sumludusIf you're gaming on a single 60 hertz 1080p monitor, this issue probably won't affect you for the life of the card (of course, if that's your setup, why blow so much on a 970 to begin with).

So who are going to feel the pains of this card as it ages? 144 hertz 1080p? 60 hertz 1440p? People with multi-monitor setups are seeing the issues now, but going forward, which consumers will be at risk?

Could be those who play games with hi-res texture mods applied and it might be possible to trigger this fault with a modded game immediately rather than waiting for the card to age with reduced performance.
You just need a game which would actually use close to a full 4GB of GDDR5 with other processed included.

I could install a GTX 970 I brought for my sister to try it with my modded Skyrim, but I would have to edit some mod settings which could be inconvient for me.
Not sure if it would work though...

awwhhh fuxxx, horrible..

My big question, did W1zz or other reviewers find this problem on their review before?
Especially on 4K...?

Well this is a disappointment this is and the preliminary response. To me this can harm the value of this card especially in the long run but its not the end of all problem to completely devastate this card. Though I find it odd that this was not caught or labeled "ok" in this state.

If they can address it easily for users in some way, more power to them but by the way things are going I do not think its going to be easy.

GhostRyderWell this is a disappointment this is and the preliminary response. To me this can harm the value of this card especially in the long run but its not the end of all problem to completely devastate this card. Though I find it odd that this was not caught or labeled "ok" in this state.

If they can address it easily for users in some way, more power to them but by the way things are going I do not think its going to be easy.

Judging by their official response it was not something that could be fixed without completely retaping the chip.

I might burn in hell for this... but....


Every game I've played so far has been maxed out with this 970 @ 1080.... that being said, I've frequently come close to 3.5GB of VRAM use and surpassed it a few times with games like Far Cry 4 and even that old dog, Skyrim.


I've never once seen or experienced any of the issues shown in the above video.


This is an odd one....

Rahmat SofyanMy big question, did W1zz or other reviewers find this problem on their review before?
Especially on 4K...?

The reviews are there for all to read, if this was a real problem you'd think more would actually mention it or more importantly the reviews would show the impact.

As it stands afterburner simply can't tell me my VRAM usage beyond 3.5GB, I'll live.

If a game isn't written with huge textures to take full advantage of the 4K just increasing the resolution may not increase the memory load, just the output framebuffer size by the difference.

XL-R8RI might burn in hell for this... but....
Every game I've played so far has been maxed out with this 970 @ 1080.... that being said, I've frequently come close to 3.5GB of VRAM use and surpassed it a few times with games like Far Cry 4 and even that old dog, Skyrim.
I've never once seen or experienced any of the issues shown in the above video.
This is an odd one....

Reading "un sourced Nvidia response" at PCPer

Nvidia response via PCPerThe GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section. When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rd party applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we use both segments.

We understand there have been some questions about how the GTX 970 will perform when it accesses the 0.5GB memory segment. The best way to test that is to look at game performance. Compare a GTX 980 to a 970 on a game that uses less than 3.5GB. Then turn up the settings so the game needs more than 3.5GB and compare 980 and 970 performance again.

Here’s an example of some performance data:

The interesting thing is this was really brought to light via a CUDA benchmark. Where the remaining 0.5GB is far slower then the rest of 3.5GB.

Scott Watson from The TechReport points out that it might be more of an issue with newer games that use a strict 64bit executable but more testing needs to be done during ALT+TAB show.

Nvidia could just market the 970 like some user pointed out over at OCN



Or a more clear explanation of the segmented memory on the box which differs from the 980 is marketed with functioning full speed 4gb .

good news!

GTX 960Ti will be here sooner than expected!

GorbazTheDragonJudging by their official response it was not something that could be fixed without completely retaping the chip.

So it would seem though they may be able to "improve it" a bit.

XzibitReading "un sourced Nvidia response" at PCPer



The interesting thing is this was really brought to light via a CUDA benchmark. Where the remaining 0.5GB is far slower then the rest of 3.5GB.

Scott Watson from The TechReport points out that it might be more of an issue with newer games that use a strict 64bit executable but more testing needs to be done during ALT+TAB show.

Nvidia could just market the 970 like some user pointed out over at OCN



Or a more clear explanation of the segmented memory on the box which differs from the 980 is marketed with functioning full speed 4gb .

That should have been done to begin with to avoid this issue.

XL-R8RI might burn in hell for this... but....


