WCCftech: Memory allocation problem with GTX 970 [UPDATE] PCPer: NVidia response

Page 11 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
If any issues actually come up with games from this configuration they will necessarily have to tweak how the memory is managed in the driver. It's possible it is ONLY a static setting in the firmware but I seriously doubt Nvidia would use such a ham fisted approach.

In addition this puts the review sites suddenly going back to FRAPS from FCAT in a bad light. Was this a Nvidia suggestion to avoid a site finding a game where this 3.5+0.5GB solution fails?
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
Well its not working at all in some games you have basically 3.5Gb card no matter what.Especially if you playing in 1080P
http://www.dailymotion.com/video/x2ejqt1_skyrim-gtx970-8xmsaa

GTX970 is not 256bit and 4GB card.
It is 208bit 3.5GB + 48bit 512Mb card.LIARS!!!!
https://www.youtube.com/watch?v=JbovJbKALzA

So you run through a town with a modded skyrim that is barely getting 30fps to begin with at 8xmsaa and you wonder why some spots drop ridiculously? Hell it did that on my 670s. Skyrim is also very well known to use an engine prone to stuttering. I'm not sure what you're even trying to prove anymore because it makes no sense.

You need to make better examples. Especially that last bit claiming it's not a 256bit memory controller lol.
 
Last edited:

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Note: FCAT usage suddenly started to dry up among reviewers with the 980/970 launch.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
LOL average fps tables when the issue is showing its face in stuttering .

Show them FCATs and frametimes nvidia then will talk

Frametimes were done by guru3d and techreport in a few games. Look at them. I don't see a problem.
 
Last edited:

Enigmoid

Platinum Member
Sep 27, 2012
2,907
31
91
So its basically like the 660Ti as far as the end effect goes.

The effect of this is going to vary. If the last 500 MB of data in Vram are things that are infrequently accessed then you will only see a minimal effect. If the whole 4 GB are constantly needed then things are going to be messy.
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
Well its not working at all in some games you have basically 3.5Gb card no matter what.Especially if you playing in 1080P
http://www.dailymotion.com/video/x2ejqt1_skyrim-gtx970-8xmsaa

Wrong. The driver creates a virtual adress room. It doesnt mean that the game needs all the vram.

Here an example from AC:Unity with my GTX980:
1440p + FXAA: 4000MB
1440p + 2xAA: 4000MB
1440p + 4xAA: 4000MB
1440p + 8xAA: 3600MB + stutterung (failmode)
 

ocre

Golden Member
Dec 26, 2008
1,594
7
81
http://techreport.com/news/27721/nvidia-admits-explains-geforce-gtx-970-memory-allocation-issue

Interesting. We explored a similar datapath issue related to the GTX 970's disabled SMs in this blog post. In that case, we looked at why the GTX 970 can't make full use of its 64 ROP partitions at once when drawing pixels. Sounds to me like this issue is pretty closely related.
Beyond satisfying our curiosity, though, I'm not sure what else to make of this information. Like the ROP issue, this limitation is already baked into the GTX 970's measured performance. Perhaps folks will find some instances where the GTX 970's memory allocation limits affect performance more dramatically than in Nvidia's examples above. If so, maybe we should worry about this limitation. If not, well, then it's all kind of academic.
they explored it here,

http://techreport.com/blog/27143/here-another-reason-the-geforce-gtx-970-is-slower-than-the-gtx-980

3dm-color.gif


the gtx 970 is built this way. Its the same GPU that everyone has tested.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
Personally I trust Nvidia engineers who designed it much more than a bunch of people on the internet claiming they need to do a recall. Especially when I've tested a bunch of games and am very pleased with the performance. They couldn't have made the driver work properly with every game and every mod out there and I have yet to find a situation where I have issues.
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
I dont think that the benchmark is flawed. More that with Cuda it is not possible to get to the other segment.
 

lavaheadache

Diamond Member
Jan 28, 2005
6,893
14
81
To me this just muddies the waters when it comes to how much ram a card has sort of like cache when it comes to a cpu
 

Cloudfire777

Golden Member
Mar 24, 2013
1,787
95
91
I dont think that the benchmark is flawed. More that with Cuda it is not possible to get to the other segment.

The benchmark is incapable of accessing the last 500MB like real software. games for example do, but instead falls down in speed because the 3500MB is full and it starts dealing with flushing of cache and dealing with system RAM, its a useless benchmark.

The GPU have also been tested extensively in GPGPU software and nobody noticed anything fishy there either, so the benchmark have no bearing.
 
Last edited:

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Frametimes were done by guru3d and techreport in a few games. Look at them. I don't see a problem.

TechReport actually switched to using mostly FRAPS once big Maxwell showed up. Why would the site cheerleading "MAXIMUM SMOOTHNESS" suddenly go back to the less accurate FRAPS?

Most of the numbers you'll see on the following pages were captured with Fraps, a software tool that can record the rendering time for each frame of animation. We sometimes use a tool called FCAT to capture exactly when each frame was delivered to the display, but that's usually not necessary in order to get good data with single-GPU setups. We have, however, filtered our Fraps results using a three-frame moving average. This filter should account for the effect of the three-frame submission queue in Direct3D. If you see a frame time spike in our results, it's likely a delay that would affect when the frame reaches the display.

We didn't use Fraps with BF4. Instead, we captured frame times directly from the game engine itself using BF4's built-in tools. We didn't use our low-pass filter on those results.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
Personally I trust Nvidia engineers who designed it much more than a bunch of people on the internet claiming they need to do a recall. Especially when I've tested a bunch of games and am very pleased with the performance. They couldn't have made the driver work properly with every game and every mod out there and I have yet to find a situation where I have issues.

