Hitting the VRAM limit

[aduman]

Hi, I've watched this vid today:

 

 

Paul mentions the 780Ti hitting its Vram limit of 3GB on some of the tests he did and when it does the card gets a huge performance drop. You can see it from the 780Ti's minimum FPS numbers. This got me worried a bit. Everyone has different opinions about VRAM issues these days. I think we can sum it up with these:

 

-The 6GB 780 would run out of GPU horsepower before taking advantage of the extra VRAM. Getting a 780Ti would be a better choice for the extra processing power.

 

-6GB of VRAM would prevent possible future bottlenecks.

 

-If you SLI two 6GB 780s the GPUs could actually be able to keep up with VRAM usage over 3GB and you don't have to worry about Vram issues.

 

I'd really appreciate your opinions on this guys.

 

Thank you!

 

EDIT: This is not relevant to the actual topic but I have to ask this. According to the latest rumours GTX 880 will not be a big improvement over the current high-end cards. Some people are saying it might not be faster than 780Ti. I'm asking this because if this turns out to be true, it's going to be a huge disappointment for lots of people...

[sandals100]

Wouldn't getting an R9 290/x be better overall? The 4GB would help with the VRAM problem and it would actually be able to use it all. It may not be as powerful however it is cheaper.

[Askew]

I think AMD is better for 4K at the moment.

[Altecice]

AMD are better at higher resolutions at the moment. Going by the rumours of the 880gtx having 8GB of VRAM on a 256-bit memory bus worries me.

[I/O]

Wouldn't getting an R9 290/x be better overall? The 4GB would help with the VRAM problem and it would actually be able to use it all. It may not be as powerful however it is cheaper.

But isn't nvidia much better because of Physx, Game wurx, drivers, quality and performance plus EVGA ?

[Katsu-K1L]

But isn't nvidia much better because of Physx, Game wurx, drivers, quality and performance plus EVGA ?

Those are features and and AMD Drivers are not as bad as everyone says they are Good not as good as Nvidia but they're good. Also you heard of a company called Sapphire..? yea they're AWESOME and Amd cards have Good build quality not the ref thou

[Bloodkil933]

But isn't nvidia much better because of Physx, Game wurx, drivers, quality and performance plus EVGA ?

The 4GB of VRAM gives the 290X quite good performance at 4K because of its massive 512bit bus.

[aduman]

I simply don't understand the reason why nvidia is going back to 256bit bus and adding a huge amount of VRAM. I mean what's the point of this?

I'm afraid in the near future 3GB won't be enough even for 1440p-1600p resolutions. jmo though.

[SlaughterSmurf]

nvidia ... best graphics in the world !

 

But I thought nvidia was .... "best graphics in the world and highest market share"

Titan Black bro, at 4k you hit the 4GB limit as well lol

[I/O]

Titan Black bro, at 4k you hit the 4GB limit as well lol

What kind of framerates at that reso ?

[D3MON]

people think im a fanboy or troll sometimes, those people completely ignore my previous experiance, cest la vie.
so let me provide arguements regarding vram from both teams and use cases.

lets assume (and this is all cardboard theory and there will be exceptions) that for 1080p you need 1gb of vram minimum 2gb preferable.
for 1gb of vram to be utilised properly you need (with gddr5) a 128bit bus at 1000mhz +/-10%.

with that as a rule typical cards like my 7850, typical a cards like 7870/gtx660 etc follow thier theory well.
256 bit bus, 2gb (usually).
7970/r9-280x is 3gb on a 386 bit bus and is designed for 1440p(or 1600p) gaming 
a little grey area is the 1gb/1080p thing where 1.5gb is probably better, 1600p is twice as many pixels as 1080p so having a 3gb card makes sense, fed by a 386bit bus.
nvidia saves money on memory controllers by instead going for faster ram, years ago this was the other way round with amd opting for gddr4 while nvidia had a 512 bit bus with ggdr3 on thier 280 (and my beloved gtx285  )

so for 4k a 4gb card on 512bit bus makes allot of sense still, im guessing amd are actually doing this cardboard math.

