Quantcast
[ 3 / biz / cgl / ck / diy / fa / g / ic / jp / lit / sci / tg / vr / vt ] [ index / top / reports / report a bug ] [ 4plebs / archived.moe / rbt ]

Due to resource constraints, /g/ and /tg/ will no longer be archived or available. Other archivers continue to archive these boards.Become a Patron!

/g/ - Technology


View post   

[ Toggle deleted replies ]
File: 303 KB, 730x503, gtx970_vram_alloc_test_opengl-glslhacker.png [View same] [iqdb] [saucenao] [google] [report]
55872456 No.55872456 [Reply] [Original] [archived.moe] [rbt]

How come the 970 doesn't have any problems addressing all 4gb on board vram in opengl if it's such a massive issue?

>> No.55872841

>>55872456
The last .5 GB are slightly slower, so manchildren complain.

>> No.55872890

>>55872841
>manchildren
How cuck do you have to be when a company lies to you and still defend them?

>> No.55872941

>>55872890
Doesn't really matter if the lie is insignificant. I've been using 970 for 2 years now, never had any problems with it.

>> No.55872967

>>55872941
This desu, its only an issue with 4k iirc but the 970 isn't a 4k card anyways - if I was going to be using a 4k display I would've bought a 980 or whatever

>> No.55873447

>>55872890
How did they lie?
They stated 4gb, and it's capable of using the full advertised 4gb.

>> No.55873470

>>55872456
It does though. For example, when you play GTA5 on ultra at 1920x1080 it works fine, but when you play it at 1920x1200 it starts to run like ass because you now need that last half a gig of slow ass shit

>> No.55873634

>>55873470
Gra v is opengl?

>> No.55873641
File: 194 KB, 603x802, Shield.png [View same] [iqdb] [saucenao] [google] [report]
55873641

>>55873447
They lied about the ROP count (56 vs 64) and about the L2 cache (1.7mb vs 2mb)

Not to mention they failed to disclose that the memory wasn't available across the full bandwidth, so the bit rate was really 224GB/sec instead of 256GB/sec

http://www.pcper.com/reviews/Graphics-Cards/NVIDIA-Discloses-Full-Memory-Structure-and-Limitations-GTX-970

>> No.55873824

>>55872967
>>55872941
>It's fine they false advertised to me I wasn't going to use it anyway.

Like clockwork.

>> No.55873899

>>55873470
It doesn't, I play at 1440p and I have no problems with VRAM, my only problems are Anti Aliliasing since I already have problems getting 60FPS without it in some games.

I have to do trade offs, shadows some other settings if I want to increase AA.


What's even worse is that at 1440p 2xMSAA is still not enough.

Truly a meme resolution.

>> No.55874170

>buying a 4gb stick of ddr4
>0.5gb of it is only capable of sdram speeds
>"wtf i want my money back"
>"but its ddr4 and you do get the 4gb you wanted"
nvcucks are the worst

>> No.55874177

>>55874170
People seemed to be ok when the 660ti did it

>> No.55874196

>>55874170
s/sdram/sdr-sdram

>> No.55875062

>>55874170
>implying people will notice with normie use
I've even got Ubuntu mate on an old ddr2 laptop and I struggle to get it over 1GB ram usage with my regular use.

>> No.55875489
File: 343 KB, 600x489, 2724.hermit-crab-plastic-f1772.jpg [View same] [iqdb] [saucenao] [google] [report]
55875489

question : what percentage of memory (vram,ddr,whatever) are you OKAY with being below the advertised speed. I ask because those who defend the 970 are clearly okay with 12.5 % .

i'm just curious .
10 % ?
20% ?
30 % ?
How much will you take without caring?
i am curious

>> No.55875799

>>55875489
I think most people agree it's shitty and Nvidia fucked over a lot of people, but the performance impact is minimal to not noticeable

>> No.55875808

>>55875799
I should also add that people wet themselves with how well the card performed when it launched, and after the issues came to light literally nothing changed

>> No.55875842

>>55875799
>>55875808
i'm not disagreeing the card was very good regardless of the memory issues.
that's not the point i'm making .

