Monday, January 26th 2015

NVIDIA Responds to GTX 970 Memory Allocation 'Bug' Controversy

The GeForce GTX 970 memory allocation bug discovery, made towards last Friday, wrecked some NVIDIA engineers' weekends, who composed a response to what they tell is a non-issue. A bug was discovered in the way GeForce GTX 970 was allocating its 4 GB of video memory, giving some power-users the impression that the GPU isn't addressing the last 700-500 MB of its memory. NVIDIA, in its response, explained that the GPU is fully capable of addressing its 4 GB, but does so in an unusual way. Without further ado, the statement.
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. 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.
Continued

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:
<div class="table-wrapper"><table class="tputbl hilight" cellspacing="0" cellpadding="3"><caption>GTX 970 vs. GTX 980 Memory-Intensive Performance Data </caption><tr><th scope="col">&nbsp;</th><th scope="col">GeForce <br /> GTX 980</th><th scope="col">GeForce <br /> GTX 970</th></tr><tr><th scope="row">Shadow of Mordor</th><td align="right"></td><td align="right"></td></tr><tr class="alt"><th scope="row"><3.5GB setting = 2688x1512 Very High</th><td align="right">72 fps</td><td align="right">60 fps</td></tr><tr><th scope="row">>3.5GB setting = 3456x1944</th><td align="right">55fps (-24%)</td><td align="right">45fps (-25%)</td></tr><tr class="alt"><th scope="row">Battlefield 4</th><td align="right"></td><td align="right"></td></tr><tr><th scope="row"><3.5GB setting = 3840x2160 2xMSAA</th><td align="right">36 fps</td><td align="right">30 fps</td></tr><tr class="alt"><th scope="row">>3.5GB setting = 3840x2160 135% res</th><td align="right">19fps (-47%)</td><td align="right">15fps (-50%)</td></tr><tr><th scope="row">Call of Duty: Advanced Warfare</th><td align="right"></td><td align="right"></td></tr><tr class="alt"><th scope="row"><3.5GB setting = 3840x2160 FSMAA T2x, Supersampling off</th><td align="right">82 fps</td><td align="right">71 fps</td></tr><tr class="alt"><th scope="row"><3.5GB setting = >3.5GB setting = 3840x2160 FSMAA T2x, Supersampling on</th><td align="right">48fps (-41%)</td><td align="right">40fps (-44%)</td></tr></table></div>
On GTX 980, Shadows of Mordor drops about 24% on GTX 980 and 25% on GTX 970, a 1% difference. On Battlefield 4, the drop is 47% on GTX 980 and 50% on GTX 970, a 3% difference. On CoD: AW, the drop is 41% on GTX 980 and 44% on GTX 970, a 3% difference. As you can see, there is very little change in the performance of the GTX 970 relative to GTX 980 on these games when it is using the 0.5GB segment.
Source: The TechReport
Add your own comment

92 Comments on NVIDIA Responds to GTX 970 Memory Allocation 'Bug' Controversy

#76
Caring1
xorbeAm I reading into it wrongly, or is the 970 basically operating in 224-bit mode most of the time, and not 256-bit mode?
Yep, you're reading it wrong, that's the bandwidth in Gb/s
Posted on Reply
#77
HumanSmoke
SteevoNOOOO NVIDIA IS OUR BULL GOD!!!!!!!
Seriously though, for most its still a good deal.
Pretty much, the numbers might change but the performance hasn't magically altered since it launched.

Having said that, I wouldn't mind one bit if Nvidia dropped prices across the board, offered AAA title game keys, and intro'd the GM 200 early just to erase the bad taste.
Posted on Reply
#78
xorbe
Caring1Yep, you're reading it wrong, that's the bandwidth in Gb/s
Are you sure? If it's only using 7 of 8 channels for 3.5GB that's 224-bit mode. Then the 8th channel is like 32-bit mode, and the two modes are exclusive. I thought AnandTech made it clear that it strides 1-2-3-4-5-6-7 - 1-2-3-4-5-6-7 and doesn't touch the remaining 8th channel until more than 3.5GB.
Posted on Reply
#79
the54thvoid
Super Intoxicated Moderator
trenterIgnorance stones? Only a fanboy in denial, or a complete moron would take nvidia's explanation as truth with zero skepticism. Performance at the time of review isn't the problem, the problem is that people were sold a gpu expecting the specifications of that gpu to be the same four months later as they were at release. Also, people don't just buy a gpu to play only the games on market up to the point of release, they buy them with future performance in mind. You don't think there are some people that may have decided to skip the 970 if they had known there could be problems addressing the full 4gb of ram in the future, especially when console ports are using up to 3.5gb at 1080p now? What about the people that were worried about that 256 bit memory bus? Nvidia pointed out to reviewers that their improved L2 cache would keep memory requests to a minimum, therefore allowing more "achieved" memory bandwidth even with a smaller memory interface. The point is they lied about the specs, and it's almost impossible to believe that this is some kind of big misunderstanding between the engineering and marketing team that they just happened to realize after 970 users discovered it.
Mmm. Read my posts much? I did say I don't believe NV were genuine. Yawn...
Posted on Reply
#80
HumanSmoke
R-T-BWait a second, ex-miners? That was so 2013. I haven't seen cards used in mining at any profitable margins since mid-2014, and even then nearly no one was doing it anymore.
There might still be a few on the market, but I doubt it.
There are still more than a few out there. A quick browse of eBay shows multiple listings. This guy has 8 used R9 290's for sale, used "in an air conditioned environment", kind of screams scrypt mining, even if the number of boards being sold isn't a big enough clue.
Posted on Reply
#81
R-T-B
Ah. Must be some alt-coin craze I was unaware of after litecoin then. Wow, you'd think people would realize GPU mining profit wasn't all that great after that and cut it out way before now. *shrugs*

