TR Forums

Sat Jan 24, 2015 7:32 pm

As an aside Nai - the creator of the little tool we've been playing with - confirmed today that it lacks exclusive access and will run headlong into existing applications in VRAM and spit out crazy numbers.

His post here - in German.

Melvar · Sun Jan 25, 2015 4:18 am

GTX 750 1GB:

And just for fun, my GTX 980:

Both done with all monitors unplugged. The test actually causes the 750's driver to crash & recover if I run it with the monitors plugged in.

Sun Jan 25, 2015 5:28 am

Both done with all monitors unplugged. The test actually causes the 750's driver to crash & recover if I run it with the monitors plugged in.

Looks like DWM in Windows didn't release it's hold despite the monitors being unplugged. It's distinctly unlikely that 128MB of the 750 is on a different segment (that and technically you have another 128MB past that the benchmark reserved). Meanwhile we know the 980 only has a single segment.

The crash you talk about will sometimes happen if it hits an area already in use by the OS.

This is why I'm not a fan of this little benchmark/test. Silly thing is stupid hard to get proper results out of. There's also implications that the tool doesn't know how to access the second segment if a card has one.

Looks like you're running Windows 7 based on the frosted glass look. You might try the test again with one change: choose a theme that cuts off Aero (the frosted glass look) and the monitors unplugged from the video card.

Melvar · Sun Jan 25, 2015 6:48 am

What is this benchmark supposed to test, aside from how much VRAM one can free up for the benchmark? It seems like the problem people are looking for, if it exists, is in the part of VRAM this program doesn't test properly/at all.

Anyhow, switching to the classic theme doesn't seem to get the DWM to release all of the VRAM (more, but not all) in the system with the 980 in it, and I want to figure out what actually needs to happen before I test the 750 again because it's really annoying unplugging/replugging the monitors on that system.

Edit: setting the theme to basic and rebooting before running the test seemed to free up all of the the memory on the 980 (that the program tests). I'm off to try it on the 750.

Even more edit:

Science!

Chrispy_ · Sun Jan 25, 2015 8:08 am

So a vanilla 750 with a chopped-down Maxwell doesn't have this problem; Interesting.

I notice super.speler's first post has the 970 running on a w7 and the 980 on an w8. I'm with Ryu on the issue that this test is more likely to be a test of what OS and how you run it, rather than divulge any useful information about the GPU you're testing.

Pville_Piper · Sun Jan 25, 2015 10:04 am

I wonder if this is what was going on with the Rog Swift monitor I saw at a Chicago Micro Center. The monitor was running a test program of a mountain scene and it was running anywhere from 120 fps to the mid 70's and running very smooth the week before when it was powered by a Mars card (dual GTX 760's). When they had replaced the Mars card with a GTX 970 in it was stuttering and running poorly with a range of upper 70's to upper 20's. I thought a single GTX 970 would out perform a the two 760s but it didn't even come close to the Mars card performance.

Sun Jan 25, 2015 2:26 pm

Chrispy_ wrote:
So a vanilla 750 with a chopped-down Maxwell doesn't have this problem; Interesting.

I notice super.speler's first post has the 970 running on a w7 and the 980 on an w8. I'm with Ryu on the issue that this test is more likely to be a test of what OS and how you run it, rather than divulge any useful information about the GPU you're testing.

Yeah, it's becoming more and more apparent as time passes that Nai's tool is just inadequate for the job. I'm certain now that the driver will not let it access a second segment on cards that sport one.

970s start showing problems at 3.25 because I think it likely that DWM assets are put into the back of the first segment (3.0GB - 3.5GB). Then when the tool hits the 3.5GB to 4.0GB range the virtual memory addresses it has been fed just maps to system RAM. The video driver totally ignores the application and doesn't direct/map it into the second segment.

We need a new tool (CUDA or some other language). One that does this test using fullscreen exclusive mode and that is designed to query both segments.

auxy · Sun Jan 25, 2015 3:32 pm

Ryu Connor wrote:
We need a new tool (CUDA or some other language). One that does this test using fullscreen exclusive mode and that is designed to query both segments.

Nai's tool does use CUDA, but you probably knew that.

With that said, I would like to see a fullscreen exclusive tool that is GPU-vendor-agnostic. I always support the creation of more benchmarks! (*‘∀‘)

GeForce6200 · Tue Jan 27, 2015 3:20 pm

After reading Nvidia's response I see the term "gaming" brought up many times. Since I have pretty limited knowledge of GPU architecture and this "issue" as a whole, I am curious if it would affect rendering. Say if the machine had a low end card to drive the display, and had 970/s specifically used for rendering would they then be affected? I do know that specific GPU rendering software can utilize all of the available VRAM. If this has already been answered I apologize, just was curious in a scenario other than gaming.

