HardOCP looking into the 970 3.5GB issue?

[Deleted member 174368]

Man, it's really weird that absolutely no one from [H] has so much as commented on this. I honestly though they'd be one of the first ones to start an investigation and post FCAT results.

Even GameSpot of all places ran a story on it.

 

[Cali3350]

Man, it's really weird that absolutely no one from [H] has so much as commented on this. I honestly though they'd be one of the first ones to start an investigation and post FCAT results.

Even GameSpot of all places ran a story on it.

Its disconcerting to say the least. Even a "We dont think its a deal" comment is at least something if that's how they feel.

 

[evilsofa]

I'll just repeat what I said earlier in the thread:

Maybe - just maybe - some publications are noticing that their readers have the pitchforks and torches out, and know that if they don't get their facts and tests all perfectly done, they'll be burned in effigy for it. So either they're staying out of the fray entirely, or they are taking the time to do proper analysis, which takes longer than a couple of hours on Monday afternoon.

 

[n=1]

Bad news, it seems Nvidia is backpedaling on the refund stance and believes there is no problem:




Peter Sierant, Senior Director of Customer Care at Nvidia edited his original post.

Original:


Updated:

You dun goofed nVidia, you dun goofed.

 

[Damar]

I'll just repeat what I said earlier in the thread:

Maybe - just maybe - some publications are noticing that their readers have the pitchforks and torches out, and know that if they don't get their facts and tests all perfectly done, they'll be burned in effigy for it. So either they're staying out of the fray entirely, or they are taking the time to do proper analysis, which takes longer than a couple of hours on Monday afternoon.

This is sort of what I've been thinking.

You've got the two extremes... one saying it matters a ton, one saying it doesnt matter at all, and then a few in the middle saying it "might matter now and then".

All I know is that my 970's have worked great since I got them. Mostly I play FFXIV with them and I've not had any issues at all while playing, and an MMO is the type of game that would show framerate drops pretty badly if something was up esp when set to "Maximum".

When I played Shadows of Mordor, I didnt see a problem either, and I had the settings up there as well.

While its pretty plain NV didnt list the specs as they should have, I cant at all knock the performance the 970's have given me for what I paid for them.

If [H] puts out an article on the whole mess I'll gladly read it and see what the real story is, since they're about the only ones I'd trust to give me the info and not blow any of it out of proportion for page hits like other sites seem to be doing.

 

[n=1]

I think the ones saying it doesn't matter are likely only running a single 970, and at which point the 970 runs into the vram wall most of the time they're getting below 30FPS anyway.

The ones saying it matters a ton are probably those running 970 SLI (or tri-SLI) and gaming at 1440p or 4K, where vram because critical because there is now enough GPU power to push 60 frames, but the card could be held back by the vram wall.

 

[Meeho]

 

[GoldenTiger]

I believe that I've called you out over bias before but after seeing your posts on this issue I feel like I should follow the example set by a couple others and mention that I've been impressed by your ability to call it like you see it and admit that I was wrong about you. I still think that some of your past comments have shown bias but a little bias based on personal experience is natural and you're clearly NOT the fanboy that I thought you were so I apologize for any rudeness.

I don't have a 970 so I don't really have a dog in this fight but it's the first Nvidia card that I've seriously considered in a few years due to it's price/performance ratio so the subject is of interest to me. I decided to try and hold out for and affordable(<$500) 6gb+ card, I tend to hold on to cards for a while these days so VRAM is important to me.

Thanks for the post... I really appreciate it.

 

[jackstar7]

tweets

smh

I thought nV had the superior marketing and PR to AMD, but they are really not showing me a compelling reason to jump to their product line.

 

[Ocellaris]

This is the best one. Don't worry guys and girl, Nvidia is on your side even though they sold something with incorrect specs and won't take any action to help displeased owners.

 

[jackstar7]

Thanks for the post... I really appreciate it.

Along the lines of other posters I'm going to say that I'd added you to my ignore list, but people quoting your recent posts and showing your about-face has brought you back into my regular viewing.

Good on you.

 

[Meeho]

Several people report memory usage going towards 4 GB even on 1080p with longer play on newer titles. I wish some sites would test that scenario too.

 

[Ocellaris]

Several people report memory usage going towards 4 GB even on 1080p with longer play on newer titles. I wish some sites would test that scenario too.

You don't need more sites to go out of their way to test things for you so you believe it. There is already plenty of information out there for you to look at.