For what it's worth, when I sold my mining cards, I did so here and sold them dirt cheap with big warnings on them. Thanks for the correction, anyhow.
Posted on Reply
#82
buggalugs
I'm waiting for an apology to the community for all the guys here that said there was no issue..... Or at least just say you were wrong.

Nvidia advertised the 970 as the same memory subsystem as the 980. That is clearly a big fat lie. and its not even a little different, its completely different. I cant see how they could "overlook" that in the specs.

Edit: ....and why does GPUz show 64 rops when there are 56?? Does GPUz actually detect or are the specs written in based on the model number??
Posted on Reply
#83
W1zzard
buggalugswhy does GPUz show 64 rops when there are 56?? Does GPUz actually detect or are the specs written in based on the model number??
GPU-Z asks the NVIDIA driver. Even if I queried the hardware units, it would still show them as active, because they are not really disabled (they can be used for AA)
Posted on Reply
#84
rooivalk
It's amusing that what many people want is:
a. apology
b. free game
c. price drop

rather than anything that solve, negate, or prevent further problem.
Posted on Reply
#85
Xzibit
W1zzardGPU-Z asks the NVIDIA driver. Even if I queried the hardware units, it would still show them as active, because they are not really disabled (they can be used for AA)
If your quering the driver and its telling you 64 does that mean the marketing department is making drivers now ?

We all just got told that the engineers knew but never communicated the correct specs to the marketing team.
Posted on Reply
#86
buggalugs
W1zzardGPU-Z asks the NVIDIA driver. Even if I queried the hardware units, it would still show them as active, because they are not really disabled (they can be used for AA)
Ok I see.....and GPUz doesn't detect cache so that went unnoticed.

The more I think about this, the worse it looks for Nvidia. There is a very long thread on Nvidia forums about this starting since just after the 970 release late last year. Nvidia only makes comment when the story is reported by major tech sites. I cant believe they didn't know about this much earlier......its like they held off as long as they could to continue the hype and get 970 sales over the Christmas period.