>> No.55875868

>>55875842
Ok, to your original point, I don't think anyone is ok with it. I certainly am not with mine, and the fact that Nvidia settled a class action lawsuit shows that.

I'm just saying that the performance impact is minimal which makes the situation less dire than it would be otherwise

>> No.55875908

>>55875868
I agree

>> No.55875915

.5gb doesnt matter, AMDrones just do a serious amount of mental gymnastics so they think it is

>> No.55875984

>>55875915
They lack the ability to do any mental gymnastics

>> No.55876018

>>55873641
I did not consider those specifics when I bought the card. I considered that it performed very well for its price for others. After I bought it, it performed very well for its price for me. Maybe you were lied to - not me.

>> No.55876019

>I need 4GB of video RAM for shitty movie games and useless fuckhuge displays that you can't tell any goddamn difference in from normal ones except via placebo "I TELL MYSELF I CAN TOTALLY TELL A DIFFERENCE BECAUSE IF I DIDNT I WOULD CRY MYSELF TO SLEEP FOR PAYING MORE FOR FUCKING NOTHING"

>> No.55876026

>>55873641
AMD cards are also involved in the lawsuit going back to the 2xx series

>> No.55876029

>>55875842
>>55875489
The percentage that would enable you to use the card without noticing the limitation.

>> No.55876038

>>55872890

Their lie was on par with crying about McDonalds lying in their advertisements about how nice the burgers look.

It's completely irrelevant and you still got the same thing you expected in the end.

>> No.55876053

>>55876026
For what

>> No.55876060

>>55876038
This. Everyone loved it before the 3.5 thing. The hardware didn't change after we found out so I think it's just a bunch of drama queens

>> No.55876067

>>55876018
>Maybe you were lied to - not me
I guess if that's your position. Doesn't make up for the fact that what it says on the box is wrong

>> No.55876074

>>55876067
But it says 4gb on the box and it has 4gb on the card.

>> No.55876086

>>55876067
>Doesn't make up for the fact that what it says on the box is wrong
No, of course not. But since I'm not obsessed with meaningless details, that is irrelevant to me.

>> No.55876093

>>55876074
That's not what we were talking about

>> No.55876096

>>55876053
false advertising.
It seems like Newegg took down the listing of all the effected cards though.

>> No.55876107
File: 37 KB, 494x299, 1405651145173.jpg [View same] [iqdb] [saucenao] [google] [report]
55876107

>>55872841
>Slightly slower
Try 1/8th the speed.
Bandwidth is fucked if you try doing anything useful with your GPU it has to stay under 3.5

This is coming from a 970 user who bought for CUDA
I'm taking my 30 bucks and putting that toward a new GPU

>> No.55876109

>>55876086
>meaningless details
If you did some research, you'd realize how important the ROPs are. And the memory bandwidth

>> No.55876123

>>55876096
http://webcache.googleusercontent.com/search?q=cache:VUPgm6Z111cJ:blog.newegg.com/nvidia-gtx-970-class-action-lawsuit/+&cd=1&hl=en-US&ct=clnk&gl=en

archived link

>> No.55876143
File: 30 KB, 717x211, Screenshot_2016-07-28-12-51-39_org.floens.clover_1469756709098.jpg [View same] [iqdb] [saucenao] [google] [report]
55876143

>> No.55876144

>>55876109
I know they are not important because the video card works well.

>> No.55876146
File: 169 KB, 1920x1080, 20160330231437_1.jpg [View same] [iqdb] [saucenao] [google] [report]
55876146

>>55876107
If you bought a Maxwell card for workstation or compute tasks you're a fucking idiot

Also, I average between 3.6gb and 3.8gb on my 970 while running GTAV with no issues

>> No.55876185

>>55876146
What the fuck did you do to Michael

>> No.55876209
File: 241 KB, 950x1462, 18_0.jpg [View same] [iqdb] [saucenao] [google] [report]
55876209

>>55876185
Tried to model him after Robert De Niro in heat

>> No.55876529
File: 27 KB, 181x220, 1460512680972.png [View same] [iqdb] [saucenao] [google] [report]
55876529

>>55876107
You mean $26.25?