Every game I've played so far has been maxed out with this 970 @ 1080.... that being said, I've frequently come close to 3.5GB of VRAM use and surpassed it a few times with games like Far Cry 4 and even that old dog, Skyrim.


I've never once seen or experienced any of the issues shown in the above video.


This is an odd one....

1080p even up to 144hz on recent games is probably not going to exceed 3gb at least by much at the moment. However 1440p 60hz+ might start in recent AAA games and 2160p sure will/does which is where most of this is going to be a problem.

Something like this could easily be avoided if they just would have handled it differently. Hopefully they can address this in some way to at least make it smoother.

GhostRyderSo it would seem though they may be able to "improve it" a bit.

Maybe, but don't count on it. They might try to reduce the issue in chip updates, but it is unlikely to be completely solved. Doesn't this occur with several Kepler and Fermi cards too?

GorbazTheDragonJudging by their official response it was not something that could be fixed without completely retaping the chip.

Well I've had my theory and it somewhat makes more sense now that they explained it, it looks like Maxwells magic sauce was probably in the cache and when they disable an sm naturally the cache for it is disabled too, so it seems they vaguely explained that here saying "hey the 970 has some disabled sm's therefore it can't use all vram efficiently" So while 4gb is present the sms can only address 3.3gb properly with the benefit of the fancy Maxwell cache. Kepler and other architectures probably didn't have that problem because they were less dependant on cache for efficiency. This explains y gtx960 is a straight up cut in half 980, that's because it looks like each sm is designed to handle a certain amount of bandwidth/bus width(16bit) but it remains to be seen as more testing is needed. I am curious about bandwidth and wether the 970 is truly a 256bit card or wether it's actually only 208 bit that is truly efficiently addressable by the cache and the rest is just there, because that's what I get out of this.

Rahmat Sofyan

awwhhh fuxxx, horrible..

My big question, did W1zz or other reviewers find this problem on their review before?
Especially on 4K...?

This doesn't happen, I've taken ACU up to the full 4096mb and i only lost a couple of FPS.

That guy did something to his system in the switch, look at his ram also.

XoriumThis doesn't happen, I've taken ACU up to the full 4096mb and i only lost a couple of FPS.

That guy did something to his system in the switch, look at his ram also.

Well yeah it surely is curious. does this tearing and purple stuff happen if you go over your vram limit? Because I've had older underpowered cards and usually if I run games over their limit I just get low fps and stuttering but I've never see. Artifacts like this lol

sergionogr aphyWell yeah it surely is curious. does this tearing and purple stuff happen if you go over your vram limit? Because I've had older underpowered cards and usu if I run games over their limit I just get low fps and stuttering but I've never see. Artifacts like this lol

Yeah hes done something to that system. Game/driver/bios/bad overclock related.
I've gone over the limit daily @4k in ACU. My desktop is set to 4k as well so It's consuming more vRam.
I lost a couple FPS when it happens maybe some stuttering if it has to change a large portion quickly. But artifacts and weird purple shit going on nonono.....


whats going on in that video is more similar to a bad core overclock. ram issues tend to be blobby like this but more stretched out.

One has to wonder if AMD announced the HMB feature, already knowing about this glitch. Hmmm...

SasquiOne has to wonder if AMD announced the HMB feature, already knowing about this glitch. Hmmm...

HBM has been in development for about a year and the joint venture was announced on Dec 2013, waaayyy before the 970 was out.





Back on topic, this issue might not affect all 970s. It would be interesting to know if there are brands or chip batches that aren't affected. I might snatch one for cheap for my HTPC if the issue becomes mainstream knowledge and people start to drop them on ebay. :peace:



My biggest gripe (disclaimer, I don't own an nVidia card, only one is a 7950GT AGP I got for a retro rig) is the way they worded the response:

The GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section.

They make it sound as if this was intended all along. o_O Since this wasn't announced before the launch and just after a lot of nagging my take is that either:

a) they didn't catch the issue in QC (because they didn't test such VRAM usage scenarios or not every chip is affected), and then came up with the "memory partition" explanation. :rolleyes:

b) they knew beforehand but thought that no one would find out. :nutkick: Apparently the post that started all of this was from a guy that owned both a 980 and a 970 and wondered why both showed a substantial difference in memory usage under the same conditions (same system, same game, same settings).

Constantine YevseyevThis "power user" probably forgot to retarget his super awesome CLI application from "All CPU" to "x64", LOL.