The 970 memory subsystem is unique, I'm not aware of a similar system at least on mid-high end graphics cards. Unless they have been doing it and we weren't aware. I just don't accept they overlooked a memory subsystem that is so unique and forgot to mention it....and I don't accept it took them 3 months to figure it out.
Posted on Reply
#87
HumanSmoke
buggalugsThe more I think about this, the worse it looks for Nvidia. There is a very long thread on Nvidia forums about this starting since just after the 970 release late last year. Nvidia only makes comment when the story is reported by major tech sites. I cant believe they didn't know about this much earlier..........and I don't accept it took them 3 months to figure it out.
It's actually quite believable.
What percentage of GTX 970 buyers encountered the problem? Sometimes people (especially the disgruntled) post multiple times over multiple forums, but for the most part, 970 owners don't seem that affected - even owners here and elsewhere say it hasn't impacted them personally (in fact the biggest outcry is from people who own AMD cards, go figure!). Nvidia stated they'd sold a million GTX 970's and 980's, and I'll go out on a limb and say that the bulk of those sales are of the former. Hundreds of thousands of cards sold, and how many individual issues (as opposed to multiple postings by individuals) reported ?
You don't have to look very far for a precedent. AMD's Evergreen series owners started questioning the stability of the cards almost from launch in September 2009( I returned two cards of mine personally and 3-4 from builds I was doing for others). It wasn't until the issue became better publicized in early 2010 that AMD started work on trying to locate and remedy the problem ( PowerPlay state voltage settings).
It would be nice to entertain the thought that this kind of stuff is acted as soon as it rears it head, but that seldom happens unless the issue is pervasive.
Posted on Reply
#88
sergionography
xorbeAre you sure? If it's only using 7 of 8 channels for 3.5GB that's 224-bit mode. Then the 8th channel is like 32-bit mode, and the two modes are exclusive. I thought AnandTech made it clear that it strides 1-2-3-4-5-6-7 - 1-2-3-4-5-6-7 and doesn't touch the remaining 8th channel until more than 3.5GB.
Another thing that was mentioned is that each smm on Maxwell is capable of outputting 4pixels per clock meaning even tho it has 58 rops, in reality it only has 52pixel/second fillrate due to having 13smm. and in other words it's more like 208bit or being able to feed 6.5 of the channels, now had it been that nvidia enabled only 12smms then 192 bit would've fed the gpu just fine. Now with all this brought to attention it only makes me realize how ignorant I been about so many of the details on gpus. How does this translate to other architectures like gcn for example? Yes Hawaii has 512bit and 64rops, but do they get fed efficiently or are they there more for compute rather than graphics? Because that's what it feels like. And with this being said everyone complained about gtx 960 having only 128bit, but to come think about it having 32rops it gets fed exactly according to how much it can handle without any bottlenecks.
Posted on Reply
#89
xorbe
When I say 256-bit vs 224-bit, I'm referring to the effective vram width (not sm/core/rop guts of the GPU, that can vary a lot as usual). I'm surprised everyone is hung up on the 4 vs 3.5, and not the 256 vs 224.
Posted on Reply
#90
TRWOV
HumanSmokeAMD's Evergreen series owners started questioning the stability of the cards almost from launch in September 2009( I returned two cards of mine personally and 3-4 from builds I was doing for others). It wasn't until the issue became better publicized in early 2010 that AMD started work on trying to locate and remedy the problem ( PowerPlay state voltage settings).
Yeah, but in AMDs case they didn't really know what was going on and had to investigate ans since it wasn't repeatable 100% of the time that made things more difficult, in this case nVidia knew everything beforehand and in fact they themselves caused the confusion by providing the press with wrong specs. Not saying that AMD hasn't done things wrong in the past just that the example you used isn't comparable IMO.

I mean, we all know AMD and nVidia cherry pick benchmarks and stuff to make their products look better but flat out giving wrong specs is a different thing.
Posted on Reply
#91
Casecutter
What sinks is that the more I read the more we uneath the smell of stinky fish... In that Nvidia looks to have found plenty of chip with 3-SM that needed to be fused, as planned. However they found many chips having one on the L2 defective on too many chip and appeared to figure out a way to weasel around it because "nobody ever checks or question L2" we take their word (spec's). I don't care if it performs alright, they duped folks because their equiptment wasn’t as advertised.

Nvidia screwed to pooch on this by saying the GTX 970 ships with THE SAME MEMORY SUBSYSTEM AS OUR FLAGSHIP GEFORCE GTX 980, and I'm not buying the... PR didn't get the message. Nvidia doesn’t need to lower price or give away games… It's the owners who purchase prior who should have a way to be compensated if they want. Nvidia should just step-up… come out and say if you want to return them we’ll refund all your money, or you can apply for some form of settlement. If Nvidia can’t bring themselves to do that then, a Class Action Suit should be brought to make the owner who were duped be provide some compensation.

What's so funny are the guy here defending Nvidia saying it not a big deal, or they all do it… OMG! If this was AMD or Apple those same folks would be calling for their heads. To allow Nvidia to sweep this under the rug would just promote other Tech companies to further deceive with impunity.
Posted on Reply
#92
HumanSmoke
TRWOVYeah, but in AMDs case they didn't really know what was going on and had to investigate ans since it wasn't repeatable 100% of the time that made things more difficult, in this case nVidia knew everything beforehand and in fact they themselves caused the confusion by providing the press with wrong specs. Not saying that AMD hasn't done things wrong in the past just that the example you used isn't comparable IMO.
I was using comparable in the sense that between the first signs of detection and subsequent actioning there is a lag. If you wanted a more apples-to-apples comparison of the companies prior knowledge of performance discrepancy but declining to publicize it, a more apt comparison would be AMD's advertising of Bulldozer as an eight core processor knowing full well that limitations of shared resources meant a degradation in core performance.
TRWOVI mean, we all know AMD and nVidia cherry pick benchmarks and stuff to make their products look better but flat out giving wrong specs is a different thing.
Agreed.
Posted on Reply
Add your own comment
Dec 21st, 2024 21:28 EST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts