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

[cmdrdredd]

So was realy because of the SMMs cut.

The statment confirms that Nvidia knew all the time as they write the driver to priorize the 3.5GB but hide this information of the customers.:thumbsdown:

Why does it have to be drivers? Why can't it just be the card's firmware that handles it?

 

[garagisti]

So was realy because of the SMMs cut.

The statment confirms that Nvidia knew all the time as they write the driver to priorize the 3.5GB but hide this information of the customers.:thumbsdown:

Hear, hear! I think a 970 was a perfectly good card, and will remain so when advertised properly about its abilities. The only thing i can think of is, perception problems at consumer end, may have led to lower average selling price, which one can suggest is good for consumers. I sincerely think that Nvidia needs to be called out for not leaving this out in public domain, and the speed at which they have come along with the information leaves no doubt that it was plain deliberate.

 

[Cloudfire777]

So was realy because of the SMMs cut.

The statment confirms that Nvidia knew all the time as they write the driver to priorize the 3.5GB but hide this information of the customers.:thumbsdown:

If the games want more than 3500MB it will gain access to the last 500MB. It really doesnt matter if its 4000MB at once (GTX 980) or 3500MB+500MB (GTX 970) since the various game tests from Nvidia show there is no difference in performance

 

[amenx]

Like they did of bumpgate? IIRC, only Charlie was speaking about it, and he was being flamed left, right and center. I wouldn't be surprised to learn that some Charlie shaped effigies were burned, and Pinatas were smashed. Even now, whenever it comes up, you could see someone suggest that it was not that big a deal.

So are you saying to disregard the entire tech world even if they were to do testing of the issue with clear and precise criteria/standards? That you will only take one answer and that it must fit your (and Charlies) viewpoint that Nvidia has knowingly and willfully screwed every 970 owner no ifs, ands or buts about it? You seem like a reasonable guy.

p.s. reminds of AMDZone where they dismissed all reviews of C2Ds, Sandybridge, etc as bought and paid for by Intel.

 

[Spanners]

The only outcome from this drama made by bored people not knowing what they are looking at is an official statement from Nvidia explaining why the benchmark is showing incorrect data, reviews from several well known sites showing it doesnt affect gaming or any other real world scenarios. Followed by denial and twisting from people who thought it was the real deal and was looking for Nvidia`s downfall because they thought recall would happen.

There are many cases where people run the benchmark correctly and still get these low bandwidth drops and not being able to use VRAM fully.

Here, I posted this earlier. Read
http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

The users that mention stuttering and stuff like that are as vague as one can be. Stuttering can happen on any machine, with any GPU setup, on certain scenarios.
I see people posting graphs from maxed out games with high resolution, lots of MSAA to get VRAM usage up and say the reduced FPS is because of reduced bandwidth, but doesnt realize that this setting can force even the most hardcore computer down on its knees regardless of any bandwidth you have available

Seems the benchmark was identifying a real-world difference between the way the 970 and 980 handle memory allocation. Validates what I was saying in point B including the fact that it may have a minimal impact real-world.

"drama made by bored people" Has motivated Nvidia to explain what was going on underneath the hood which is good for everyone.

It would be interesting to see frametimes as the 0.5GB portion is being used, it seems to have minimal effect on fps.

 

[cmdrdredd]

Seems the benchmark was identifying a real-world difference between the way the 970 and 980 handle memory allocation. Validates what I was saying in point B including the fact that it may have a minimal impact real-world.

"drama made by bored people" Has motivated Nvidia to explain what was going on underneath the hood which is good for everyone.

It would be interesting to see frametimes as the 0.5GB portion is being used, it seems to have minimal effect on fps.

Logic? Sound reasoning? What sorcery is this?

Seriously though this makes sense. I just don't think the problem is as bad as forums would have you believe.

 

[Erenhardt]

My 2GB card was running games fine with vram maxed. Some memory swapping doesn't hurt performance much, maybe occasional stutter here and there. But if you go too far and some often used data needs to be swapped, then the stutterfest begins.

I think current games (aside from few examples) don't use enough VRAM to show have and impact. Nonetheless, 970 will begin stutterfest 0.5 GB earlier than 980.

 

[amenx]

Big question is... the 970s which have been out for 4+ months and sold in the thousands, why has this issue only came to the fore in last couple of weeks? If it was as bad as it is suggested to be, why wasnt it detected early on? Have you seen the 970/980 thread and how big it is? Any complaints there?

 

[cmdrdredd]

My 2GB card was running games fine with vram maxed. Some memory swapping doesn't hurt performance much, maybe occasional stutter here and there. But if you go too far and some often used data needs to be swapped, then the stutterfest begins.

I think current games (aside from few examples) don't use enough VRAM to show have and impact. Nonetheless, 970 will begin stutterfest 0.5 GB earlier than 980.

No it won't, we proved this by running close to 4GB and no stutterfest. Again, and I've repeatedly said this, if the bandwidth dropped off as much as this synthetic test suggests then the fps drop would be significant enough to show visually and you would not need to fcat it. You'd see it in real time.