Here are some very interesting 4K 970 SLI benchmarks from Reddit

Brian@NVIDIA · Wed Jan 28, 2015 2:53 am

super.speler wrote:
I bought a 4GB video card, and knowing that I can only use about 3.5GB of it doesn't sit well with me.

Hey Super.speler
Just lurking the forums, and stumbled upon your thread. I work for NVIDIA out here in Santa Clara, CA.

There has been a lot of back and forth about the GTX 970 vram issue. Let me just jump in and say that while the GTX 970 is just as amazing today as it was when the card launched, our communication as a company has clearly been problematic.

I understand why GTX 970 owners are upset. We posted incorrect and misinformed specs and we didn't properly explain the memory architecture. We never intended to deceive anyone but despite our best intentions many of you got wrong information/specs that impacted your buying decision which we have ultimately let you down.

The GTX 970 is still an amazing GPU and still deserves the praise it has received throughout the community, review at launch.

But, with that said, you and others may feel different. You might feel mislead and feel there isn't much you can do. Well you DO have an option.
If any of you don't want the GTX 970 you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange. You should do what will give you the best gaming experience possible and if you need help to get that done let me know, we'll help.

super.speler · Wed Jan 28, 2015 5:35 am

Brian@NVIDIA wrote:
super.speler wrote:
I bought a 4GB video card, and knowing that I can only use about 3.5GB of it doesn't sit well with me.

Hey Super.speler
Just lurking the forums, and stumbled upon your thread. I work for NVIDIA out here in Santa Clara, CA.

There has been a lot of back and forth about the GTX 970 vram issue. Let me just jump in and say that while the GTX 970 is just as amazing today as it was when the card launched, our communication as a company has clearly been problematic.

I understand why GTX 970 owners are upset. We posted incorrect and misinformed specs and we didn't properly explain the memory architecture. We never intended to deceive anyone but despite our best intentions many of you got wrong information/specs that impacted your buying decision which we have ultimately let you down.

The GTX 970 is still an amazing GPU and still deserves the praise it has received throughout the community, review at launch.

But, with that said, you and others may feel different. You might feel mislead and feel there isn't much you can do. Well you DO have an option.
If any of you don't want the GTX 970 you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange. You should do what will give you the best gaming experience possible and if you need help to get that done let me know, we'll help.

I feel like my biggest issue, and many others', is that I purchased my GTX 970 before the holiday season, in anticipation for the holiday season. I have already contacted the retailer and they simply told me that it was past the RMA date, instead pointing me to the manufacturer, which claims that the product is working as designed. If you could shed some light on how you plan on helping, that would be appreciated.

Kougar · Wed Jan 28, 2015 7:56 am

In all fairness, the new understanding of how the 970's VRAM works doesn't change the card's initial launch day performance.

At this point I do wonder about non-gaming applications such as Crunchers, Miners, and Folders. Or anyone that runs simulation or other computational loads on a 970. As far as I know Folding@home won't be using anywhere near that much VRAM for some time though.

LoneWolf15 · Sat Jan 31, 2015 8:59 pm

Brian@NVIDIA wrote:
super.speler wrote:
But, with that said, you and others may feel different. You might feel mislead and feel there isn't much you can do. Well you DO have an option.
If any of you don't want the GTX 970 you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange. You should do what will give you the best gaming experience possible and if you need help to get that done let me know, we'll help.

Hi Brian,

I had a Geforce GTX 970. I feel they have very good performance in current games. With no disrespect to you, that's really not the point. Even given that, what I purchased was different than what I thought I was buying.

My two thoughts on this debacle are similar to how I felt when I bought a Geforce 6800 and found it had broken PureVideo, a feature nVidia advertised, and didn't stop advertising even when the brokenness became known (the PDF remained available on the website). It's also similar to how I felt when my nForce4 mainboard had a broken hardware firewall.

In those cases, nVidia acknowledged the issues --sort of, grudgingly; and then we were stuck with what we had, beyond the RMA period. Yes, those are old examples. But then again, I wonder how a GTX 970 will perform not just now, but in future games. I'm glad you're offering to help people return their cards; how well nVidia does in offering that help to people beyond their RMA period will determine how I view you as a company. I don't expect perfection from companies, but it's a big a deal to me how a company handles their mistakes. My history with how nVidia handled their past ones meant I didn't purchase their cards for awhile (even when they had the best gear out there), so I hope to see in this case your company does differently.

In my case, I sold my 970 at a loss, and (shame on me, but it is the best card) bought a GTX 980. It's a great card, but it may be the last nVidia card I buy for some time. I hope you truly are telling card vendors that you'll make good on customer returns if that's their choice, considering a lot of vendors have exchange-only terms or limited return periods on graphics cards. It would go a long way towards helping people trust your company. Thank you.

LoneWolf15 · Sat Jan 31, 2015 9:12 pm

GeForce6200 wrote:
After reading Nvidia's response I see the term "gaming" brought up many times. Since I have pretty limited knowledge of GPU architecture and this "issue" as a whole, I am curious if it would affect rendering. Say if the machine had a low end card to drive the display, and had 970/s specifically used for rendering would they then be affected? I do know that specific GPU rendering software can utilize all of the available VRAM. If this has already been answered I apologize, just was curious in a scenario other than gaming.

Here are some very interesting 4K 970 SLI benchmarks from Reddit

If I were doing rendering, I would probably purchase different cards than the Geforce 9xx series, unless it was hobbyist rendering. This is because the Geforce 900 series has been designed principally with gaming in mind; as such, nVidia is focusing these cards on single-precision floating point ops. This is different than their professional workstation cards, which also have a a focus on double-precision floating point.

Cites:
http://www.anandtech.com/show/8526/nvid ... -review/20

http://www.tomshardware.com/reviews/gef ... 50-16.html

For professional rendering work, you'd want workstation graphics cards instead.

super.speler · Sun Feb 01, 2015 3:21 am

Brian@NVIDIA wrote:
super.speler wrote:
I bought a 4GB video card, and knowing that I can only use about 3.5GB of it doesn't sit well with me.

Hey Super.speler
Just lurking the forums, and stumbled upon your thread. I work for NVIDIA out here in Santa Clara, CA.

There has been a lot of back and forth about the GTX 970 vram issue. Let me just jump in and say that while the GTX 970 is just as amazing today as it was when the card launched, our communication as a company has clearly been problematic.

I understand why GTX 970 owners are upset. We posted incorrect and misinformed specs and we didn't properly explain the memory architecture. We never intended to deceive anyone but despite our best intentions many of you got wrong information/specs that impacted your buying decision which we have ultimately let you down.

The GTX 970 is still an amazing GPU and still deserves the praise it has received throughout the community, review at launch.

But, with that said, you and others may feel different. You might feel mislead and feel there isn't much you can do. Well you DO have an option.
If any of you don't want the GTX 970 you have purchased, knowing what you know about the performance in your system, you should return it. Get a refund or an exchange. You should do what will give you the best gaming experience possible and if you need help to get that done let me know, we'll help.

Four days has passed and you haven't really replied to my question asking how you plan on helping. Did you actually plan on following through or did you just stop by to post a friendly message that seems to be copy and pasted across numerous websites?

l33t-g4m3r · Sun Feb 01, 2015 5:57 am

LoneWolf15 wrote:
In my case, I sold my 970 at a loss, and (shame on me, but it is the best card) bought a GTX 980.

https://www.youtube.com/watch?x-yt-cl=8 ... 9428#t=134

LoneWolf15 · Sun Feb 01, 2015 9:52 am

l33t-g4m3r wrote:
LoneWolf15 wrote:
In my case, I sold my 970 at a loss, and (shame on me, but it is the best card) bought a GTX 980.

https://www.youtube.com/watch?x-yt-cl=8 ... 9428#t=134

Saw it already, laughed at it, thought it was hilarious. I do have a sense of humor, even with the knowledge I did exactly what it said.

My issue is really that I want a vendor to be up front at a point where I get to make the decision. If I bought a GTX 970 knowing everything about it because all of that information was there for me to make the decision, fine. And if I'd had that information, maybe I'd have started with a GTX 980, or --I'd have chosen to wait longer with the hardware I had before. Either way, I'd have been able to respect them for giving me all the information to make an informed decision.

Most of us here have seen enough GPU architecture that we already expect that each block of memory on a graphics card will have equal access/speed on the bus to the GPU. In fact, I'd say we assume it, unless a vendor tells us otherwise; that's how it has been for years. What I draw from this experience and others is that from now on, I should wait at least six months to buy any product based on nVidia technology to make sure everything they said at launch about their product is correct. And if everyone were to do this, it would mean that nVidia loses out on the best profit margins from opening day product, but also that their competitors get a bonus six months to come out with something better, giving the chance to choose from that as well.

G8torbyte · Mon Jul 17, 2017 4:16 pm

sschaem wrote:
Hopefully we are past the 4GB confusion by now.

To recap: The OP didn't highlight any issue with addressing 4G. but the fact that the GX970 is suffering performance issues >3.2GB, while the 980 doesnt.

I dont think any game truly leverage >3GB today (its mostly texture cache) , so even if nvidia, in their drivers, limit the usage of the 970 to 3.2GB, nobody would be the wiser.

But, if anyone can prove that the GTX 970 is doing that, nvidia will be hit with a serious class action suit.

Yep, already done. The claim period was last year to submit your proof of purchases. I think the cutoff date was in Nov or Dec. I received my settlement check today for the claim $30 per video card. If you applied in time look for yours too in the mail soon.

TR Forums

Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Re: Nvidia GTX 970 VRAM Limitation

Who is online