670 SLI or 980?
Buying a second hand 670 is 100€, buying a new 980 is more than 4 times as much.
Considering I will not use more than 1080 resolution the 2 vs 4 Gb ram should be of no importance.
There are people who say that SLI rigs although having same fps look less smooth. Wtf is that all about?
http://www.geforce.com/whats-new/articles/nvidia-livestream-may-6th-at-6pm-pst
GTX 1080 or GTX 1070
>>54405811
Sure, I could just spend even more money. Go away nvidia shill
>>54405800
>calling people nvidia shill
>Only asking about nvidia cards
Get a 380 if you want the best bang for bucks or 390 if you want to max out and profit of dx12.
>>54405800
Gtx 670 sli user here. Gigabyte windforce.
One card overclocks like shit but I'm still getting decent scaling regardless.
Check your mobo, I'm limited to 16x/4x pcie 2.0. If you're on pcie 3.0 though you'll get better scaling
>>54406194
I should add, i only play battlefield 4.
I'm getting 100+ fps all the time, great for my crt.
>>54406194
Thanks for the tip. Mind putting it on all the way? My mobo is a P8Z68-V Pro.
>>54406235
Mines a gigabyte p55a ud3r usb3.
Ask any questions, I'll be here for a few minutes before bedtime. Gotta be up early for work.
>>54405800
>Considering I will not use more than 1080 resolution the 2 vs 4 Gb ram should be of no importance
Nice meme. Pity it's not 2010 any more. Literally any big release in the past couple of years will break the 2GB barrier at 1080p unless you turn textures way down.
Only a retard would consider doubling down on their gimped, outdated Kepler setup at this point, but it's your money. Enjoy that ever-dwindling SLI support fading away to nothing once DX12 becomes the norm and developers have to implement mGPU themselves.
>>54406256
Do you have temp or noise issues? Summer is coming so I don't want to put my house on fire
>>54406272
Multiadapter!
Yiss
>>54406357
Second gpu ran hotter, so I modded the case and put a small 80mm fan on top of the cards blowing cool air in for the second card. It went from a 10+ degree difference to a 2 degree difference
>>54405800
>not waiting for Volta
fucking pleb.
>>54405800
Not worth it. You only SLI when you already have the biggest card avaiable and want more.
>>54405811
This, maybe a 1060Ti
>>54405936
Retarded, also fuck DX12. 380 wouldn't be any upgrade and he could get a 290 much cheaper than a shitty rebrand. 290X is currently the best price-performance ratio on the market, but I would wait for the new cards.
>>54406272
I would even get a 780Ti instead of 2 670s. But not anymore, the Tis still go for ~300 €, 290X are only 200 if youre lucky +1 GB VRAM. Get a 1060Ti for equal performance with 6 GB VRAM, you can never have too much.
All those shitty 3.5 and 4 GB cards will die off once the new ones have arrived at the market, Entry cards already get 6 GB, Performance ones 8, Enthusiast / highend 16-32 GB.
>>54406694
>Entry cards already get 6 GB
pls be true
>>54406751
I mean this at least for the 1060Ti, the 1050/60 will possibly only feature 4 GB. Maybe they release 8 GB cards too.
My 2011 GT 530 has more VRAM than a 580, but its always only DDR3.
single gpu > dual gpu
unless you already have the best single gpu and then you can double up.
>>54406694
>VRAM meme
>still on a 780ti gaming is still great on high settings at 1440p
>it only has 3GB of ram
>>54406787
VRAM is not a meme. I'm on a 780Ti and play in 4K. Whatever I play it is constantly full. Sometimes I need to lower texture quality. It is OK for the current state but won't last very long. As soon as the bigger sizes come into the market developers start to use higher texture resolutions.
>>54406829
>780ti
>4K
You play games at <20fps?
>>54406779
This
>>54405800
As another owner of a 670 that got completely neglected by Nvidia after Maxwell, I just can't make myself throw good money after bad.
A 670 is actually a distinctly stronger GPU than a 960, but you'd never be able to tell by looking at newer games' benchmarks.
