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

[Sencha]

I'm not so concerned about the 3.5gb of ram its more that some cards seem to **** the bed once they go beyond that

 

[dewi]

Glad I am not the only one.....

I have a Ref 970 on Pre-order, now would I be better off cancelling that order and putting the money towards buying a Ref 980 or should I just stick with the 970.

Gaming is 24" @1080 with G-Sync Monitor, (being delivered today) if that helps any. But I want some future proofing as well.

Sorry if its off topic but I am that confused and really would like some peace of mind over it, as I don't have a lot of money to spare so I try to make the right purchase first time if I can.

I'm in exactly the same position. I also have an OcUK reference 970 on pre-order. I'm undecided whether I should stick with the 970, which I think would be fine for my current needs, or pay the extra for a 980 to get a more future-proof solution, given that future games will probably use more VRAM.

 

[CAT-THE-FIFTH]

I'm not so concerned about the 3.5gb of ram its more that some cards seem to **** the bed once they go beyond that

If its dependent on what part is disabled there are probably GTX970 cards which are fine at full 4GB VRAM usage.

Thats why we need to start that thread,once a reliable testing methodology is devised first.

 

[stooeh]

If you want to future proof wait for the 8GB cards

At the settings that require 4GB the frame rates are pretty low anyway.

 

[opethdisciple]

What hasnt been quantified is who much IQ difference 500mb actually makes?

 

[Gregster]

Think I was unlucky TBF! It was an awesome card for the time.

Yer, mine was a real step up from my previous card (hercules I think) and just kept on going and going.

James seems to have a decent game test going on, so people who are having issues, should try and replicate James's settings and report back. Use ShadowPlay to record perhaps or even screenshot the VRAM usage.

 

[almighty15]

If the GPU has 3.5Vram and another memory pool of 500mb is it still Vram or is it using normal RAM? like DDR3?
Is this why people experiencing slow downs? Like if you run out of RAM your PC will start to use the Hard-drive Page file and also slow down the system..

I still classed as VRAM.

It slows down because the bandwidth from this 512Mb pool is much lower then the rest.

You have:

3.5Gb @ 224Gb/s bandwidth
0.5Gb @ 45Gb/s bandwidth

That's why performance tanks as the that smaller pool completely bandwidth starves the card.

 

[mattcrane]

I'm in exactly the same position. I also have an OcUK reference 970 on pre-order. I'm undecided whether I should stick with the 970, which I think would be fine for my current needs, or pay the extra for a 980 to get a more future-proof solution, given that future games will probably use more VRAM.

I have pretty much decided to change the order to a 980.

Now I just need to decide which one to get Pallit,MSI or Zotac as I really need the reference cooler.

 

[tangey]

As a very poorly informed amateur, I've been reading this thread as I've just bought a 970 card.

It appears to me that the overclockers forum in general has many members that are highly proficient in GPU board / hardware details, advantages disadvantages etc etc.

It is also clear that the forum has even more people who are really into gaming, know which resolutions/feature sets/cards are best suited to which games etc, and are highly versed in configuring rigs for best performance.

Unless I've missed something, until that now famous memory access test was posted a few days ago, there was no perceived issue with 970 cards. The hardware guys didn't have an issues, and the gaming guys didn't have an issue. Also none of the well respected graphics card testers (3dguru etC) had an issue.

My conclusion as it stands is that this *may* be a theoretical issue with the 970 cards, which can be exposed by a test. But in practical terms the entire extensive Nvidia knowledge base on overclockers has not detected a defect in real use situations, unless attempting to subsequently contrive situations that might expose it. Interestingly, either no one generally uses these contrived situations, otherwise they would have already been talked about as being troublesome by reviewers/overclockers, or the issues now being seen and *NOW* being attributed to the memory access issue, were determined previously not to be a defect, but just a normal result of the card struggling in set-ups that pushed it to/beyond its limits.

At this time, I'm very happy that when I install my 970, I'll get the same great graphics performance the rest of you have been having since before christmas.

That's not to say I won't continue to watch this thread for updates and further Nvidia responses. It it turns out to be an issue that Nvidia ultimately accepts and offers redress, I'll be right in the queue along with the rest of you

 

[dewi]

Think of it like this: the 980 has 16 SMMs (Streaming Multiprocessor Maxwell). all 16 need to be in use at the same time to achieve the quoted bandwidth figures. Now, the 970 has 13 of them, the other 3 being disabled. Because of this, bandwidth and therefor memory performance is affected. It seems that nVidia have attempted to optimise bandwidth by segregating the memory in to two chunks, 3.5gb and 0.5gb, the 3.5gb chuck having a 'higher priority' according to nVidia. It seems like the right approach to me but it would appear to be causing some people more problems than others once you start using that smaller chunk.

