Middle Earth: Shadow of Mordor GeForce GTX 970 VRAM stress test
Click here to post a comment for Middle Earth: Shadow of Mordor GeForce GTX 970 VRAM stress test on our message forum
Fox2232
I know it and explained it several times int hat nasty long thread 🙂
Impact of this slow vram region would be very painful if you had 60+fps before filling it. (it will drop your fps to somewhere around 40-45fps even if game did solid 120fps before)
But all test are done at situations where games already run 25-40fps unstable by time card utilizes 3.5GB.
And since utilization is increased by means of putting another unreasonable load on GPU, it crumbles to 15-30fps where is part of this impact 0.5 of slow vram. (1 maybe 2fps here)
That is reason why test which HH in good will did is not way to show what this 0.5GB can do.
And that is why I wrote in 970 thread that people would be better not crossing 3.5GB ever.
I did 4 model situations how can nVidia approach this and one they are using is best in situation where they already sold cards as 4GB. I would presonaly made them with 7 (3.5GB) chips, sold them as 224bit and they would work. And would come cheaper.
Undying
From best card in last few years to scandal. Worst GPU configuration ever, this should've been 3.5gb card without 0.5gb slow ram, it only makes things worse.
Like HH said, lets hope we never see this crap again. 😀
Hilbert Hagedoorn
Administrator
And I'll say it again, your methodology is stupendous. Feel free to continue that discussion in the allocation thread, other then that I like this discussion thread back on topic now in direct relation to the article. Thanks.
omnimodis78
"After some internal testing here over the weekend we could quite honestly really reproduce stutters or weird issues other than the normal stuff once you run out of graphics memory."
So you could or couldn't? The way that sentence is put together implies that you couldn't reproduce stutters or weird issues. Please fix that if you meant to write couldn't
Hilbert Hagedoorn
Administrator
could not, let me look that up.
Primey0
You reached 60fps for 1 second lol. I wouldn't brag about that because you were constantly around 50 - 55
Mineria
That is how I understood it, even if you missed out with a "not" in the sentence.
I bought my card after checking out your review, since it seemed to be perfect for 1440p.
And it has been so far for even new AAA titles with all their shininess enabled, so I'm not to worried about the findings with exception that Nvidia should never release false specs for any of their products.
This one was easy to hid for them too, just imagine if we couldn't se the amount of memory on the cards and they had sold them as 6GB cards...
Hell would have broken loose. 😀
rl66
agree with you
about the "stuttering" it's the new "fancy" word.
like those with the I7 1366 extreme who were suffering from "bottleneck" with the 580gtx...
a lot talk about, but few have it 🙂
in 99% of the case the problem is due to other things (driver, heavy download in background (don't laugh i have seen it ), having a 10 years old HDD for system disk etc etc ... and of course the main cause of issue on computer are beween the sit and the screen ("this ati/nvidia card is so much powerfull than the ati/nvidia card i get 1FPS more in CoD/BF/WoT" or "i have a lot of stuttering i drop of 2 fps regulary").
NVidia have fail in com, for sure
but also when you buy a 970gtx you won't get a 980gtx despite they are very close. and at this level it is a nice performance/price/Watt despite the 3.5g +0.5g mem 🙂
Agent-A01
Uh that's obvious why isn't it?
Hilbert Hagedoorn
Administrator
Chaps, over in the VRAM allocation thread Guys, I've got somebody from Nvidia EU who will join up in that thread today.
The goal is to answer your questions and concerns, feel free to vent .. but do keep your comments respectable. Most of you have shown exemplary behavior and I respect that very much.
MrBenB is the verified representative from Nvidia.
omnimodis78
I have no issues with any of this. My 970 is getting better with each driver release and I thoroughly enjoy gaming with it. I do have one question though, you write that "...it should have been called a 3.5 GB card with a 512MB L3 GDDR5 cache buffer." Is this accurate? Correct me if I'm totally off the mark, but isn't this 512 not being accessed at all until the 3.5 GB point is reached, whereas the L3 cache would always be part of the overall equation, that is, the entire memory package would have access to the L3, but in this case, it doesn't. Am I right or not?
Solfaur
Well, since there has been more than enough trolling and flaming going on, I'll just say this: regardless of how nvidia handled it (and I agree with Hilbert, they did a poor job), the 970 is still the same card that got praised all over (including by me) 3-4 months ago.