I will point out that it isnt all that straight forward as jayztwocents has pointed out that if you have less vram you can still be fine, itll jsut have less overhead and be more susceptable to drops.

what ie been worried about for ages is where on a thread people are like "shall i get a 4gb 770?" amd im thinking...would my card make sense if it was a 4gb card? my card in crossfire is effectively an r9-290 (2x4gb 2x256 bit bus's etc) which noone has a problem with but when we see these threads people often says sure its fine or just get a gtx780.

doesnt the fact you can get a 4gb 5670 not make people think "hey maybe this isnt as good as i think it is", im not comparing cards obviously but cmon, if you want your card to last more than 2 years shouldnt you opt for more vram and bandwidth to back it up if you can get it.
the 880 having 4gb on a 256 bit bus is crazy and an 8gb version? unless they're running gddr6 at 3000mhz (12000 effective) thats just plain stupid....

then again maybe the rhumours of the 880 being lower tier being true, maybe nvidias still trying to f*** with peoples heads (hence i use amd) how long did nvidia sell the 690 against the 780? and they sold the gtx650ti along side the 750ti....yeah cus theyre alike..
the whole first letter beign generation, 2nd letter being range is completely broken, shouldnt what we refer to the 750 be the 740?
shouldnt the 750ti be the 750? theres a huuuuge gap from 750ti and 760 what are they going to do make a 750ti-i or a 755 and a 755ti?

eugh ill move on.
ive used a 1gb gddr5 128bit 6670 and was good, ive used a 2gb gddr5 128bit 7770, was good but the extra vram was USELESS
id say 1gb/128bit bus is a minimum that i'd reccomend per 2MP and of course the gpu power shoudl corolate to the resuolution and settings your gaming at I bet id suffer if my 7850 was a 1gb version so i wont buy them even though they're allot cheaper (2nd hand yo!).

for now a 4gb r9-290 will work for 2160p/4k as well as a gtx780 but once games get a bti more demandsing and 4k becomes a bit more common where you say "do i buy a new card or just get another r9-290" it'll be a close call because the vram will start to be used fully (expect to see msi afterburner saying you're using 3708mb of 4096mb),m for anyone running a 780/780ti runnign sli probably wont be a worth while option, then they'll have to go and buy an 880 with its lacklustre memory bus  purely because it has the vram required (very clever nvidia.

my friend has a g92 cored 8800gts (same as 9800gtx+) 512mb, hes at 1360x768 and while 256bit of gddr3 is plenty of memory bandwidth for most games he plays at that res even world of tanks max's out the vram so its worthless him playing on it, yeah yeah sucks to be him but the lesson we can take from this is just like buying a motherboard with usb3 over the same board without, its probably worth the extra few bucks to get more vram proving the card can make use of it.

this isn't even my longest post..

[aduman]

people think im a fanboy or troll sometimes, those people completely ignore my previous experiance, cest la vie.

so let me provide arguements regarding vram from both teams and use cases.

lets assume (and this is all cardboard theory and there will be exceptions) that for 1080p you need 1gb of vram minimum 2gb preferable.

for 1gb of vram to be utilised properly you need (with gddr5) a 128bit bus at 1000mhz +/-10%.

with that as a rule typical cards like my 7850, typical a cards like 7870/gtx660 etc follow thier theory well.

256 bit bus, 2gb (usually).

7970/r9-280x is 3gb on a 386 bit bus and is designed for 1440p(or 1600p) gaming 

a little grey area is the 1gb/1080p thing where 1.5gb is probably better, 1600p is twice as many pixels as 1080p so having a 3gb card makes sense, fed by a 386bit bus.

nvidia saves money on memory controllers by instead going for faster ram, years ago this was the other way round with amd opting for gddr4 while nvidia had a 512 bit bus with ggdr3 on thier 280 (and my beloved gtx285  )

so for 4k a 4gb card on 512bit bus makes allot of sense still, im guessing amd are actually doing this cardboard math.