>>54406779
Yes. SLI/Crossfire is only good for benchmark scores. Actual performance sucks because of microstuttering. It will be unplayable even though you will be rocking 60+ FPS all the time because of choppy gameplay.
** DO NOT BOTHER WITH MULTI GPU CONFIGURATIONS. BUY THE BEST SINGLE GPU CARD YOU CAN AFFORD **
>Yeah I am serious about that statement.
>>54406859
No, 60 stable. BF4 works on ultra without MSAA. I also tried shitty games like Witcher 3, only runs with 40, so I won't even try.
I downgraded from a Titan X, got the 780Ti to avoid 3.5 shit, maybe I get a 1070 when its not gimped and below 350 €. I'll buy a 1080Ti whenever it is avaiable and Mass Effect Andromeda has an official release date.
This should also explain why I spent 1000 € on a shitty Titan X.
>>54406904
>It will be unplayable even though you will be rocking 60+ FPS all the time because of choppy gameplay.
That's a problem that's solved with a FPS cap in RTSS. If you system can run 60FPS+ then just lock it to 60.
>>54405800
>670 SLI or 980?
>100 Euro on a piece of shit that's been deprecated for 3 years
That's a great waste of money! Good job.
Since you want to buy 2 cards * 100 Euro each - and Nvidia cards sli like shit btw - I can only tell you to wait 3 months and buy a mid range Polaris junker.
At least you'll get a card capable of VR.
The 670 are too old to be viable.
>buying old cards when new die shrink is around the corner
>>54406931
Gsync
>>54406931
You can't solve micro-stutter in any way.
It's a low level driver problem and limiting fps will just make the choppiness more pronounced.
future wise you better buy a 390 (if you cant wait)
games are becoming more and more compute heavy (qb) and naturally nvidia will fall behind
given how consoles is now locked 100% on amd also given how adreno chips on mobile sockets(amd desing too) are going to be the norm
and given that amd is pushing for dual gpu on consoles (polaris 10 one apu and one discrete gpu) getting an nvidia now will most likely result on your biggest mistake because the white paper of p100 showed to us that they have become more like gcn 1.1 but they are still doing static scheduling there fore again they wont have any serious async or even compute+copy capability
>>54407060
>given how adreno chips on mobile sockets(amd desing too)
Anon, AMD sold off the Imageon line to Qualcomm in 2009, which was around the end of the fixed-function pipeline era or maybe the very beginning of the TeraScale era.
They have very little in common with modern GCN designs.
>>54405800
Pascal is literally months away bruv...
>>54406210
Why would you buy a 980 for a CRT?
Isn't the highest resolution those things get like 800x600?
Might as well get the 670
>>54407167
actually the whole front end of adreno is a copy paste of gcn they do have patent agreement with amd there is a reason as to why they said they almost didnt had to do anything on s7 to create a game on vulkan
>>54407207
I don't know where you got that from, my last CRT from 2007 could run 1600*1200@85HZ
Lg flatron f900p master race
>>54407207
my flat 22" Iiyama trinitron does 1600x1200@100Hz, m8.
> too bad it's been in storage for like 8 years since I bought my HP LP3065 30" IPS because of muh productivity.
3 9 0
>>54405800
670 SLi here.
It works well, but I'm probably going to upgrade when AMD and nvidia launch their next lineup.
What games are you going to play?
The Witcher 3 ran pretty well for me, just had to turn a few things down. I usually get 90-98% usage out of both of them, but I had to OC my 2500k. Before that I was only getting 60% on both of them.
It says SLI disabled because I just updated my driver.
>>54407006
Not true.
Gta v is choppy with hd5770 crossfire, but the choppiness completely disappears when i enable vsync
>>54407218
I hope that AMD is at least getting some respectable royalties if they're pushing their better tech off to third parties.
>>54406938
I already have a 670.
>>54407207
I think he meant he has 2 670s for his crt anon.
>>54407328
Also, never goes below 100fps in overwatch with literally everything maxed out.