 

[rinaldo00]

It's not that simple. I did some more testing using Shadow of Mordor, SLI enabled and disabled, sub-3500mb and over 3500mb VRAM consumption. Frametimes stay within normal variation/acceptable consistence when in single-card mode regardless of VRAM, but going to over 3500mb in SLI causes wild and rampant stutters/hitches with vastly fluctuating frametimes to match.

What are you using to record frametimes, FRAPS?

 

[GoldenTiger]

What are you using to record frametimes, FRAPS?

MSI Afterburner, which does it the same way FRAPS does. Judging by what I see on-screen during these tests, the data appears fully accurate to me.

 

[dr/owned]

smh

I thought nV had the superior marketing and PR to AMD, but they are really not showing me a compelling reason to jump to their product line.

Drivers that were made in this century?

Let's not forget AMD's year long crossfire stuttering fuckup before we get out the torches and pitchforks for NV.

 

[Meeho]

You don't need more sites to go out of their way to test things for you so you believe it. There is already plenty of information out there for you to look at.

Oh, I am aware of the issues. It is for those that don't believe anything not written by big sites. People should remember how CF stuttering issues went on for years without coverage from most of them, with mostly only user reports (kudos to HardOCP on that one).


Some more 970 SLI reports:
Dying Light 970 VRAM High Vs Medium Textures
https://www.youtube.com/watch?v=a8Q6jmg_qik

This is exactly what happens and I get a huge judder/stutter when the game hits its max VRAM and tries to load in additional textures. The same happens with FC4 and Assassin's Creed Unity.&#65279;

I have the exact issue on my SLI GTX970's as you in Dying Light. Performance is all over the place with max settings when I hit the 3.5GB limit. When lowering textures to medium all good as the vram is below 3GB.
Tested it on Shadow of Mordor, Far Cry 4 and Lords of the Fallen with exact same issue only solvable by lowering textures or details to not go close to the 3.5GB VRAM.

 

[skizzled]

This is the best one. Don't worry guys and girl, Nvidia is on your side even though they sold something with incorrect specs and won't take any action to help displeased owners.

This is... I don't even... UGH.

They have yet to prove they are on anyone's side except their own. At first, it seemed like they may take care of it but now, it seems like it's full-on denial mode that anything could possibly be wrong with their product. Typical.

I'm still undecided on whether or not I'm going to bother returning this thing. Current performance is amazing for the games that I play but at the same time, what happens when I absolutely NEED 3.6GB+ of VRAM? Maybe I won't even actually need it in this card's lifetime. Who knows. But, it's starting to lean me toward returning it based on principle alone.

 

[Bladestar]

I'll just repeat what I said earlier in the thread:

Maybe - just maybe - some publications are noticing that their readers have the pitchforks and torches out, and know that if they don't get their facts and tests all perfectly done, they'll be burned in effigy for it. So either they're staying out of the fray entirely, or they are taking the time to do proper analysis, which takes longer than a couple of hours on Monday afternoon.

They need to state what they are doing/waiting for.

A simple "We are looking into it" is enough for now.

Staying silent screams of Nvidia is doing everything they can to hide this issue. If that means going after sites by blacklisting if they don't tote the company line, like they have done in the past over bad press, remains to be seen.

 

[jackstar7]

Drivers that were made in this century?

Let's not forget AMD's year long crossfire stuttering fuckup before we get out the torches and pitchforks for NV.

Drivers are a non-issue. And the stuttering is now better on AMD 29x series than on nV.

If nV can make similar changes and improvements coming out of this, then we all win.

 

[Dahkoht]

Good god , I would have thought Nvidia would have much more business savvy in the marketing department than what they are doing now.

From editing a post where the PR Mgr says he'll help with refunds originally , to the "it's awesome and we are on your side" tweets , it's a 101 in how to not handle a situation like this.

Oh and as aside , GoldenTiger , it sucks you sent your Acer 32" 4k you got super cheap back thinking the issue was with that and looks like it might be the 970's with the issue, I'm still in waffle land on which 4k to get and also whether anything is on the horizon better than the 980 sli (ATI or Nvidia) in the next few months for 4k.

 

[Rizen]

Several people report memory usage going towards 4 GB even on 1080p with longer play on newer titles. I wish some sites would test that scenario too.

This is hard to quantify though because some game engines reserve as much VRAM as possible, even if the game is not actually use that space. Frostbite (BF3/BF4 engine) does this to an extent. I remember when I upgraded from a card with 1.5GB of VRAM to a 2GB card in BF3, my VRAM usage showed 2GB, even though I was running the same settings and nothing else had changed.