I will point out that it isnt all that straight forward as jayztwocents has pointed out that if you have less vram you can still be fine, itll jsut have less overhead and be more susceptable to drops.

what ie been worried about for ages is where on a thread people are like "shall i get a 4gb 770?" amd im thinking...would my card make sense if it was a 4gb card? my card in crossfire is effectively an r9-290 (2x4gb 2x256 bit bus's etc) which noone has a problem with but when we see these threads people often says sure its fine or just get a gtx780.

doesnt the fact you can get a 4gb 5670 not make people think "hey maybe this isnt as good as i think it is", im not comparing cards obviously but cmon, if you want your card to last more than 2 years shouldnt you opt for more vram and bandwidth to back it up if you can get it.

the 880 having 4gb on a 256 bit bus is crazy and an 8gb version? unless they're running gddr6 at 3000mhz (12000 effective) thats just plain stupid....

then again maybe the rhumours of the 880 being lower tier being true, maybe nvidias still trying to f*** with peoples heads (hence i use amd) how long did nvidia sell the 690 against the 780? and they sold the gtx650ti along side the 750ti....yeah cus theyre alike..

the whole first letter beign generation, 2nd letter being range is completely broken, shouldnt what we refer to the 750 be the 740?

shouldnt the 750ti be the 750? theres a huuuuge gap from 750ti and 760 what are they going to do make a 750ti-i or a 755 and a 755ti?

eugh ill move on.

ive used a 1gb gddr5 128bit 6670 and was good, ive used a 2gb gddr5 128bit 7770, was good but the extra vram was USELESS

id say 1gb/128bit bus is a minimum that i'd reccomend per 2MP and of course the gpu power shoudl corolate to the resuolution and settings your gaming at I bet id suffer if my 7850 was a 1gb version so i wont buy them even though they're allot cheaper (2nd hand yo!).

for now a 4gb r9-290 will work for 2160p/4k as well as a gtx780 but once games get a bti more demandsing and 4k becomes a bit more common where you say "do i buy a new card or just get another r9-290" it'll be a close call because the vram will start to be used fully (expect to see msi afterburner saying you're using 3708mb of 4096mb),m for anyone running a 780/780ti runnign sli probably wont be a worth while option, then they'll have to go and buy an 880 with its lacklustre memory bus  purely because it has the vram required (very clever nvidia.

my friend has a g92 cored 8800gts (same as 9800gtx+) 512mb, hes at 1360x768 and while 256bit of gddr3 is plenty of memory bandwidth for most games he plays at that res even world of tanks max's out the vram so its worthless him playing on it, yeah yeah sucks to be him but the lesson we can take from this is just like buying a motherboard with usb3 over the same board without, its probably worth the extra few bucks to get more vram proving the card can make use of it.

this isn't even my longest post..

 

Thanks for your time man you made so many good points there. Here's my opinion: nvidia knows 780Ti is a very powerful GPU so they have to cripple its performance to make room for their new releases. The VRAM is the first thing that comes to mind. They could have easily given this card 4GB or more but they didn't. They didn't use a bigger bus either, instead they made the memory faster. It's really simple actually... They have to make money somehow and this is their way of doing it. Is it ethical? Hell no! But this is just how it works guys. Btw, maxwell being power efficient is just a cover-up imo, they are just going to add some of the things they used to ''cripple'' 780Ti.

 

This is coming from a guy who had only nvidia GPUs.

[Najuno]

Thanks for your time man you made so many good points there. Here's my opinion: nvidia knows 780Ti is a very powerful GPU so they have to cripple its performance to make room for their new releases. The VRAM is the first thing that comes to mind. They could have easily given this card 4GB or more but they didn't. They didn't use a bigger bus either, instead they made the memory faster. It's really simple actually... They have to make money somehow and this is their way of doing it. Is it ethical? Hell no! But this is just how it works guys. Btw, maxwell being power efficient is just a cover-up imo, they are just going to add some of the things they used to ''cripple'' 780Ti.

 

This is coming from a guy who had only nvidia GPUs.