>>54407328
Mainly Ark survival evolved and DayZ (yes, I'm an alpha / beta / pre release addict) my cpu is an 2600k at 4.5GHz
>>54407182
You mean less than a month, plus reveal (and maybe tech site reviews, not just rumors) starting in just 8 hours.
>>54407387
Ark runs shit on anything. I tried it once and refunded after 30 minutes. I haven't played ARMAII in a long time so I'm not sure about that.
It seems like you don't play too many demanding games, so just wait until the new crads are out.
>>54407340
they never got for the mobile
look like they destroyed g sync with adaptive sync..
currently the number of monitors with it is almost 5 times larger than g sync they only advertise it on their cards..
or hbm..they spend years perfecting it and they never asked for anything but you know
semiaccurate already are reporting problems with nvidia version of hbm on hpc this is going to be good too good
also remember amd practicly forcing the next consoles to be crossfire native it means they are getting a very big revenge regarding how gameworks keeps sabotaging amd
>>54407429
Ark runs wayyyyyyyyyy better now
>>54407447
>semiaccurate
Charlie ripping on nvidia is kind of a broken clock situation.
GP100 *not* having issues would surprise me, but given that it's all but vaporware until Q1'17, yield and other quality issues don't have much bearing yet.
I'm just shaking my head that anyone in the consumer sector even cares about that chip whatsoever, given that it's basically Fermi's successor and not at all designed around gaming.
> tfw 10.6 fp32 TFLOPS with 3.5k shaders
> tfw Fiji is 8.6
> tfw 6k-8k fp32 shader consumer cards probably not coming till 2018
>>54407914
you should see some idiots on anand forums thinking a gp102 will come directly from this card...
also it was bound to happen nvidia rushed with hbm amd said they had similiar problems 2 years ago while testing and its natural for them to release something like this on that segment since "if its broken fix it your self" kind of attitude nvidia has its just suits them good
>>54407914
The Pascal architecture has a maximum of 3840 shaders, I think they'll release a GP102 card without DP units for the consumer market. Both with HBM2, but the Ti version will have a slightly reduced amount of shaders, maybe all for the Titan and 3584 for the Ti. Nontheless I go for the Ti, they get cheaper and I could have gotten 2 980Tis for the release price of my Titan X. Fuck the VRAM, 16 GB HBM2 is enough for 4K.
>>54407963
my gut feel is
> Q1'17:
> GP100 - heavily GPGPU focused, still sold to retarded consumers who didn't learn from Fermi
> Vega - largely GPGPU focused like GP100 so they don't loose entire enterprise segment
> 2018:
> GP200 - the Maxwell (stripped down) version of GP100 with more shaders and little fp64 capacity
> Navi - similar, with possible addition of "NextGen" memory, though this part might well get dropped
I strongly suspect that Polaris 10/11 and GP104 will have 2+ year lifetimes in the market and that AMD/Nvidia will just make the higher end 14/16nm chips less shit over the next few years.
>>54408560
navi if everything is correct will be the attempt of amd to bring dual gpu card to the masses
given that they practicly will have cf on consoles its logical..
vega will most likely be the last big (die) gpu amd will ever do
>>54408507
GP100 is clearly a 4096 shader GPU with HEAVY fusing/binning going on, being a huge chip on a new lithography node.
The point about the architecture is that like Hawaii, every ALU can either do 1 fp64 mult+add per cycle or 2 for fp32, at the cost of more area used and power needed than pure fp32 ALUs.
It's not like there are extra shaders for fp64, just the ability to use the same slightly expanded logic in multiple distinct ways.
Any GP102 will be as different from GP100 as Maxwell is from Kepler.
>>54408560
But what is "next-gen" memory? HBM2? The coming enthusiast cards will feature up to 32 GB HBM2, 32 confirmed for the GP100, with a bandwith around 1 TB/s (Tesla GP100 = 720 GB/s).
I don't think they will release a GP200, 2018 is already Volta time. Don't forget the GP102 is already in the driver, so we can sure this GPU will be somehow used. Same for the GP104-150, which will be a stripped down 1070, closing the gap between the X60 and X70 lineup. GP106 comes later this year.
