• 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.
I would put me off if i was looking to get a couple of cards to try DSR for 4k. I would deffo hold back on the trigger until i got some good proof and test done to prove otherwise not saying or supporting the case for, just saying if it was me i would be slightly put off yes.

I still think this has been well blown out of proportion and is no where near as bad as it sounds.

If I was buying 2 cards for 4k or to try dsr/avr I would be buying a Titan or 2 290x 8gb or at a push 2 980's not 2 970's. Just my opinion.
 
You would know how frames tank badly whilst it access's the page file. Ig that guy who tested ran out of VRAM, it would tank and show on those graphs.

if that test is just to say it hasnt run out of vram i agree it proves that
it just does not prove there isnt a performance problem with the last 500mb
a small portion of the vram being slow is big difference to no vram!

and if you are only using 200mb of that slow ram that is accessed very randomly you wouldnt be able to tell much on a 2min test with not much to compare it to, thats my point
my guess is it would show up as stutters or more frametime but nothing as dramatic as running out
 
if that test is just to say it hasnt run out of vram i agree it proves that
it just does not prove there isnt a performance problem with the last 500mb
a small portion of the vram being slow is big difference to no vram!

and if you are only using 200mb of that slow ram that is accessed very randomly you wouldnt be able to tell much on a 2min test with not much to compare it to, thats my point
my guess is it would show up as stutters or more frametime but nothing as dramatic as running out

Would the frame time not suffer when it is using the supposed high latency memory?
 
Would the frame time not suffer when it is using the supposed high latency memory?

Yer it would and would show on the graph. Some just seem to want to keep seeing something that isn't there. Nobody has said what games bring the issues and I am bored of going over the same thing time and again, so I will drop out until something new comes up.
 
hes a very frustrating man!
he looks at a 2min graph and goes "well that proves that no performance issues here!! job done!" and im the one who is closed minded!

im going to look for cake! :mad:
 
What would you suggest then Mei?
Seems rude me poking my head in this late into the thread, but been busy :p

I haven't read the thread since around page20ish, are people still arguing over whether the 970 has 3.5gb 'usable' or whether the other 500mb is slower?
 
So, what I'm reading on the Nvidia forums and elsewhere, in terms of what people are saying, can be summed up thus below... and it SOUNDS like a good argument, but is this totally incorrect...?

Basically, the 970 was sold as having 4GB of VRAM at 224Gbps, as seen in their official specifications. Is that true? Because it would APPEAR (based on certain tests which may be wrong) that the VRAM is essentially divided in to two... one 3.5GB running @ 224 Gbps and one 512MB @ at 52Gbps. Is this totally false?

I would say it is totally incorrect.

970-specs.jpg


Taken from here

This shows the specs of the card, it does have 4GB of ram, it does have 224GB/s of bandwidth.

If people want to argue that it doesn't have 224GB/s over the entire range then good luck with that, it is just as mad as the point I tried to make earlier that a lot of video cards will throttle below their listed base clocks when using something like furmark, is that grounds for a replacement?
 
If people want to argue that it doesn't have 224GB/s over the entire range then good luck with that, it is just as mad as the point I tried to make earlier that a lot of video cards will throttle below their listed base clocks when using something like furmark, is that grounds for a replacement?
I see what you're saying, but the card isn't throttling, rather the way the card has been designed and what's claimed in the specs. Assuming certain people are correct in their assertions, the card sees a performance hit when it goes over 3.5GB and accesses the remaning 500mb. Now, this may be wrong as others say, but if true then it would SEEM to suggest that the 224GBs claim is false, and it SHOULD come with a proviso that this speed is only guaranteed for 3.5GB of RAM, and the remaining 12.5% (500mb) is not. But, as I say, I don't know really know what's true.
 
What would you suggest then Mei?
Seems rude me poking my head in this late into the thread, but been busy :p

I haven't read the thread since around page20ish, are people still arguing over whether the 970 has 3.5gb 'usable' or whether the other 500mb is slower?

i would suggest less insane and more cake for me :D
mostly if its slower/restricted i think
but can pick your argue!
 
I see what you're saying, but the card isn't throttling, rather the way the card has been designed and what's claimed in the specs. Assuming certain people are correct in their assertions, the card sees a performance hit when it goes over 3.5GB and accesses the remaning 500mb. Now, this may be wrong as others say, but if true then it would SEEM to suggest that the 224GBs claim is false, and it SHOULD come with a proviso that this speed is only guaranteed for 3.5GB of RAM, and the remaining 12.5% (500mb) is not. But, as I say, I don't know really know what's true.

But you are taking the 224GB/s figure as an absolute, that all memory access is at that rate. if that is the case then the base clock is an absolute as well and if when working flat out it drops below that then it is faulty, hence the furmark reference.

Or of course, much more realistically the 224GB/s figure is up to 224GB/s, in which case Nvidia have stated nothing wrong in the specifications, because the card does have 4GB ram and can access up to 224GB/s.
 
I know Ubi titles aren't anything to go by but I do think something's up with vram performance on these cards.

Unity was hitching like a boss with TXAA. I even got a two-second freeze when I tried turning around rapidly. GPU-Z reported vram load at 3537mb and that was the max it would show so I haven't the faintest idea how much the game was actually chewing up.

Far Cry 4 used about 3300mb with TXAAx4 and felt rather jittery when running around Kyrat (it was fine in the Himalayas sections). It didn't felt nearly as bad as Unity but something definitely felt off. The fps in both games hovered between 32-45 fps so still "respectable" but it seemed lower than that. The thing is, I can't really put my finger on whether I just became so sensitive to sub-50 fps that I'm making **** up or whether something's really wrong.

I really like this card so I hope it'll be able to cope with vram-heavy future releases, at least through 2015.
 
This really sucks... I was playing star citizen 1440p maxed out and I am getting stuttering every 30 seconds or so, checking MSI Afterburner my VRAM hits about 3580mb usage and I get a big dip in GPU usage, which in turn causes the stuttering. I would have got a 980 if I knew this! The card is as good as useless if you want stutter free gaming in Star Citizen and newer high res texture games going forward.
I hope this issue can be resolved in some way or form, or I will be moving back to AMD next time, not good Nvidia...
 
But you are taking the 224GB/s figure as an absolute, that all memory access is at that rate. if that is the case then the base clock is an absolute as well and if when working flat out it drops below that then it is faulty, hence the furmark reference.

Or of course, much more realistically the 224GB/s figure is up to 224GB/s, in which case Nvidia have stated nothing wrong in the specifications, because the card does have 4GB ram and can access up to 224GB/s.
That may be, and such an explanation could be how they side step any claims of fault. But if it turns out there IS a performance hit within that 500mb (under 4GB of course), in certain circumstances as a result of them building the card this way (as many people DO seem to be reporting, accurately or not I don't know), then they will find themselves utterly pilloried by their customers, and will end up sending many AMD's way in the future I am sure. But we shall see how it unfolds.
 
Ran the rest on my other Asus Strix 970. It drops off to 22gb/s at only 2.9gb ffs :(

In Star Citizen and Far Cry 4 the performance hit is massive at 2.9 onwards.
 
Last edited:
They will be pilloried (attacked ridiculed) for the 970 from this point forth, regardless of what happens now. The damage has been done, mainly by the mass of scaremongering that has gone on from the moment this story first broke.
 
Status
Not open for further replies.
Back
Top Bottom