Hate to tell you but that´s not entirely true.

 

First of all yes the GTX 780 Ti is really fast. I´ve got 2 of them in SLI and it rocks from 1080p-1440p all the way all the day . But the 3 GB VRAM is too less for 4K since this resolution is very demanding in terms of memory.

 

Then why does nVidia not allow their board partners to come up with a 6GB GTX780 Ti?? And the answer to that is more than obvious, isn´t it? Because of the Titan Black... for gamers it would make this card completly obsolete because no gamer is really after the full double precision performance. And other than that the Titan Black and the GTX 780 Ti are the same. So you pay almost 400 USD more for 3 GB VRAM. And man I am a real enthusiast, but I´m not stupid. I know that we´re not there on the GPU side right now with 4K in gaming, so this would have been a waste of money for 1440p (unless Skyrim with many mods) to go with 2 Titan Blacks instead of 2 GTX 780 Ti.

 

Okay, but why does a GTX 780Ti then have 3 GB VRAM instead of 4?? Well, to understand this, you have to go a lot deeper into the memory interface and its controller. But a 384 BIT memory interface can´t handle 4 or 8 GB VRAM. It can only handle either 3 or 6. So neither nVidia nor the board partners can come up with a custom solution here to make the GTX 780 Ti more suitable for 4K. With that being said the new upcoming Maxwell GPU such as the GTX880 will probably have a 256 BIT memory interface with a bigger cache and can therefore handle 4 or 8 GB of VRAM. Many people are afraid of a 256 BIT interface that it will bottleneck the card big time, but nVidia will not make such a mistake as they want to overcome this issue with the cache.

 

Now take this, what I write here, with a big grain of salt, because this is my very personal opinion on nVidia´s Maxwell: The first cards we see in 28nm architecture will be just the jump into the next gen GPUs from nVidia. Neither will they be very expensive nor very highend (even if they might be advertised as such). Sure the chances are that the GTX880 will be slightly faster than the GTX780 Ti for a way better price. But the reality behind this is, the GM204 (successor of the GK104) will be nothing but a performance chip and a place holder for the GM200 (successor of the GK110). The GM200 will be the highend chip that many enthusiasts wait for. And if nVidia gets it right this could be really amazing. Just imagine a DIE size that allows you to hold the temps on the lower side and get this in 20 or even 16nm with a 512 BIT interface and 8 GB VRAM. And that would be the point were I personally would drop my jaw get 2 or 3 of the cards, put watercooling on them (however they will be called Titan 2 or GTX980) and go all day long for 4K.

 

Well but what about the 4GB VRAM on the R9 290X? The R9 290X (I own one my self in my other machine) is a great card and a very good investment for a fair price. Yes I call this a fair price, because I do not have a budget on building a computer what so ever. The raw power is not as fast as the GTX 780Ti but it is pretty close within 10 percent. And concerning the price and what you get for it is amazing. I just wrote in another thread what AMD in my opinion should have done better at launch. The temps were never too much an issue for me, because I water cool all my stuff anyways in my benchmarking/gaming machine. But as much as I like the Hawaii chip as much I hated the cheap VRAM AMD put initially on the cards. Sure nowadays you have as well Samsung or Hynix RAM on the PCBs which improved this by like a light year. Okay you might think now, that the 4GB VRAM are sufficient for 4K, and yes in most cases it is. But there are games out there where you will depend on the 6GB VRAM of the Titan Black. But that´s not the main problem here. The AMD R9 290X struggles with the same issues that nVidias GTX780 Ti and Titan Black struggle with. These GPUs are just too slow in raw computing power and nothing can change this. Even the dual highend GPUs in 4 CF/SLI can´t resolve the issue due to the bad scaling of more than 2 cards.

 

So what´s the conclusion here? Well for me this is crystal clear a conclusion is always a very subjective point of view and my personal conclusion is clear, I will wait until AMD and nVidia releases and next gen GPU that can handle 4K with rich details and reservers before I even think about 4K. If I have to use a SLI or a CF I don´t care and I don´t mind at all. But they will really have to bring something up on the table that is innovating and not an insult for people that are willing to put down so much money into a hobby. Sure people will say just stick with 1080p and 1440p. And I agree that´s probably the best case scenario for right now and the next 1-2 years. But you need the innovation and people that are willing to be poineers in order to make new technologies cheaper and available to the mass.