>>54408652
I know that, and it makes zero reason to release such a GPU on the gaming market. Maybe on some overpriced card like the Titan. I'm very sure the GP102 is a gaming optimized cut version of the GP100.
>>54408765
search the recent amd patents and you will see what it is..
(basicly they are trying to put hbm modules inside the gpu as a cache and remove any memory controller they have)
>>54408765
>But what is "next-gen" memory? HBM2? The coming enthusiast cards will feature up to 32 GB HBM2, 32 confirmed for the GP100, with a bandwith around 1 TB/s (Tesla GP100 = 720 GB/s).
It's almost certainly going to be directly stacked memory, which eliminates the interposer and provides the potential for even wider/faster buses, but will be a bitch to assemble and to properly cool.
There's also the possibility of rearranging the DRAM subarrays to reduce power draw at the cost of some density, since row/column selection power drain is starting to strongly outstrip IO power.
>I don't think they will release a GP200, 2018 is already Volta time. Don't forget the GP102 is already in the driver, so we can sure this GPU will be somehow used. Same for the GP104-150, which will be a stripped down 1070, closing the gap between the X60 and X70 lineup. GP106 comes later this year.
Driver name entries don't mean much, especially when products are this far from release and the vendors want to spread as much disinformation as possible.
I just find it difficult to accept that Nvidia would pay to produce another ~600mm^2 chip this early in the 16nm process's lifetime, since it's unlikely that they would be able to produce something the next year that's substantially better in the consumer space.
>I know that, and it makes zero reason to release such a GPU on the gaming market. Maybe on some overpriced card like the Titan. I'm very sure the GP102 is a gaming optimized cut version of the GP100.
GP100 is GPGPU from the ground up, with few (3.5k binned 4k) but complex shader ALUs and a massive 14MB (binned from 16MB) register file. The true successor to GM200 will likely be essentially pure fp32 (6k+ shaders probably) and have a smaller and less power hungry scheduler and register file. The only things it will have in common are the HBM, the 16nm lithography, and the rough size. Calling it a "cut version" of GP100 is simply nonsensical.
here the patent of amd
http://patents.justia.com/patent/9331053
>>54409085
> stacked thermal phase change memory on a GPU
color me skeptical that this will be what they actually end up using
>>54408641
here's my guess as to what Navi could be.
the "NextGen memory" and "scalable" clues definitely make something like this conceivable...
>>54408641
>>54409398
> fuck, forget entire point of post...
>>54407467
getting 20-30fps walking through the forest on high on a $300 card isn't better.
>>54406944
only worth it at vega
>>54410851
depends where you're at and what you're willing to pay.
As a GTX 660 user, a <$300 Polaris 10 would be a pretty reasonable upgrade for me.
EVGA 980ti hybrid
Faster than a GTX 1080 that isn't even out yet.
>>54409031
OK, I remember to heard Nvidia talking about "stacked DRAM".
>>54412320
Not sure, the 1080 will perform similar to a 1550 MHz 980Ti. I hope you know that, it runs on stock 1800 MHz.
But I will wait, maybe get a used 1070 or 980Ti below 350 €, or survive with my 780Ti until 1080Ti comes out.
>>54413274
>OK, I remember to heard Nvidia talking about "stacked DRAM".
Either Pascal or Volta was supposed to be Hybrid Memory Cube instead of HBM, but Nvidia supposedly had to switch because of unspecified technical challenges they couldn't get past.
They have called both HMC and HBM "stacked memory", but those standards both use silicon interposers to link GPUs/CPUs with the memory.
What people are speculating about AMD's "NextGen" memory is that DRAM dies will be stacked directly atop GPU dies instead of next to each other on an interposer.
If this happens, Nvidia will no doubt be short behind them, though they've not yet hinted at anything along the lines of direct stacking.
>>54413399
also, HMC is generally a dumb idea for GPUs since it's designed to optionally support full mesh topologies, which necessitates narrow high speed serial buses for each interface, which increases heat, latency, etc. compared to the fat, slower clocked, non-networked IO of HBM.