>> No.55876859

most of the problems people see when getting to the 4GB usage mark are probably due to switching to system RAM rather than the slower 512MB section. I believe it is difficult to measure this but there is a significant performance drop on any card when this happens.
It has been revealed that the card is quite intelligent about when and how it uses the last 512 so that the performance impact is minimal.

>> No.55877085
File: 201 KB, 1827x1258, 1469997916477.jpg [View same] [iqdb] [saucenao] [google] [report]
55877085

Daily Reminder that the 1070 GTX did the same thing except it can only address 6.7 GB of the 8GB it advertises.

>> No.55877098

>>55877085
Except it didn't and has been disproved on this very board.

>> No.55877112
File: 84 KB, 533x700, fag.jpg [View same] [iqdb] [saucenao] [google] [report]
55877112

>>55877085

>> No.55877119
File: 315 KB, 1024x628, gtx970memory.jpg [View same] [iqdb] [saucenao] [google] [report]
55877119

>>55872841
>slightly

>> No.55877269

>>55877098
>disproved
Nope
>>55877112
Gonna keep posting and there is nothing that you shills can do about it.

>> No.55877307

>>55877269
Well since you posted it, where's the proof?

>> No.55877316

>>55877307
Where's the disproof?

>> No.55877327

>>55877316
I wasn't the one that said it was disproven. But you posted a supposed fact now back it up

Burden of proof faggot, lies with you not me

>> No.55877334
File: 422 KB, 1920x1080, 1469998061888.jpg [View same] [iqdb] [saucenao] [google] [report]
55877334

>>55877307
>>55877327

>> No.55877345
File: 416 KB, 1920x1080, 1469998127443.jpg [View same] [iqdb] [saucenao] [google] [report]
55877345

>>55877307
>>55877327
There you go faggot

>> No.55877377
File: 1.24 MB, 245x180, lolpoop.gif [View same] [iqdb] [saucenao] [google] [report]
55877377

>>55877334
>>55877345
https://www.reddit.com/r/nvidia/comments/4mvoj0/gtx_1070_ram_test_ram_used_only_67gb/


https://www.google.ca/search?q=6.7+gb+vram+nvidia+1070&ie=utf-8&oe=utf-8&gws_rd=cr&ei=lGeeV67XIKHLjgTmgZ_ICQ

http://forums.guru3d.com/showthread.php?t=407953


TL;DR:

total bullshit
has been disproven
op is a faggot.

moving on

>> No.55877404

>>55877377
Those links present no evidence of it not being true. Just a redditor mod of the Nvidia subreddit saying it's false and linking to a pajeet forum.
And a thread on guru 3d where nobody can explain why that happens.

>> No.55877519

>>55877377
>>55877404
Nvidia shills btfo

>> No.55877521

>>55877404
That test is flawed anyway, it says how much of the vram it has allocated for the test at the top, and also to ensure nothing else is in vram before the test starts, which is impossible inside of Windows

A 980 only tests 3200MB with the same bench, and there is no question it has the full 4096, so unless you have something else besides a flawed test, I call bullshit.

>Texture unit count drops from 160 on the GTX 1080 to 120 on the GTX 1070 thanks to a loss of five SMs.
>It looks like NVIDIA has disabled one complete GPC rather than disabling SMs piecemeal across the GPU. ROP count remains the same though at 64, and of course the memory bus stays at 256-bit to go along with them.

http://www.pcper.com/reviews/Graphics-Cards/GeForce-GTX-1070-8GB-Founders-Edition-Review

This was the problem that caused the 970 memory issue, which they didn't do on the 1070.

>> No.55877532 [DELETED] 

>>55877519
A single flawed "test" isn't proof

see
>>55877519

>> No.55877543

>>55877519
A single flawed "test" isn't proof

see
>>55877521

>> No.55877569

>>55877521
Except it's getting more and more clear that they did the same on the 1070. They lied about it for months with the 970 and had shills like you lie about as well. But now we finally have a class action suit that proves it.
How much will be this time $50?

>> No.55877590

>>55877569

this is getting a little sad now

go to bed raja

>> No.55877592

>>55877404
Except that makes no sense.