One final word about my person, I am building computers already for almost 20 years and yes indeed I´m crazy about hardware. I build up computers for chosen customers (have a little business on the side of my true job) and friends and sure I do service and maintenance on their machines. I have three computers atm in use in my place (due to my passion and better chance of making hardware diagnostics). Most builts are custom made so I do a lot of research when it comes to hardware.

[D3MON]

Thanks for your time man you made so many good points there. Here's my opinion: nvidia knows 780Ti is a very powerful GPU so they have to cripple its performance to make room for their new releases. The VRAM is the first thing that comes to mind. They could have easily given this card 4GB or more but they didn't. They didn't use a bigger bus either, instead they made the memory faster. It's really simple actually... They have to make money somehow and this is their way of doing it. Is it ethical? Hell no! But this is just how it works guys. Btw, maxwell being power efficient is just a cover-up imo, they are just going to add some of the things they used to ''cripple'' 780Ti.

 

This is coming from a guy who had only nvidia GPUs.

might of been a good idea to snip my post .

yeah ive had multiple intel and nvidia rigs over the years and moved to the underdog because i was tired of intels monopoly and nvidia bullcrap schemes and have never been happier, i will move back when they stop, and im sure they'll do that once they stop making money  (never)

gk110 isnt technically crippled (780 is, its like cut down to 75%!!) its the 770 and lower rebrands that suffer (though the customers will say otherwise, because customers know better than the manufacturer right?)

why is 770 (680) kepler, but 750ti maxwell?

so 750ti is maxwell but higher end maxwell is 8xx series? sorry if i dont want to be on my knees sucking that while im being headf******.

imagine if amd released their cards as the 8 series so...

8450, 8570, 8650, 8670, 8750, 8770, (750ti competitor).......

but thier higher tier was, 9830, 9850, 9870, 9950, 9970

then imagine if some of the lower tier was GCN 1.0 and some of the higher tier was GCN 2.0

now to bring it to the current time amd (nvidia) is about to come out with some new stuff on their "BRAND NEW GCN 2.0" architecture with a completely new name...

so nvidias releasing maxwell cards and calling them the 8xx series....but the 750/750ti is maxwell but 7xx series?

also why is the gtx750 a gtx?.....shouldnt it just be a gt? and shouldnt the 750ti be a gts or something? isnt gtx meant to be enthusiast/ high end? did they make a gtx630Ti superduperclock when i wasn't looking?

are we going to have the 850, 850ti, 860, 870, 880, (followed a few months later by an 880ti which hasnt been castrated...) only to find out that theres a 950ti, 960, 960ti, etc that are actually the high end, why does nvidia need to split thier cards across 2 generation numbers, they're screwing people around so that they cant get a grip.

i'll play devils advocate here and say that amd's naming scheme is a little off in places but i can see why they did it. the r7 is low to medium and the r9 is mid to high end...fine, but the r7-265......its my 7850 and is basically a 7870 which when it came out was mid-high end, if you didnt have a above the 78xx line you had a 7950 or 7970 AND THAT WAS IT, so why nuke its name?

the amount of people that think the r7-260x is good and the 265 only a little better dont know the difference (try doubel the memory bandwidth to feed double the frame buffer), IMHO as someone who had a 2gb 7770 the r7-260x/7790 is just a 7770 on steroids without the grunt (or rops) to push the pixels the 2gb frame buffer can hold when its not being limited by its 128 bit bus.

nvidia-hardware aside needs to stick to a propper naming scheme and stop messing around with peoples heads just because they know their customers will blindly follow them regardless.

amd-hardware aside they need to rethink thier naming scheme a little, the r7/r9 thing is a good idea, much like intels i3/i5/i7 line people can easily relate to that even if they dont know the difference between a 4350 and 4460, i think having the r7-350, r7-360, r7-370, r7-380, r7-390 etc would be a better move (with the odd x edition, sure) then next year have the r8-XXX and so on. or just copy what they both did in the past and start from "amd radeon 250", 260, 260x, 270, 270x...

oh and btw guys memory controllers dont like that.