 

[Cloudfire777]

Big question is... the 970s which have been out for 4+ months and sold in the thousands, why has this issue only came to the fore in last couple of weeks? If it was as bad as it is suggested to be, why wasnt it detected early on? Have you seen the 970/980 thread and how big it is? Any complaints there?

Because the issue doesnt exist and a german guy made a benchmark that threw out numbers which didnt apply and gave the wrong picture.
Nobody have noticed anything in games because the issue doesnt exist

My 2GB card was running games fine with vram maxed. Some memory swapping doesn't hurt performance much, maybe occasional stutter here and there. But if you go too far and some often used data needs to be swapped, then the stutterfest begins.

I think current games (aside from few examples) don't use enough VRAM to show have and impact. Nonetheless, 970 will begin stutterfest 0.5 GB earlier than 980.

Based on the "pitchforkers guide to GPUs?"

You have the PCPer link where Nvidia used over 3.5GB and had no issues.
You have a reddit thread posted earlier where people tested several games up toward 4GB and had no performance drop or issues

What more do you need?

 

[ShintaiDK]

I see the lack of endgame drama have made some people frustrated.

Flawed benchmark, works fine in games. No stuttering no, performance penalty. Just pure full 4GB enjoyment for the owners.


If you continue posting flamebait like this, you will be viewing these forums from the outside in.

-Rvenger

 

[Cloudfire777]

I see the lack of endgame drama have made some people frustrated.

Flawed benchmark, works fine in games. No stuttering no, performance penalty. Just pure full 4GB enjoyment for the owners.

I know several that expected Nvidia to go bankrupt and hope that they would do a recall on over a million GPUs because of this lol

Bet OEMs and hardware shops was answering thousands of phonecalls from frustrated 970 owners the last few days, just because of this benchmark made by this german fellow.

He must be proud of himself.

 

[Spanners]

I see the lack of endgame drama have made some people frustrated.

Flawed benchmark, works fine in games. No stuttering no, performance penalty. Just pure full 4GB enjoyment for the owners.

The benchmark highlighted a real-world difference between the 970 and 980, one that caused Nvidia to release a statement explaining the cause behind it. I wouldn't call that flawed. I'd say it's been quite useful.

Pure full 3.5GB + 0.5GB enjoyment?

Because the issue doesnt exist and a german guy made a benchmark that threw out numbers which didnt apply and gave the wrong picture.
Nobody have noticed anything in games because the issue doesnt exist

It gave the only picture it could, you couldn't expect the author to have knowledge of the underlying architecture when Nvidia never disclosed it.

Doesn't exist depends a lot on how you frame it, seems to have a very tiny impact on game performance but the issue/thread certainly hasn't been completely based in fantasy like some have implied.

 

[jj109]

I see the lack of endgame drama have made some people frustrated.

Flawed benchmark, works fine in games. No stuttering no, performance penalty. Just pure full 4GB enjoyment for the owners.

Actually not a flawed benchmark. This seems to be a CUDA bug.

http://www.computerbase.de/forum/showthread.php?t=1435408&page=7&p=16912375#post16912375

The last 500 MB isn't loading into the GPU's VRAM for some reason. If it can happen during a simple malloc + add test, I think it can happen during games. It would explain why some people report massive stuttering at 3.5G if anything above 3.5G is always being swapped over the PCIe bus instead of the 2nd partition.

 

[SteveGrabowski]

There is a 3% drop in performance from less than 3.5GB to more than 3.5GB VRAM usage which is neglible, and could very well be from increased textures and the fact that 970 have less TMU`s or was just a coincidence. Its certainly within margin of error

Its certainly isnt a DOOM situation where its 22GB/s bandwidth that drags the entire card down with it like many had hoped for because the benchmark showed a massive drop in bandwidth. And why did it show 22GB/s? Because the benchmark could only access 3.5GB and used the system RAM for the remaining 500MB. 1600MHz DDR3 have a bandwidth of 22GB/s. Flawed benchmark.

Its a non issue and it really doesnt exist

I'm glad to read this. I was pissed I got a 970 instead of a slightly cheaper 290x reading about this in the morning. I rarely play games on DSR (usually just use whatever settings GeForce Experience recommends) so I haven't come close to any of the memory limits playing at 1080p I think.

 

[garagisti]

So are you saying to disregard the entire tech world even if they were to do testing of the issue with clear and precise criteria/standards? That you will only take one answer and that it must fit your (and Charlies) viewpoint that Nvidia has knowingly and willfully screwed every 970 owner no ifs, ands or buts about it? You seem like a reasonable guy.

p.s. reminds of AMDZone where they dismissed all reviews of C2Ds, Sandybridge, etc as bought and paid for by Intel.

Right... so we already know you think bumpgate didn't happen. Great!

What are my views on 970? I have mentioned that it is a very good card, regardless of this issue. It was in a post just above yours suggesting what colours i wear. Am i wrong to think that it may have affected selling price of cards? No. Do i think silence till now was deliberate? Yes and 'am not alone.