Think about it, when you run a 970 on the benchmark it shows all 4096MB, but upon passing the 3587MB mark immediately starts to drop down in performance. If you put a 980ti in there, it won't detect all of the ram either; or so some people on the site say. I don't know. I don't have a 980ti.

If anyone has a GTX 1080, I suggest running the benchmark and seeing if it will detect the full 8GB of ram or if it will find 6.7GB as well.

If you like I can try and load up a game and get a screenshot of my 1070 using over 6.7GB of VRAM.

>> No.55877595

>>55877543
>Tests that prove me wrong are flawed.
Kek

>> No.55877606

>>55877569
>Except it's getting more and more clear that they did the same on the 1070
I'm just waiting for hard evidence instead of grasping at straws with a flawed test.

As I said before the way they cut down the chip is different than with the 970, which was what caused the problem

I've made my counter claim, I really don't care if you don't agree with it

>> No.55877611

>>55877606
See
>>55877595
You'll claim anything that proves to wrong is flawed anyway. I already proved what I had to prove

>> No.55877619

>>55877595
I've explained why it's flawed, it's not meant to test memory capacity, only bandwidth, as well as explained how Nvidia changed the SM disabling

>>55877611
You didn't prove shit, and I showed how your methodology is wrong

>> No.55877624

>>55877592
The game will just use ram instead of vram after the 6.7 mark

>> No.55877634

>>55877619
It's not flawed, you can select how much memory you want the bench to use.

>> No.55877651

>>55877624
Right, but the tool I use (Precision XOC) measures VRAM usage, specifically. If it's thrashing, it will show only the amount used on the card (2048MB in the case of my old 670.)

Gimme a sec.

>> No.55877714

>>55876019
Amazing how ncucks immediately turn on their damage control when someone criticises nvidia for doing something objectively wrong. smdh

>> No.55877744

>>55876143
This

>> No.55877791

If you compare >>55877345 to >>55877119 you will see there is a massive difference in how the card acts approaching the end of the test.
On the 970 (which had 3840MB assigned under proper testing conditions of no dwm, no background applications allocation VRAM) as it approached the final 512MB it slowed down. The 970 assigns that memory last as it is slower.

On the 1070 it was not run under ideal conditions, dwm is running and there are background applications, Chrome and maybe others in the sys tray, however there is no slow down as it reaches the end of the test.

This means that it is not, at least, the same configuration as the 970 which has 3.5+5, or in this case 6.7+1.3, which it couldn't have anyway because the 1070 has eight 1GB RAM chips (these have been seen with the cover off, there is no hiding what they are), at best it would be 7+1.

If someone with a 1070 would like to nail this shut I would suggest getting Nai's benchmark, running a game that will push your VRAM usage over 2GB and then run Nai's benchmark with GPU-z running to show usage, be sure to screenshot Nai's benchmark results (the DRAM bandwidth part, not L2 cache and GPU-z.
You will see a hit to the VRAM bandwidth due to the load of the game on the GPU but you will get full VRAM utilization and we would also see if there is any slow down on the final part of the VRAM (if it is like the 970 it will be assigned last so Nai's benchmark will be utilizing it, not the game).

>> No.55877820
File: 30 KB, 387x488, Untitled.png [View same] [iqdb] [saucenao] [google] [report]
55877820

http://i.imgur.com/jgHyWmA.jpg
http://i.imgur.com/uumYQHq.jpg
http://i.imgur.com/4vm2as6.jpg

Here's three screenshots of my GTX 1070 going above 6.5GB.
Tested in Battlefield 4 on Ultra settings at 3840x2160 with 4x DSR (7680 x 4320).

The graphics card has a 175MHz core overclock, memory clock is stock.

I attached a screenshot of my GPU-z info. Like I mentioned, PXOC doesn't show system memory usage, it only shows VRAM usage. If this isn't enough I can post a screenshot of OpenHardwareMonitor showing over 6.7GB of VRAM usage.

