrise]

Of course, there are not very many who own the card out there right now. So it's not affecting a massive amount of people, so if it is a hardware issue, Nvidia will be more than able to offer some form of replacement or reimbursement for those who do. I had one picked out for my build, but with this knowledge, i'm going to wait until the issue is resolved, or splurge for the seemingly unaffected 980.

[Astorrin]

Check how many that got this problem with they own 970, there is a lot that got this problem.

 

Google /hint

[TheGosuStandard]

Still a problem, if you want to play far cry 4, hope it will be fixed soon.

 

I've played far cry 4 on ultra settings not sure what I am looking for. 

[Astorrin]

I've played far cry 4 on ultra settings not sure what I am looking for. 

I play it fine to, the thing to check is your VRAM, some people have low fps because they hit that bug with ~3.5GB VRAM, like this is all about lol, some hit 3.2GB VRAM and have 10-20 fps.

[Uphorics]

Check how many that got this problem with they own 970, there is a lot that got this problem.

 

Google /hint

 

...

 

Everybody has this issue with the 970...

[AmbellinaThePrise]

I've played far cry 4 on ultra settings not sure what I am looking for. 

Apparently stuttering and quality degradation when you're using anywhere between 3-3.5GB VRAM is what to look for. But it's not unreasonable to say that your card could be one of the rare few i've seen that are noting that their card is not having this issue.

[Astorrin]

I turned off Shader Cache and Edit some config files, yeeh for ubisoft lol, but most people that hit that limit VRAM have bigger problems, not all want to play with low settings.

[Guest]

A lot of this is knee jerk reactions fuelled by wild speculation. A bunch of people posting 'benchmark' results everywhere, but with no consistent benchmarking procedures, and very few examples of issues actually being game effecting.

 

I'd wait for some some well known hardware/benchmarking sites to properly examine the issue before running out into the street and screaming the sky is falling...

 

Its certainly possible that there's an issue, but I'm definitely loling at all the people raving about how there is going to be a recall, or free gtx 980 upgrades already.

[SteveGrabowski]

EDIT: Nevermind, I'll wait to hear from Nvidia

[Spev]

Wow....this is really interesting I'm going to follow this...luckily I don't have to worry I guess, I used to have a 970 but I have a 980 now. I will sleep easy tonight!

[Dnoble24]

I don't think Valley will use >3.5GB of VRAM.

 

Well, this issue, combined with the 970 downclocking too much *rumor* is what brings me to these feelings. 

 

I mean, Amazing power consumption to performance ratio at good price points doesn't happen by magic. It seems some corners were cut in a few areas for the 970 is all i'm saying. 

[NutzInTheHead]

I got 2 of them MSI gaming 970s back in november of last year. but I can't test this as my monitor is on warranty repair. I will investigate this on my cards once I get my monitor back early next week.

I really hope nvidia does something about this. I didn't exactly win the lottery. I worked hard for these cards.

[SteveGrabowski]

Nai's benchmark sounds like bullshit

 

http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

[Guest]

Nai's benchmark sounds like bullshit

 

http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

might not even be a problem with the benchmark itself per say, rather it seems to be more of an issue with a bunch of noobs that don't know how to benchmark properly running around posting screenshots of this benchmark everywhere, claiming that its 'confirming' the issue.

 

The problem is, that no one is running the benchmark in a consistent manner, and most of the screenshots that are being posted don't state the conditions in which the benchmark is being run.

 

Some are running it with aero enabled, some are running with aero disabled, some are running it off the igp with the dedicated gpu in headless mode (and the latter is supposed to be the proper way to run the benchmark for accurate results). Running this benchmark on a gtx 980 with aero enabled seems to 'confirm' the same issue, memory performance trailing off after 3500mb, where running it in headless mode does not:

 

http://www.reddit.com/r/pcgaming/comments/2tiifp/used_nais_benchmark_on_a_gtx_980_to_test_the_gtx/

 

 

 

Hi Guys, so I've seen a lot of threads about this and I decided to try it out on my GTX 980. When I heard about this issue it was explicitly said in the article that this does not affect the GTX 980, well if you believe this benchmark it in fact does. As you can see the memory bandwidth tailed off after 3500MB.

I re ran the benchmark again and this was my second result but this time my bandwidth is massive throughout. What gives? well the second bench I ran when my Windows desktop was being powered by my Intel HD 4600. There was no memory usage overhead by Aero because the card had nothing to do.

I looked at MSI Afterburner and I saw that during desktop use my VRAM usage was about 400MB sometimes slightly higher and sometimes slightly lower but always around that range. When I turned off the fancy compositing effects of Aero I noticed a slightly higher increase in the bench results after 3500MB, not much but definitely measurable.

I would recommend anyone who is worried about this to re run the test again but with your integrated card powering your desktop(if you have one) and see if it helps. I'm really starting to believe this is a non issue.

[AddonCustoms]

might not even be a problem with the benchmark itself per say, rather it seems to be more of an issue with a bunch of noobs that don't know how to benchmark properly running around posting screenshots of this benchmark everywhere, claiming that its 'confirming' the issue.

 

The problem is, that no one is running the benchmark in a consistent manner, and most of the screenshots that are being posted don't state the conditions in which the benchmark is being run.

 

All you do with Nai's benchmark is run the EXE and let it do its thing, nothing else is needed, not sure how reliable this tool is. (unless I am not understanding you correctly which is a huge possibility as its gone midnight and I want my bed )

 

My Benchmark - EVGA 970 SSC ACX 2.0 STOCK Clocks

[xAcid9]

Some are running it with aero enabled, some are running with aero disabled, some are running it off the igp with the dedicated gpu in headless mode (and the latter is supposed to be the proper way to run the benchmark for accurate results). Running this benchmark on a gtx 980 with aero enabled seems to 'confirm' the same issue, memory performance trailing off after 3500mb, where running it in headless mode does not:

 

http://www.reddit.com/r/pcgaming/comments/2tiifp/used_nais_benchmark_on_a_gtx_980_to_test_the_gtx/

 

What is headless mode?

[maizenblue]

You have to laugh at all the Nvidia fan boys gushing over their 970s in sli, totally ignoring the shit performance they were getting in some games at higher than 1080p res.

Their fan boyism outweighed what they were experiencing.

[Razzaa]

Response from Nvidia

 

https://forums.geforce.com/default/topic/803518/geforce-900-series/gtx-970-3-5gb-vram-issue/post/4432672/#4432672

 

Also: I have no issues. My card uses all its memory when needed.