FYI, "Any CPU" means x86 on x86, x64 on x64, IA64 on IA64, and ARM on ARM. It is what Visual Studio uses to denote a platform-neutral binary. This is almost exclusive to virtual machines (.NET and Java) where the same byte code is actually executed by distinct binaries.


The problem might not be with the card and instead either the driver or something on Windows mucking it up. There are really too many variables. NVIDIA needs to find the heart of it.

XzibitWhere the remaining 0.5GB is far slower then the rest of 3.5GB.

The only explanation I can think of for that is those chips are far away from the GPU. I don't trust that idea.

If this is the case, class-action lawsuit incoming.

well, 960 already in market, someone, wth 960 shoould bench or test it for memory utilization along with bandwidth, may be it comes to flaw as well. its time ofor quick 960Ti draw fromNVIDIA. rumors turn out to be long long critic discussion, i am sure in other forums as well, where nvidia co social people monitoring the complaints about the performance.

this can effect the 960 sales and preferences, mot by much but still people will have their concern regarding these memory utiliztio problems on 960 too. NVIDIA address on it. on the other hand, AMD will get a good opportunity to market their best product at good price.

The Nwell, 960 already in market, someone, wth 960 shoould bench or test it for memory utilization along with bandwidth, may be it comes to flaw as well. its time ofor quick 960Ti draw fromNVIDIA. rumors turn out to be long long critic discussion, i am sure in other forums as well, where nvidia co social people monitoring the complaints about the performance.

this can effect the 960 sales and preferences, mot by much but still people will have their concern regarding these memory utiliztio problems on 960 too. NVIDIA address on it. on the other hand, AMD will get a good opportunity to market their best product at good price.

I don't think this will affect nVidia that much unless there are reports all over the web or something but, ironically, nVidia's response might be the ultimate cause for a meltdown.

So far a handful of outlets have covered it and, going by the posts on OCN, it's not universal. Besides, lots of posters were using the benchmark incorrectly (Nai's benchmark must be run on a GPU that's not the primary display device) and many post are showing bad results even for older cards like the 660 :banghead: so it might even be a smaller issue (bad batch?) but now that nVidia has answered the potential for a FUBAR scenario is high since you can bet that an nVidia announcement is going to be covered. Lot's of people that haven't experienced ANY problem with their 970 will then (incorrectly) use Nai's benchmark and decide that their card has a problem. :laugh:

FordGT90ConceptThe only explanation I can think of for that is those chips are far away from the GPU. I don't trust that idea.

If this is the case, class-action lawsuit incoming.

Someone will end up suing someone for sure. You know Samsung is smiling at this with them having sued Nvidia for false advertising on Tegra and might want to prove its a standard practice or go as far as financially backing legal suits against Nvidia.

Anyone checked Radeon R9-290X for this (since it too has 4GB of VRAM)? Just out of curiosity...

RejZoRAnyone checked Radeon R9-290X for this (since it too has 4GB of VRAM)? Just out of curiosity...

It's AMD they don't do a memory mistake, only bad drivers.... :laugh:

sry, i couldn't help myself, but i doubt this here is a layout or memory problem of the GTX 970 maybe GPU problem, that can be fixed with a bios update even this here is taking too long to find the issue.

puma99dk|It's AMD they don't do a memory mistake, only bad drivers.... :laugh:

sry, i couldn't help myself, but i doubt this here is a layout or memory problem of the GTX 970 maybe GPU problem, that can be fixed with a bios update even this here is taking too long to find the issue.

Well, driver mistakes can be fixed. Broken hardware feature, not so much. Apparently this was pretty much hacked into the GTX970 because it's a cut down GTX980 and it simply entiely lacks that part.

RejZoRWell, driver mistakes can be fixed. Broken hardware feature, not so much.

AMD had to redesign crossfire implementation after Tahiti as it was dysfunctional. PR for CF would have showed a misrepresentation of frame rates, implying smooth visual experience, which was far from the truth. So yeah, AMD have had very similar problems. But they fixed it with XDMA implementation in Hawaii, for DX10 forwards.
HOWEVER, I would have bought a 4Gb card expecting it to be able to handle memory usage up to 4Gb. If my fps tanked well before that I'd be pissed.
Nvidia have grossly misrepresented the cards memory capacity. It should have stated in the PR it had 3.5Gb effective gaming memory, or something similar.