>> No.55877829
File: 1.33 MB, 2560x1080, all 6gb nigga 1.png [View same] [iqdb] [saucenao] [google] [report]
55877829

>>55877592
I'd like to add when I ran Nai's bench on my 1060 it only allocated 5120MB to the benchmark, however when I used two games to push my VRAM usage higher the benchmark would then use all the remaining VRAM.
There was a performance hit to the VRAM bandwidth, but that is understandable from the games hitting the GPU.

>> No.55877846

>>55877085
Daily reminder AMD paid shills are trying to push FUD in /g/.

>> No.55877859

>>55877791
Will do, on it!

>> No.55877955

>>55877820
Thank you anon

>> No.55878027
File: 192 KB, 1614x1151, Untitled.png [View same] [iqdb] [saucenao] [google] [report]
55878027

>>55877955
No worries.

>>55877791
So, I did what you said. Gave it a run in Nai's and gave it a run in BF4.

Look below -- what we see is when we're using about 700MB on idle we're seeing about 7579MB of usage in Nai's Benchmark. You can also see that, in Nai's, the VRAM usage is FLAT. Because throughout the entire benchmark, it allocates a chunk of RAM, not releasing it until the program is closed. You can actually see that in the idle screenshot, that it looks like a solid block. If we take 7579 MB (the highest it reached in Nai's) and subtract 719 MB (the average I was getting on idle) we get about 6.7 to 6.8 GB. So, pretty much exactly what Nai's is saying is available, give or take a few megabytes. Also notice that it's not even close to capping the VRAM.

In BF4, however, it's using practically the entire VRAM. There's only 20 MB free. It was higher at some places and lower at others. We're talking a few dozen megabytes here at most.

Screenshot this if you like -- if anyone comes about spreading this article again, feel free to call them out because it's complete BULLSHIT.

>> No.55878209
File: 111 KB, 592x459, 1425433905905.png [View same] [iqdb] [saucenao] [google] [report]
55878209

>>55877119
>why does the more powerful gpu have faster vram

>> No.55878411

>>55878027
The main thing I think would be trying to disprove the behavior of the 970.
See >>55877119
When the 970 gets to the ~3200 mark it slows down drastically (the benchmark allocated 3840, if we assume it allocated all the available VRAM that would mean 256 is allocated else where putting that 3200-3328 block in the 3456-3584 which is roughly the entry point into the final 512MB area).

So the goal would be to get Nai's benchmark to allocate the rest of the VRAM so that it can be tested for the same behavior. Like I said in >>55877791 and >>55877829, if the 1070 has a slower pool of VRAM like the 970 does it would be allocated last, so the best way to test this would be to run a game that would use a portion of the VRAM (around 2GB, not all of it) and then run Nai's benchmark, which will then most likely allocate all the remainder of the VRAM.

I do find the consistent allocation of 6784MB on the 1070 interesting though, it makes me wonder if in the driver there is a limit set on these Pascal GPUs to the amount of VRAM CUDA can allocate, preventing full allocation.

>> No.55878717

>nvidia admits they lied and is giving out $30 for compensation
>they didn't lie
I know you're desperate to suck Huang's cock, but blindly shilling isn't the way to do it.

>> No.55878874

>>55878717
Actually, when I read the full disclosure I am more on the side of NVidia about not lying about the memory (aside from the mistake about how many RDPs and the bus width on the initial information release).

The memory is still there and running at full speed but because of the way they disabled an SM on the 980 GPU to create the 970 the way it communicated with the final 0.5GB had to change.
Technically they could have disabled that last 0.5GB completely and sold the card as 3.5GB but the way they did it the card can at least still address that last 0.5GB at a rate much faster than using system RAM, which is what graphics cards do when they run out of VRAM anyway.

>> No.55878935
File: 529 KB, 1340x1588, 1461090171933.jpg [View same] [iqdb] [saucenao] [google] [report]
55878935

Reminder that AMD are the real jews here.

>sells you a re-brand of a perfectly capable card for $130 more with an extra unusable 4gb vram slapped on.