They should have clearly explained the 970 memory scheme ala mixed mode in previous cards. So that the individual potential buyer could decide whether they will trust Nvidia to make sure this doesn't affect their gaming experience.

Heck we weren't even sure to believe the people who discovered this quirk in the 970s VRAM. People were saying they were imagining it.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
TechReport actually switched to using mostly FRAPS once big Maxwell showed up. Why would the site cheerleading "MAXIMUM SMOOTHNESS" suddenly go back to the less accurate FRAPS?

The thing for me is though that there would be severe drops if this was a problem. That doesn't appear. Even FRAPS would show the drastic drop off.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
The thing for me is though that there would be severe drops if this was a problem. That doesn't appear. Even FRAPS would show the drastic drop off.

One of the reasons sites started using FCAT is that FRAPS doesn't capture all forms of stuttering.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
They should have clearly explained the 970 memory scheme ala mixed mode in previous cards. So that the individual potential buyer could decide whether they will trust Nvidia to make sure this doesn't affect their gaming experience.

Heck we weren't even sure to believe the people who discovered this quirk in the 970s VRAM. People were saying they were imagining it.

The thing is though, games still run great. That's why people buy these cards is it not? The "issue" is overblown...to the point I'm surprised I don't see memes popping up everywhere like the XBox 360 RROD.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
One of the reasons sites started using FCAT is that FRAPS doesn't capture all forms of stuttering.

I'm not talking about stuttering. If the bandwidth dropped from 150G/s to something like 12 as people suggest then I can assure you the game would throw up all over you to let you know something is wrong.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
The thing is though, games still run great. That's why people buy these cards is it not?

Games right now, but even then there is a sudden dip of FCAT usage which would best catch any microstutter.

There is a difference between being oblivious to something, in this case due to Nvidia not disclosing this information, and deciding that Nvidia will make sure it doesn't cause you any problems. They were upfront with the 550 Ti split memory bus.
 

Enigmoid

Platinum Member
Sep 27, 2012
2,907
31
91
TechReport actually switched to using mostly FRAPS once big Maxwell showed up. Why would the site cheerleading "MAXIMUM SMOOTHNESS" suddenly go back to the less accurate FRAPS?

Most of the numbers you'll see on the following pages were captured with Fraps, a software tool that can record the rendering time for each frame of animation. We sometimes use a tool called FCAT to capture exactly when each frame was delivered to the display, but that's usually not necessary in order to get good data with single-GPU setups. We have, however, filtered our Fraps results using a three-frame moving average. This filter should account for the effect of the three-frame submission queue in Direct3D. If you see a frame time spike in our results, it's likely a delay that would affect when the frame reaches the display.

We didn't use Fraps with BF4. Instead, we captured frame times directly from the game engine itself using BF4's built-in tools. We didn't use our low-pass filter on those results.

http://techreport.com/review/27067/nvidia-geforce-gtx-980-and-970-graphics-cards-reviewed/5

FCAT is difficult to use and generates a ton of data (AT and TR complained about this).

Also, our FCAT video capture and analysis rig has some pretty demanding storage requirements. For it, Corsair has provided four 256GB Neutron SSDs, which we've assembled into a RAID 0 array for our primary capture storage device. When that array fills up, we copy the captured videos to our RAID 1 array, comprised of a pair of 4TB Black hard drives provided by WD.

They also said right there that there is little difference for single GPU setups.

For multigpu FCAT is a must.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
I'm not talking about stuttering. If the bandwidth dropped from 150G/s to something like 12 as people suggest then I can assure you the game would throw up all over you to let you know something is wrong.

It would only be when accessing things in that 500MB section, so game behavior would vary. Microstuttering would be a definite possibility.
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
They didn't go in the driver and set programming for every game and every mod. That's kind of impossible. I have a pretty big library and some of my older games can use more than 3.5GB of memory at high resolution. I did some 4k tests to see if I could use DSR in some older titles and didn't find any drops there either. The FPS drop off would be extreme enough to see quite easily.
 
Last edited:

wand3r3r

Diamond Member
May 16, 2008
3,180
0
0
TechReport actually switched to using mostly FRAPS once big Maxwell showed up. Why would the site cheerleading "MAXIMUM SMOOTHNESS" suddenly go back to the less accurate FRAPS?

Most of the numbers you'll see on the following pages were captured with Fraps, a software tool that can record the rendering time for each frame of animation. We sometimes use a tool called FCAT to capture exactly when each frame was delivered to the display, but that's usually not necessary in order to get good data with single-GPU setups. We have, however, filtered our Fraps results using a three-frame moving average. This filter should account for the effect of the three-frame submission queue in Direct3D. If you see a frame time spike in our results, it's likely a delay that would affect when the frame reaches the display.

We didn't use Fraps with BF4. Instead, we captured frame times directly from the game engine itself using BF4's built-in tools. We didn't use our low-pass filter on those results.

http://techreport.com/review/27067/nvidia-geforce-gtx-980-and-970-graphics-cards-reviewed/5
I can't think of a bigger flip-flop position from a fairly popular review site. IIrc they were basically in bed with NV bringing FCAT to demonstrate how much "smoother but slower" kepler was, even in single card and it's importance in demonstrating the lack of smoothness, and now to the complete reverse! :thumbsdown:

I believe they were basically NV's mouthpiece with the original debacle and apparently they still are since they have reversed course 100% despite justifying the need themselves! I would expect more integrity but I guess they are just shamelessly promoting NV, even if it requires reversing their positions.
 
Status
Not open for further replies.