Nothing changed, except the fact that we now know more of the card, from an engineering point of view. I still think it's a GREAT bang for the buck, because I genuinely am satisfied by the perfomance for the money I payed for them. I passed 3.5GB vram in several games since I play at 1440p with at least 2xMSAA and never have I encountered these issues.
Agent-A01
Correct...
The card will use ALL memory when it has to, it just tries to keep that last bit of memory segregated via drivers..
Tbh all this crap is overhyped, its still a 4gb card and in the situation where it is using all 4gb of memory on these new games its still gonna perform the same regardless if the last 512mb of vram wasn't slower.
If a games not playable on the 970, its not going to be playable on the 980, so as usual people will make something out of nothing/miniscule and blow it up.
GoldenTiger
Your testing is only showing you as using ~3.6gb of VRAM... this means that only a tiny fraction of the data being accessed would ever even be in the slow segment, and less frequently, compared to testing with more of the card's full capacity being in use. Having run a 4K native setup with SLI GTX 970 cards, my own experiences match up with PC Game Hardware's testing (http://www.pcgameshardware.de/Geforce-GTX-970-Grafikkarte-259503/Specials/zu-wenig-VRAM-1149056/) that shows many and frequent frametime spikes, where it was very obviously stuttering and didn't look smooth even with framerates remaining at 60+ the whole time in various games from BF4 to Shadow of Mordor and others.
I think the testing you've done, while appreciated, isn't representative of the issue at play here due to only barely going above the 3.5GB mark.
ivymike10mt
DSR / 4K or HQ textures users, with 970 SLI setups, may be bit dissapointed.. Coz 970 SLI May be enough for handle 4K almost easy.
Also upcomming games, will not use smaller amount of memmory.
Anyway. Many maxwells still have worse? driver TDR or voltage regulation issues.. (hopefully its bad drivers line). But its completly another topic.. :s
CK the Greek
Bravo for the test Hilbert.Nice news about a Nvidia's representative joining in.
I really hope that all of users will respect the main reason of all this and stay on subject.
----------------------------
I say again, the main worst thing about this is
1) any customer needs to know exactly what he will buy with no marketing hidden tricks
2) it will be a near future maybe-small-but-not-tiny issue in near future. An optimized game's engine will benefit for all Vram is available to use for caching,better quality stuff,better AA and if MFAA will be better and more supported(and it will be) still we will need to use some AA.It will be better for performance reasons to have some AA(like 2xMSAA+MFAA or even SMAA) up to 1440p(most) than higher resolutions without the need for strong AA method(or none) . Same for Nvidia's hairworks and all other *works.
970 is a very good card and I enjoy it now but I hate how they advertise(d) it. This method is never good for business future plans.
HunnoPT
GTX 970 technically has fewer ROPs than the GTX 980 as well as less than the full 2MB cache of its older brother. Basically, the GTX 970 only has 56 ROPs and about 1.7MB of L2 cache available to it. It can still hit the theoretical peak bandwidth of 224 GB/s but that is only while using both blocks of memory .
Advertise to be 4Gbs at full bandwitch, when its 3.5GB at high bandwitch and 0.5GB low bandwitch , advertise to be a 384 bit when in fact its a 224bit one , the ROPS to be 64 , when they are 56 etc..
And still people saying , its ok! It performs OK! Its like saying i had 500 bucks in my pocket, then i got robbed, but they only robbed me 100 bucks, so its OK!
I really hope that kind of attitude is not adopted by the massives, otherwise in the future, they will do the same thing, because ...its OK!
Denial
No it's not like saying that at all. People bought the card based on it's performance, not theoretical performance based on L2 Cache and ROP count. You aren't out any money or performance.
And it's obviously not OK, they need to get their marketing materials right. But to make it out to be this huge ordeal is absolutely ridiculous.
Loophole35
No it's like thinking you had $500 in your pocket when you only had $400 then claiming you got robbed. In the end you still only had $400 to start with.
Denial
Nah I think it's more like you have $500 in your pocket, you reach in and only see 4 $100 bills -- you panic then notice 2 - $50's.
People bought the card based on the performance, not the the ROP/L2 Cache count. The performance is the same -- how it's getting there is different.