>> No.55879122

>>55878935
Don't forget giving the 390 driver specific features that they excluded from the 290

>> No.55879160
File: 29 KB, 480x480, 1465915660854.png [View same] [iqdb] [saucenao] [google] [report]
55879160

>>55877269
Let me go get your ban for shit posting

>> No.55879172

>>55878935
AMD SHILLS DEFEND THIS BULL SHIT
>AMD ON SUICIDE WATCH

OH THE MERCY

CAN'T
MAKE
THIS
SHIT
UP

>> No.55879180
File: 29 KB, 500x509, 1464351997346.jpg [View same] [iqdb] [saucenao] [google] [report]
55879180

>>55879172

>> No.55879190
File: 54 KB, 540x496, 1470112836938.jpg [View same] [iqdb] [saucenao] [google] [report]
55879190

>>55879180
>>55879172
>>55878935
Pajeet is on suicide watch just like amd

Proof right here /g/ you saw it here first

>> No.55879205

>>55872841
Like it when you can /thread on 1st response.
(There is a reason it still won benchmarks and recommendations for months after launch, then someone realized box specs were off and AMD thought they could shill it up and make benchmarks go away and have a chance at selling a card again)

>> No.55879238
File: 110 KB, 983x1192, Nvidia legacy products.png [View same] [iqdb] [saucenao] [google] [report]
55879238

Keep on truckin' /g/.

>> No.55879255

>>55879238
Isn't anything that isn't produced anymore a legacy product?

>> No.55879256

>>55872841
The whole point to VRAM is that it's fast. If it's slow it's unusable.

The analogy would be like selling you a car that when you put 4 people in it goes 2 MPH but with 3 people it goes 60 MPH. You could just get out and walk faster.

So they basically lied about the ram, and then "fixed" it in a firmware update by disabling the slow chip they tacked on for false advertising. enjoy your $30 payouts if you bought the card coming soon.

There's a class action in progress and the 960 up to the 980 TI could be eligible, and even some AMD R series I believe. Keep an eye out for news about the lawsuit.

>> No.55879281
File: 85 KB, 539x991, link of lies.png [View same] [iqdb] [saucenao] [google] [report]
55879281

>>55879256

>disabling the slow chip

They did no such thing. What Nvidia does do is hand tune each game ready driver for the 970's unique memory layout (same thing they did for the 660ti). The second actual optimisations for the 970 stop anytime a game slams the vram with no fucks to give the card will choke.

>> No.55879284

>>55879238
>not even being manufactured any more

what does previous-generation mean to you?

>> No.55879285
File: 188 KB, 720x1280, Screenshot_2016-08-02-19-05-20_com.android.chrome.png [View same] [iqdb] [saucenao] [google] [report]
55879285

>>55879238
Not in production = Legacy
Nvidia is better than AMD, they don't drop old hardware, older cards are still getting game ready drivers that improve performance.

Wish I could say the same for AMD, dropping legacy cards like the HD6970.
Fun fact, the HD6970 was as fast as a 660 ti, or a 7870.

Now look how far behind it's fallen thanks to being dropped by AMD.

>>55879256
That last 0.5GB is still much faster than regular ddr3 ram buddy. It's not like getting out to walk would be faster in this scenario.

>> No.55879311
File: 28 KB, 500x610, CoD BO 3 1080p.png [View same] [iqdb] [saucenao] [google] [report]
55879311

>>55879285

>kepler

>> No.55879316

>>55879256
>and then "fixed" it in a firmware update by disabling the slow chip they tacked on for false advertising
That's bullshit.
1) All the chips are the same speed, it is jut the way the GPU connects to the final chip is different due to one SM being disabled.
2) It is still faster to use the final 512MB than than using system RAM, which is what happens when the VRAM runs out.
3) If they had disabled the final 512MB of VRAM it would have HARMED performance even FURTHER.

So basically stop your misinformation bullshit. You're certainly no better than NVidia, except you won't be required to pay out for making up shit.

>> No.55879342
File: 70 KB, 317x266, 1446224602529.png [View same] [iqdb] [saucenao] [google] [report]
55879342

>>55879311
Im not here to argue about GPUs but COD and that engine is pretty much irrelevant.

>> No.55879359

>>55879311
>bo3
>using the most broken game as any form of proof

an i3 gets superior performance to an i5 or even i7 in this game.

