• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

970's having performance issues using 4GB Vram - Nvidia investigating

Status
Not open for further replies.
Buckster said:
I',m a bit confused - if the 0.5gb above 3.5gb is the same memory bandwidth as the rest of the VRAM

why partition it off ?

and why not on 980 ?
Click to expand...

See kaaps post above, it's down to how the disabled modules effect memory allocation/management on the 970, it's not an issue on the 980 as its a fully enabled part.

As a result the memory is allocated in two portions, a primary block of 3.5gb and a secondary block at 0.5gb, it would appear no monitoring programs can natively detect this 0.5gb making it appear the card has only 3.5gb usable vram.

It would also appear this memory management method gives circa 3% less performance then a card (980) with the full single memory allocation.

Of course this is the most factual (and repeatable) explanation yet, alternatively you could believe nvidia have intentionally ripped thousands of customers off, kicking a few puppies along the way :p

It always entertains me when one set of people instantly think the worst of one of these 'faceless corporations' to suit their own preferences and agenda.
 
eddyr said:
Nvidia release stated this is intentional - essentially to prevent reduced bandwidth from 0.5GB affecting performance all the time, by preferentially using the unhindered 3.5GB.Seems to be speculation, not sure there is any evidence. Or at least none presented in this thread. (CBA myself to trawl the other forums, not that interested tbh although the engineering aspect itself is interesting)

Perhaps a few % going by Nvidia's provided benchmarks.
Click to expand...

String and myself both ran the Shadow of Mordor bench using max settings @1080p. As you can see from the pics below GPUZ is reporting quite a difference in VRAM used between the two cards. I think this is down to GPUZ not seeing the 0.5gb portion on the 970.

String
970
vPhhURfl.gif.jpg


Myself
980
pITrgII.jpg
 
Just reading around the GeForce thread people are complaining of a lot more than 2%.

If you were to read Nvidia's response, which basically say's there isn't actually any difference in performance. you can't help but feel complainants are exaggerating something they wouldn't even see (In which case why are they complaining in the first place?), or Nvidia are not being entirely forthcoming.
 
Last edited:
The AnandTech guys state they'll look in to it, lets see what they come back with before getting the flaming pitchforks out, eh? ;)
 
So, as previously suggested, there is a by-design system to try and keep to the first segment - suggesting that they were well aware of the scenario. However, it looks like the impact of it is wildly variable, from some things having barely any hit as nVidia suggest and others having serious performance problems. At present I'm assuming the difference is how often the data in the 'overflow' 0.5GB needs to change as it seems to be a bandwidth issue more than anything.

Worth mentioning that people reporting problems have been saying that it appears jerky but still claims high FPS, so the 'look, FPS are not dropping' response doesn't really cover it. Perhaps frame timings are varying more or something? I've not got a 970 so can't try and do any analysis on it :(
 
Kaapstad said:
I think NVidia have got it about right with their statement.

What I think is causing a problem is monitoring software can only see the 3.5gb part of the memory and will report that the card is using that even if it really is using 3.5, 4.0 or trying to use more than 4.0gb. It is the last bit that I think is causing real world problems where the card tries to use more than 4gb and performs badly and users only see 3.5gb reported as used in something like GPUZ.
Click to expand...

Unfortunately that theory seems to be flawed, the monitoring software on gpuz appears to see the whole 4GB as per this image from overclock.net:

http://www.overclock.net/t/1535502/gtx-970s-can-only-use-3-5gb-of-4gb-vram-issue/530#post_23455333

Note it's measuring 3751mb in use.
 
Kaapstad said:
I think this is down to GPUZ not seeing the 0.5gb portion on the 970.
Click to expand...

I disagree. I think it is due to the benchmark not actually requiring >3.5GB as per the nvidia statement posted earlier:

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
Click to expand...
 
Has anyone actually been able to determine that of the 4gb of vram the 970 has, it uses the full amount of memory as per the cards specifications, I'm still not clear on this.

To me when I brought the card It looked (and still does) that all 4gb of memory should be running exactly the same, if it’s not then I will feel like a reasonably significant ‘feature’ wasn’t made obvious and should have been, even if that ‘feature’ does in reality only result in a small performance decrease.
 
Just to throw more speculation into the mix ;) .....how about this for an explanation.Taking shadow of mordor as a test case when Kaapstad said his 980 was using 4gb and at the same settings the 970 was reporting 3.5 gb,maybe (if we assume monitoring software is not reading the last 0.5gb correctly) the 970 WAS actually using 4gb,but because of this sectioned off 0.5 gb it`s just not showing.

If we then push beyond that by adding settings that would increase the 980 usage beyond 4GB,lets say to 4.3gb would it be possible that the monitoring software could potentially be reading this as only 3.8 gb on the 970,thus giving appparant stutter at less than 4gb.

This does sort of hinge on the 980 stuttering at 4.3 GB usage of course,but while everyone else is throwing all sorts speculation out there ....why not eh :D
 
Because it can. ;)

From the same nvidia quote 'but may report more for GTX 980 if there is more memory used by other commands.'
 
Surely it should just be a yes/no answer

does the card have 4gb of 224gb/sec of memory

the bandwidth includes both the physical design of the card and the chip memory speeds itself
 
pgi947 said:
See kaaps post above, it's down to how the disabled modules effect memory allocation/management on the 970, it's not an issue on the 980 as its a fully enabled part.

As a result the memory is allocated in two portions, a primary block of 3.5gb and a secondary block at 0.5gb, it would appear no monitoring programs can natively detect this 0.5gb making it appear the card has only 3.5gb usable vram.

It would also appear this memory management method gives circa 3% less performance then a card (980) with the full single memory allocation.

Of course this is the most factual (and repeatable) explanation yet, alternatively you could believe nvidia have intentionally ripped thousands of customers off, kicking a few puppies along the way :p

It always entertains me when one set of people instantly think the worst of one of these 'faceless corporations' to suit their own preferences and agenda.
Click to expand...

cutting down the SMX causes data path issues
yw! :)
 
Kaapstad said:
String and myself both ran the Shadow of Mordor bench using max settings @1080p. As you can see from the pics below GPUZ is reporting quite a difference in VRAM used between the two cards. I think this is down to GPUZ not seeing the 0.5gb portion on the 970.
String
970
vPhhURfl.gif.jpg


Myself
980
pITrgII.jpg
Click to expand...

Perhaps. Alternatively the 980 could be preemptivly caching or holding unneeded data which the 970 is actively clearing in order to avoid using the final 0.5GB.
 
Status
Not open for further replies.
Back
Top Bottom