This issue is muddied by the benchmark in the op, which apparently wasnt very well written and is possibly exaggerating the results, so further clarification needed there. In addition to this, there is a question over whether the various reporting programs (afterburner etc.) are reporting the right memory usage. It's possible that these programs are not reporting usage of the 500mb chunk, meaning people could be using 4gb of ram but afterburner etc reports <3.5gb. I'm not sure about this because we've had AB reporting 4gb and over so we need clarification there also; This would also be a good time for comments from the authors of these programs, such as Unwinder who wrote rivatuner (which afterburner is based on).

So, too many questions unanswered at the moment. But what does seem apparent is that these problems arent consistent from one 970 to the next. This could be down to which SMMs the 970 has disabled - most if not all 970s will be produced from rejected 980 cores, so disabling different parts of the GPU may have different affects on memory access.

Thanks, James. This is the best summary I've read.

I can't help feeling people are making a bigger fuss than is warranted. We always knew that the 970 had fewer SMMs than the 980. That was bound to limit performance in some way.

The benchmark is probably correct in showing that access to the upper 0.5GB is much slower than access to the lower 3.5GB. However, this doesn't necessarily translate into dramatically lower frame rates when more than 3.5GB is used, as the first 3.5GB of the VRAM will still be accessed at full speed.

I am a bit wary of Nvidia's performance data. When they say they've tested <3.5GB and >3.5GB, they could have tested 3.49GB and 3.51GB. What I would like to see would be a performance test using 3.5GB and 4GB, preferably by changing the texture detail and not the display resolution.

What does seem worrying is that the performance could be different on different 970s, depending on which SMMs have been disabled.

I'm still undecided whether to stick with my pre-order for an OcUK reference 970 or to pay the extra for a 980.

 

[Cooper]

This non issue still fuelling the internet with nerd rage I see.

 

[Rangerjr1]

This non issue still fuelling the internet with nerd rage I see.

Id be mad too to learn that i could only use 3.5 out of my 4gb of RAM until bad stuff went down

 

[Cooper]

Id be mad too to learn that i could only use 3.5 out of my 4gb of RAM

Except that's not the case is it...at least not on my system, in games that I actually play

 

[LambChop]

I still classed as VRAM.

It slows down because the bandwidth from this 512Mb pool is much lower then the rest.

You have:

3.5Gb @ 224Gb/s bandwidth
0.5Gb @ 45Gb/s bandwidth

That's why performance tanks as the that smaller pool completely bandwidth starves the card.

As much as I love Nvidia, if that is the case, that is a blatant advertising lie then as 4GB of mem does not run at 224 gb/sec

GTX 970 Memory Specs:

7.0 Gbps Memory Clock
4 GB Standard Memory Config
GDDR5 Memory Interface
256-bit Memory Interface Width
224 Memory Bandwidth (GB/sec)

 

[dewi]

I have pretty much decided to change the order to a 980.

Now I just need to decide which one to get Pallit,MSI or Zotac as I really need the reference cooler.

Out of the reference 980s, I'd choose the Zotac for the 5 year warranty. I'm also considering the Asus GeForce GTX 980 DirectCU II OC Strix, since I have a full tower case and I don't plan to use SLI.

 

[Rangerjr1]

Except that's not the case is it...at least not on my system.

Heh, ask Almighty.

 

[mattmos]

Not impressed with Nvidia's behaviour on this issue.

Its more of a problem for applications which use the full memory allocation. I've been rendering with Redshift3D, which before they implemented a hack fix, was running around 20% slower than previous generation cards, and now runs around 10% faster. The hack fix was to stop access to that last 500Mb of memory - so we effectively have only 3.5Gb usable, which is a real shame for a GPU renderer.

If they advertise these cards as 4Gb cards, that full 4Gb should perform at the same speed, not have a chunk of it which limps along bringing down the overall performance.

 

[Jinjur]

Has anyone from OCUK dropped any info in here yet?

If not, i guess they are sitting tight until an official line comes out, I guess that probably makes sense. It would still be reassuring to hear something......

J

 

[jwilliamson47]

Surely Nvidia are going to have to do more than their basic statement? I wouldn't be surprised if retailers are having to deal with people asking for refunds.

 

[Jinjur]

It could get messy I guess if nVidia are taking the line of "you want 4 gig? you got 4 gig, now ** No disguised swearing, all swearing must be fully starred out ** yourselves"

Im no sure how refunds could happen until nv accept a "problem" rather than the way it is configured.

This is going to be painful isnt it?

J