>> No.55879364
File: 100 KB, 705x896, Screenshot_2016-08-02-22-26-41_com.android.chrome_1470140817958.jpg [View same] [iqdb] [saucenao] [google] [report]
55879364

>>55879311
I too, can cherry pick benchmarks.

Notice how the 780 ti is actually getting better min fps than 970?
Yeah, that's not gimping, you're just being a cunt.

>> No.55879375

>>>/v/

>> No.55879377

>>55879364

>390x on par with a titan x

Something is very off here.

>> No.55879397

>>55879377
1. bo3 is broken

2.computerbase are HEAVILY biased towards amd.

>> No.55879406
File: 30 KB, 650x453, 7YFXC.jpg [View same] [iqdb] [saucenao] [google] [report]
55879406

>>55879375
>People whining about only manchildren play games
>le go to /v/
>Getting upset because someone likes something you don't is ulitmate /v/ logic

You are the one that belongs on /v/

>> No.55879410
File: 73 KB, 624x469, 59c6ef77_computerbase.jpg [View same] [iqdb] [saucenao] [google] [report]
55879410

>>55879311
Nothing wrong with kepler

>> No.55879412
File: 268 KB, 1924x1083, 1468921006593.png [View same] [iqdb] [saucenao] [google] [report]
55879412

>>55879397

>2.computerbase are HEAVILY biased towards amd.

Upon what do you form this opinion?

>> No.55879435

>>55879406
when you take playing video games as an insult, you know there is something wrong with playing video games.

>> No.55879447
File: 300 KB, 512x384, 1394006862631.png [View same] [iqdb] [saucenao] [google] [report]
55879447

>>55879435
>when you take playing video games as an insult, you know there is something wrong with you.

FTFY

Fucking neo /g/

>> No.55879453
File: 41 KB, 500x1130, perfrel_1920_1080.png [View same] [iqdb] [saucenao] [google] [report]
55879453

>>55879410

A 770 should be beating a 280x.

>> No.55879458

>>55879316
>Misinformation bullshit
>It's not a slow chip, it's just a slow chip BECAUSE
>They didn't disable the RAM, they just patched the driver for all games to not use the ram.

mmmkay. That's an oddly specific way to say that i'm wrong because I didn't write a 2 page report on why the 970 was falsely advertising specs and instead summarized it for non ]-autistic-americans.

>> No.55879480

>>55879412
just briefly, they always get massively different results to every other benchmark sites out there eg they tested the 1060 in witcher 3 and found it to be like 10 fps slower than the 480 (both stock) and the gtx 970 was at like 26 fps. that means they're skewing their results to favor amd or they're running the game with hairworks 8xmsaa but turning tessellation down purposely in the amd control center to make the amd cards look better.

it not just with their 1060 vs 480 tests either. they've had weird results in their 390 vs 970 tests.

i'll try and find some and post it.

>> No.55879496

>>55879453
Yeah, in dx11.

You forget, the 770 wasn't designed for dx12 and as such it actually loses performance. AMD wins with dx12, but there's still nothing wrong with the 770.
I've got a 670 and it's still performing great for its age. 7950 is performing better in dx12 games, but I'm about to upgrade soon so idgaf, I don't buy my tech to enjoy them in 4 years, I buy what I can get the most out of at the time of purchase.

>> No.55879518

>>55879496

>Yeah, in dx11.

Except it doesn't in DX11 either, as per TPU's performance summary charts.

>> No.55879560
File: 38 KB, 500x1050, perfrel_1920_1080.png [View same] [iqdb] [saucenao] [google] [report]
55879560

>>55879518
It's exaggerated in his chart, here's one with less difference

>> No.55879578

>>55879560
>>55879518
Also, I forgot.

>770
>2gb
Vs
>280x
>3GB

Probably aged better because it has 50% more vram.

>> No.55879602

>>55879578

Does remind when a certain group was quite vocal in claiming that 2gb is all you ever need for 1080p.

>> No.55879608

>>55879458
>>It's not a slow chip, it's just a slow chip BECAUSE
The chip is not any different from any of the other chips on the board, so it is not a slow chip. I am sorry but that is just a fact.
>>They didn't disable the RAM, they just patched the driver for all games to not use the ram.
The final 512MB will not be used until the rest of the VRAM has been used. It is not disabled, it is just reserved until the rest of the VRAM runs out.