So when you test for VRAM usage in scenarios like what's going on with the 970, you need to make sure the card is actually using that memory.

 

[Yakk]

Staying silent screams of Nvidia is doing everything they can to hide this issue. If that means going after sites by blacklisting if they don't tote the company line, like they have done in the past over bad press, remains to be seen.

*sitting looking at tinfoil hat*

Reading at the bizarre results popping up on some sites, and the really hard *they lied, but that's OK, really!* spin some other nvidia bastion sites are doing, I'm starting to be inclined to wear the hat.

 

[Meeho]

This is hard to quantify though because some game engines reserve as much VRAM as possible, even if the game is not actually use that space. Frostbite (BF3/BF4 engine) does this to an extent. I remember when I upgraded from a card with 1.5GB of VRAM to a 2GB card in BF3, my VRAM usage showed 2GB, even though I was running the same settings and nothing else had changed.

So when you test for VRAM usage in scenarios like what's going on with the 970, you need to make sure the card is actually using that memory.

True, but the problem is they also start to experience stuttering at that point, so it is actual RAM usage, not just allocation. Any frametimes change at that point would show if there is an issue.

 

[rinaldo00]

Man, it's really weird that absolutely no one from [H] has so much as commented on this. I honestly though they'd be one of the first ones to start an investigation and post FCAT results.

Even GameSpot of all places ran a story on it.

Kyle has said in the past that [H] does not have the equipment to do FCAT testing.

 

[MavericK]

Wow, from the look of those Tweets, nVidia is backpedaling FAST.

 

[polonyc2]

I'll wait to play 'The Division' on my 970 at 1920 x 1200 before whipping out my pitchfork

 

[Meeho]

Wow, from the look of those Tweets, nVidia is backpedaling FAST.

Arlt joins the likes of Amazon, OcUK, Caseking...taking the matter into their own hands and doing right by their customers, despite Nvidia's continuing lack of support.



Business as usual for Nvidia, it seems. Google translate:
Heise.de - Comment: Nvidia cards dirty tricks

Nvidia GTX-970-cheating is just the tip of the iceberg: In the past, the company kept falling on by questionable shenanigans. Nvidia has finally reputable and trustworthy, commented Martin Fischer.

The graphics chip maker Nvidia behaves always wrong. Of course, praise all companies have their respective products in the skies, but Nvidia takes it strikingly with the truth often not accurate enough. Nvidia has finally become trusted - and honestly communicate.

The jewel with the GeForce GTX 970 is not an isolated case, but just the latest link in a chain of events that customers and developers angry. End of 2012, Nvidia led purchaser as to the alleged DirectX 11.1 capabilities of Kepler graphics cards astray. As Nvidia Tegra 3 2011, announced its performance should reach five times the predecessor. But were meant double CPU and GPU performance three times - the Nvidia expects to flat five high. Such computational tricks remember the business practices of dubious street vendor.

2010 held Nvidia CEO Huang proud of the supposedly first Fermi graphics card in the air a little later than zusammengedengelte with wood screws dummy was exposed.

Outweighed the numerous, but Nvidia never actually declared failures of mobile graphics chips G84 and G86 series. Only the buyer renowned notebook brands were half-hearted substitute, owners of small producers let Nvidia in the rain. To date, technician repair hot air soldering machines affected mobile computer. Public product lists errors as they maintain AMD and Intel least for many of their products are reasonably careful, you will not find at Nvidia.

It it does not have to like Linux inventor Linus Torvalds hold in his 2012er Wutrede stuck his middle finger and Nvidia as "by far the worst company" titled, with whom he had ever dealt with. But Nvidia should not be surprised that self-satisfied GTX 970 owners are pissed off about such behavior.

Nvidia has to understand the current turmoil as an opportunity to rethink the communication strategy. And above all, the company must continue to simply honest with those behave that revealed the memory of the GTX 970 jewelry - namely true GeForce buyers who put more than 300 euros for such a card on the table. But perhaps more than 70 percent market share in the graphics card of the company are already too gone to his head.

 

[Meeho]

 

[n=1]

Someone should make a meme out of nVidia's tweets and make them go viral

 

[Meeho]

Preferably pasted over frametime stuttering graphs, like GoldenTiger's one.

