Can somebody explain to me how the fuck it was possible that the "muh 3.5gb" issue of gtx 970 got revealed after a fucking HALF OF A YEAR after it's debut?
>>53309000
With a new product, people do not know what performance to expect, any drops they see they usually put down to immature drivers. And, ironically enough, no one expects a company like nVidia to lie to it's customers and investors.
Only once people star to mod games do they then pay attention to how much vRAM is being used should there be significant stuttering. And only then, a handful of people will ever notice a trend and investigate further.
>>53309000
I had GTX 970 from MSI. That card would not be bad if they left it with 3GB or 3.5GB of VRAM, but thanks to the way that memory is split it will start shitting itself once it reaches 3.3GB +/-.
Its defective product that loses performance once you reach certain amount of VRAM and they will not lock it to 3.5GB because that would mean massive law suit incomming. Just stay away from that card and pick 390/390x or 980
>>53309000
jews
The hilarious thing it's that people still buy them when 14nm is close and when the r9 390 is miles better than this shit
>>53309223
390 would be good if AMD sticked with 4GB VRAM and crawled under Nvidia prices, but nah they had to slap HBM onto it to match Nvidia prices.
>>53309240
Would you care to revise your post? Your facts seem to be a little off.
>>53309240
>>53309259
In which part exactly? Also I did not state it as fact but rather as opinion
>>53309000
The card technically has 4GB of RAM, it's just that the memory controller for the remaining 512MB was chopped up since the die is literally cut and not disabled cores, and thus has abysmal bandwidth. Since graphics performance is a lowest common denominator deal, when you reach that part of the RAM, everything slows to a crawl.
Without extensive performance testing however its not possible to tell if anything is wrong, and truthfully the hardware isn't really powerful enough to push datasets as large as 4GB under normal circumstances. People only started noticing this problem on games that are RAM hogs or when trying to push high resolutions in SLI, as you'll never see this happen at 1080p in 99% of games.
>>53309337
>you'll never see this happen at 1080p in 99% of games
Dying Light, the card is shitting itself in that game especially in The Following
>>53309289
>In which part exactly?
Right here.
>>53309240
>but nah they had to slap HBM onto it
390 is 8GB GDDR5. As is the 390X.
The 290/290X both have 4GB and 8GB variants. Both GDDR5.
If you're attempting to loosely refer to the Fury series of cards, they're a different architecture to that the R9 3** series cards.
Oh, and also here;
>390 would be good if AMD sticked with 4GB VRAM and crawled under Nvidia prices
The 390 does exist with 4GB GDDR5. It's called an R9 290. And AMD did price it lower than the comparable nVidia product.
The only part of your post that can be passed off as opinion is;
>390 would be good if AMD sticked with 4GB VRAM
Even then. With 8GB GDDR5. The 390 is still cheaper than a 970 (Depending on country of course, prices will always vary)
>>53309351
Yeah, Dying Light is one of the exceptions due to the retarded amount of framebuffer effects is uses, I'm sure modded Skyrim/Fallout 4 would pose a problem too.
>>53309364
I though that 300 series was fully equipped with HBM...I am really retarded then...
>>53309082
I bought it in the launch day
>>53309390
>I though that 300 series was fully equipped with HBM
No.
Not at all.
>>53309392
After 3.5+0.5 was revealed you could easily get a full refund based on false marketing claim.
>>53309386
Witcher 3 had some problems to what I'm told.
As does Shadow of Mordor (When running the HighRes texture pack)
Skyrim most certainly does give problems if you go crazy with ENB's, but generally you can keep that game under 2GB usage when 'heavily' modded (IE, 300 or so mods, maybe 2 ENB's)
GTA V gave people problems, then they realised they can turn down settings.
Metal Gear Solid: Phantom Pain ran fairly high in vRAM usage, some areas would cause stuttering.
>>53309425
You could from the aftermarket retailers, but nVidia were, and still are refusing refunds themselves.
>>53309451
Surprisingly Witcher 3 never went over 2-2.5GB VRAM in 1080p. It was smooth after you lowered shadows by one degree. And ofc you had do disable Hair Works.
>>53309499
HairWorks is expected to fuck everything up.
Who the fuck insisted that you need 64x tessellation on hair, and grass?
Past 8x Tessellation only serves to reduce performance in increase vRAM usage. There is no visual difference.
>>53309515
Well people claim that it was made to compromise AMD card but lets look at it this way:
Isnt it like High End card feature (980Ti, Titan X)?
Also you can disable it completely
And wasnt Tessellation main feature of DX11?
>>53309499
This
im playing stable 60 fps on max without hairworks
>tfw sapphire nitro r9 390
always glad im amd master race
>lower temperature
>lower price
>performs better in all games except some on which nvidia shilled
>better on dx12 by far and on 1440p
>>53309545
>Isnt it like High End card feature (980Ti, Titan X)?
Hardly. nVidia incorporate dedicated Tessellation units in products throughout their line.
>Also you can disable it completely
You can disable hairworks, yes, that is a godsend for everyone.
>And wasnt Tessellation main feature of DX11?
For the most part, but to go over the top with it serves only reduce performance past a certain point with no visual improvement. Ideally, 8x is the cutoff, it provides a clear visual benefit over 4x, while not giving a massive hit to performance.
>Well people claim that it was made to compromise AMD card
They aren't far wrong. AMD not including dedicated tessellation units meant that something like Hairworks which forced 64x would indeed harm performance of an AMD card moreso than an nVidia one. Which obviously, something like that would market itself.
>>53309000
is it the evga model only
>>53309584
Its stupid to talk about DX 12 yet because:
1. There are no DX12 games (You cant really count this "we have two DX12 features" marketing bullcrap)
2. 900 a 300 series are rather bridge between DX11 and DX12
3. Polaris and Pascal will show, who will be leader of starting DX12 generation.
>>53309000
Because cases where you would actualy encounter this as an issue are quite rare
Not many people run 4k yet or skyrim with heavily modded shit
>>53309515
nah you can see the difference between 8x and 16x; 16x to 32x is much harder to see but still possible with direct comparisons; 32x to 64x is pretty much pixel equal.
I should note the DX11 spec has a maximum of 64 layers of tessellation. They never expected anyone to reach or exceed it 7 years ago when the spec was written.
>>53309000
Because it isn't as likely of an issue as people make it out to be. You rarely exceed 3gb of vram usage anyway.
>>53311248
Having to worry about using more than 3/4 of your card's VRAM
NVidia shills actually defend this.
>>53309635
>>53311248
Nvidia Official shill and samefag detected.
>>53311464
>>53311446
I'm not defending it. I was stating an objective fact. I'm just giving a possible reason why it took so long to become a known issue. Hence the reason I quoted the OP asking the question.
I still think it is an unacceptable amount of false advertisement and I would personally never use a 970.
nothing happened after using more than 3.5gb
>>53311545
>Framerate: 29 fps
>FPS: 63 fps
ok
>>53311579
EVGA Precission x misscounts FPS
>>53311722
Then how do we know it doesn't miscount vram usage?
>>53309082
>loses performance once you reach certain amount of VRAM
that's every gpu ever.
>>53311545
Careful it's about to explode
Buying a 970 is like buying a fast car but when you go into 6th gear the max speed drops to 30mph