Again, because you keep mixing things up, let me suggest that all of it is independent from the fact that 970 is a good product.

Also, your statement seems to suggest that trolls only exist on AMDzone... which is patently incorrect.

 

[Cloudfire777]

I'm glad to read this. I was pissed I got a 970 instead of a slightly cheaper 290x reading about this in the morning. I rarely play games on DSR (usually just use whatever settings GeForce Experience recommends) so I haven't come close to any of the memory limits playing at 1080p I think.

Good choice.
DSR and MFAA (which is an awesome feature to get good graphic and more FPS than MSAA) is something you will like.
Plus the GPU runs super cool.

I have GTX 970M SLI but MFAA isnt available for SLI and DSR still havent rolled out for mobile GPUs yet.

It gave the only picture it could, you couldn't expect the author to have knowledge of the underlying architecture when Nvidia never disclosed it.

Doesn't exist depends a lot on how you frame it, seems to have a very tiny impact on game performance but the issue/thread certainly hasn't been completely based in fantasy like some have implied.

But people and certainly the author should be able to draw conclusions that a 22GB real bandwidth performance over 3500MB must have been a big error from his benchmark.
Every reviewer that tested the 970 with 1600p should have been able to easily spot that. Not to mention users that push their rigs to the limit.

I agree that we atleast learned something about the 970 which we wouldnt if it wasnt for the benchmark. Thats the only positive I personally got from this

 

[guskline]

Ihave bothR9 290s and GTX 970. Happy with both.

 

[wanderer27]

One oddity from this test I haven't really seen addressed is the Bandwidth discrepancy between the 970 & 980.

They should be the same, but they're not.

Also it's been argued that the 970 has a 208 bit bus, but mathematically it looks like a 216 bit bus - bear with me.

For 980
6,000,000,000 * 256 / 8 /1024/1024/1024 = 178 GB/s

For 970 @ 208
6,000,000,000 * 208 / 8 /1024/1024/1024 = 145 GB/s

For 970 @ 216
6,000,000,000 * 216 / 8 /1024/1024/1024 = 150 GB/s


Now it's stated that this is a 256 bit card, and it is, but it looks to be split like this :

For 970 @ 208
208 bit => 145 GB/s
48 bit => 33 GB/s
256 bit => 178 GB/s

or

For 970 @ 216
216 bit => 150 GB/s
40 bit => 27 GB/s
256 bit => 177 GB/s (~178 when rounding accounted for)


I'm not sure if the upper segment is really running at this slower speed or is not really affected.


Anything to address here, or is this all irrelevant?




.

 

[Gikaseixas]

I wouldn't be so quick to side with or be against Nvidia here. It's early days so let's wait and see what happens in the next few weeks.

Lot of emotional posters defending / attacking NV, but i'm not really surprised

 

[Spanners]

Anything to address here, or is this all irrelevant?
.

Someone said earlier that it's shader bound until you reach the 0.5GB section. Hence the difference.

 

[tential]

I wouldn't be so quick to side with or be against Nvidia here. It's early days so let's wait and see what happens in the next few weeks.

Lot of emotional posters defending / attacking NV, but i'm not really surprised

Most of it has nothing to do with the issue at hand. It's just a "My daddy is better than your daddy" type fight going on.
Both sides just look for evidence that proves their point, and ignore all other evidence that doesn't.

 

[cmdrdredd]

Actually not a flawed benchmark. This seems to be a CUDA bug.

http://www.computerbase.de/forum/showthread.php?t=1435408&page=7&p=16912375#post16912375

The last 500 MB isn't loading into the GPU's VRAM for some reason. If it can happen during a simple malloc + add test, I think it can happen during games. It would explain why some people report massive stuttering at 3.5G if anything above 3.5G is always being swapped over the PCIe bus instead of the 2nd partition.

I'm afraid I don't trust these "people reporting massive stuttering at 3.5G". I and others here have tested for themselves. Sites have benched these cards in both single and SLI configs and never noted "massive stuttering".

 

[flexy]

Nobody have noticed anything in games because the issue doesnt exist

YES the issue DOES exist and Nvidia knows that!

The just released "damage control comparison picture" is meaningless. It shows average FPS.

THE ISSUE is that the 970 incurs stuttering and massive increase of frame-time once 3.2GB-3.5GB VMEM is exceeded. The damage control image does not show what the actual issue here is.

The problem is real, it doesn't exist on the 980.

 

[cmdrdredd]

>>
Nobody have noticed anything in games because the issue doesnt exist
>>

YES the issue DOES exist and Nvidia knows that!

The just released "damage control comparison picture" is meaningless. It shows average FPS.

THE ISSUE is that the 970 incurs stuttering and massive increase of frame-time once 3.5GB VMEM is exceeded. The damage control image does not show what the actual issue here is.

It is was massive, it would be noticeable visually. Stop saying massive, it isn't. Besides that you don't know that what you're saying is even factual. I'm sure in many minds they would love it.