 

[teiresias]

I love how the twitter account suggested someone experiencing problems with stuttering on a 970 install Geforce Experience to get "optimizations." This nonsense is pathetic.

 

[Meeho]

 

[n=1]

Well done. Now we just need to make them go viral lol

Btw there's at least a sliver of silver lining to this shitty situation: look at this piece that PCWorld ran. Obviously the entire piece was carefully worded to put as positive of a spin on it as possible, but nonetheless the title cuts straight to the chase "Nvidia employee retracts comments about planned GTX 970 driver to improve memory allocation performance".

 

[Meeho]

Yeah. They are pretty pissed at them over in Germany, too. Hopefully more bad press will force them to do something right.

To remind people why this is such a big issue, besides performance problems, here is an example from Anandtech's review:

Compared to GTX 980 and its full-fledged GM204 GPU, GTX 970 takes a harvested GM204 that drops 3 of the SMMs, reducing its final count to 13 SMMs or 1664 CUDA cores. It is otherwise fully enabled, keeping the same 64 ROPs and 256-bit memory bus as its bigger sibling.

Along with the reduction in SMMs clock speed is also reduced slightly for GTX 970. It ships at a base clock speed of 1050MHz and a boost clock speed of 1178MHz. This puts the theoretical performance difference between it and the GTX 980 at about 97% of the ROP performance or about 79% of the shading/texturing/geometry performance. Given that the GTX 970 is unlikely to be ROP bound with so many ROPs, the real world performance difference should much more closely track the 79% value

Elsewhere the memory configuration is unchanged from GTX 980. This means we&#8217;re looking at 4GB of GDDR5 clocked at 7GHz, all on a 256-bit bus. Compared to the GTX 770 that the GTX 970 replaces, this is a welcome and much needed upgrade from what has been the 2GB VRAM standard that NVIDIA has held to for the last two and a half years.

 

[{NG}Fidel]

I cant believe the approach they are taking now...
Good on that german retailer for doing the right thing by their customers.

 

[Meeho]

Can Nvidia really pretend like nothing is wrong?

A poll from ComputerBase.de:

 

[Meeho]

These are gold Found them on Neogaf.

 

[Meeho]

More tests found on Neogaf. Look at the frametime variance and also notice how the GPU is stalling while waiting for memory data.

Just in case there's anyone out there who isn't tired of frametime tests:

Assassins Creed Unity test:

4690K @ 4400
GTX 970 G1 @ 1595/7400

Two runs, about 8-10min each, did some parkour in the same area (tried to stay into populated ones), killed some templars, helped some guys being F*ed up by nvidia...i mean...thugs, opened some chests, all of that.

First run at 1440p and second at 1080p. All settings besides resolution remained the same.


note: on the 1080p run I opened the menu twice, hence the two frametime spikes (the menu is pretty damn heavy). On the 1440p run I remembered and kept it on game play all times.

alright guys, last one from me on this unless anyone needs anything specific. i wasn't too interested in waiting to see what nvidia had to say any longer after finding out that about the post where the guy at nvidia said he'd work with that one individual to get a refund and the nvidia announcement that they're going to try to improve the hardware issue through drivers. ever since i bought the card the option to upgrade to the 980 was on the table and i took it. i may burn myself down the road for $100 or so MAYBE, if nvidia offers like a cheap upgrade program or anything like that, which i doubt, but i felt way more comfortable returning the card within my local store's window and upgrading to the 980. i haven't had a chance to test it extensively but from what i have been able to do: WOW, what a difference. comparing my 1460MHZ OC'd 970 to the stock 1316MHZ 980. both MSI Gaming 4g cards Shadow of Mordor @ 1080p/60; every setting maxed w/ HD texture pack GTX 970

970

980

edit: i shrunk the graph down. it looks like there was something maybe wrong with the monitoring as it doesn't really show ANY movement in gpu/mem usage for a long period of time. i'll redo one tmrw and throw it up. final thoughts remain the same.

the stuttering and tearing are gone and the card now uses the full bank of memory. there's not much more i can say about that. Shadow of Mordor is the only game i'm aware of right now where you can hit the 3.5GiB mark without affecting performance too badly. if games stay on this trend and barring nvidia doing some serious magic with the driver update, the 970 is not where you want to be going forward, from my experiences with it.

 

[chenw]

TBH, from his graphs, 980 isn't fairing much better. from his settings 970 was stuttering at 3.5gb mark, and 980 was barely under the 4gb mark, anymore than that 980 isn't much better off.