So, yes, you are spreading misinformation bullshit.

>> No.55879614

>>55876529
niceme.me
Nie kys

>> No.55879635

>>55879578
hardware unboxed did a video on this and concluded that it isn't nvidia cards being gimped but it's just that the older hardware of the nvidia cards are causing bottlnecks eg less vram, smaller memory bus etc. amd cards generally had the beefier hardware but that in turn resulted them to have massive amounts of power consumption compared to the nvidia competition and also caused them to have housefire tier heat output.

>> No.55879678

>>55879635
Babeltech had a similar conclusion

>>55879602
I remember that too, and if I'd gotten a hd7950 I probably would have waited longer before upgrading, but my 670 had a good life and there's no point being salty over it. It's been a good 3 years, I'll probably upgrade to a 490 early next year, or a 1070 if the 490 is underwhelming

>> No.55879740

>>55879608
>It's not a slow chip. It's just a slow BECAUSE the transport layer!
>I'm not a fat person, I'm just larger BECAUSE of my increased food intake and lack of exercise.
The chips still slow and you're still fat. That's a fact of life.

>> No.55879844

>>55879740
>>It's not a slow chip. It's just a slow BECAUSE the transport layer!
Exactly. The chip isn't slow.
They didn't cheap out and put a lower speed chip in that position, access to that chip is slower due to the disabled SM in the GPU and they worked around that to keep performance impact to a minimum.

And unfortunately for you I am only about 63kg. Get rekt.

>> No.55879877

>>55876107
>This is coming from a 970 user who bought for CUDA

970 only has 1700 CUDA at best.

You fucked up on so many levels.

>> No.55879942

>Works perfectly fine in 1080p
>Never was good enough to play 1440p/4K
>3.5 vram memes when 99/100 games you try to play in 4K will shit themselves to sub 30 fps before you get anywher near 4 gigs of ram.

Also no, GTA5 doesn't shit itself cuz of vram on higher resolutions. The card just isn't good enough. Set your gta 5 to something like 1440p/4x msaa - now set your texture details to low/medium to drop below 3.5gb vram - No difference in performance.

>> No.55879967
File: 114 KB, 1754x618, 970.png [View same] [iqdb] [saucenao] [google] [report]
55879967

Dunno why but vram use didn't go over 2300 mb with 30 textures
I tried 60 textures and it didn't go past 3500mb

>> No.55880010
File: 103 KB, 1755x611, 970-2.png [View same] [iqdb] [saucenao] [google] [report]
55880010

>>55879967

>> No.55880041

>>55880010
>no bus interface load
Well it isn't using system RAM then, so I guess you need to load it more.

>> No.55880060

>>55880041
The tool is just broken

>> No.55880062

Are there any tools for addressing all 8gb of VRAM?

I haven't been able to find anything to test my 1080 on that front yet.

>> No.55880078

>>55879942
I play gta v on decent framerates (50-60fps) on 1440p on a 970 with MSAA off and soft shadows and grass quality set to high

>> No.55880195

>>55880062
I suggest this method >>55878411 >>55877829
You can get pretty much full usage of the VRAM and check for any massive slowdown when it gets to the top area (if any VRAM is handicapped like the 970 it will be addressed last and by Nai's bench).

>> No.55880199

>>55880062
Try Rise of the Tomb Raider on 4K

>> No.55880261

>>55880062
mirrors edge hyper settings at 1440p or 4k. it was specifically designed to tax the 8gb in the 1070/80.

>> No.55880323
File: 184 KB, 1282x724, 1467654831276.jpg [View same] [iqdb] [saucenao] [google] [report]
55880323

>>55880261

Hyper settings does horrible, horrible things to 4gb cards.

>> No.55880360

>>55880323
they patched that. it runs fine on 4gb cards now.

https://youtu.be/JjTUtUF6gxg?t=4m50s

>>
Name (leave empty)
Comment (leave empty)
Name
E-mail
Subject
Comment
Password [?]Password used for file deletion.
Captcha
Action