128, 256, 512 bit bus=512, 1gb, 2gb, 4gb, 8gb.

192, 384 bit bus=768, 1.5gb, 3gb, 6gb etc.

typically its 256mb of vram per 64bit bus.

[Najuno]

might of been a good idea to snip my post .

yeah ive had multiple intel and nvidia rigs over the years and moved to the underdog because i was tired of intels monopoly and nvidia bullcrap schemes and have never been happier, i will move back when they stop, and im sure they'll do that once they stop making money  (never)

 

gk110 isnt technically crippled (780 is, its like cut down to 75%!!) its the 770 and lower rebrands that suffer (though the customers will say otherwise, because customers know better than the manufacturer right?)

why is 770 (680) kepler, but 750ti maxwell?

so 750ti is maxwell but higher end maxwell is 8xx series? sorry if i dont want to be on my knees sucking that while im being headf******.

imagine if amd released their cards as the 8 series so...

8450, 8570, 8650, 8670, 8750, 8770, (750ti competitor).......

but thier higher tier was, 9830, 9850, 9870, 9950, 9970

then imagine if some of the lower tier was GCN 1.0 and some of the higher tier was GCN 2.0

now to bring it to the current time amd (nvidia) is about to come out with some new stuff on their "BRAND NEW GCN 2.0" architecture with a completely new name...

so nvidias releasing maxwell cards and calling them the 8xx series....but the 750/750ti is maxwell but 7xx series?

also why is the gtx750 a gtx?.....shouldnt it just be a gt? and shouldnt the 750ti be a gts or something? isnt gtx meant to be enthusiast/ high end? did they make a gtx630Ti superduperclock when i wasn't looking?

are we going to have the 850, 850ti, 860, 870, 880, (followed a few months later by an 880ti which hasnt been castrated...) only to find out that theres a 950ti, 960, 960ti, etc that are actually the high end, why does nvidia need to split thier cards across 2 generation numbers, they're screwing people around so that they cant get a grip.

i'll play devils advocate here and say that amd's naming scheme is a little off in places but i can see why they did it. the r7 is low to medium and the r9 is mid to high end...fine, but the r7-265......its my 7850 and is basically a 7870 which when it came out was mid-high end, if you didnt have a above the 78xx line you had a 7950 or 7970 AND THAT WAS IT, so why nuke its name?

the amount of people that think the r7-260x is good and the 265 only a little better dont know the difference (try doubel the memory bandwidth to feed double the frame buffer), IMHO as someone who had a 2gb 7770 the r7-260x/7790 is just a 7770 on steroids without the grunt (or rops) to push the pixels the 2gb frame buffer can hold when its not being limited by its 128 bit bus.

nvidia-hardware aside needs to stick to a propper naming scheme and stop messing around with peoples heads just because they know their customers will blindly follow them regardless.

amd-hardware aside they need to rethink thier naming scheme a little, the r7/r9 thing is a good idea, much like intels i3/i5/i7 line people can easily relate to that even if they dont know the difference between a 4350 and 4460, i think having the r7-350, r7-360, r7-370, r7-380, r7-390 etc would be a better move (with the odd x edition, sure) then next year have the r8-XXX and so on. or just copy what they both did in the past and start from "amd radeon 250", 260, 260x, 270, 270x...

oh and btw guys memory controllers dont like that.

128, 256, 512 bit bus=512, 1gb, 2gb, 4gb, 8gb.

192, 384 bit bus=768, 1.5gb, 3gb, 6gb etc.

typically its 256mb of vram per 64bit bus.

 

Well,

 

the GTX750 Ti is more a live engineering sample to me. nVidia just wanted to show what´s possible with their new architecture in order to have some performance with very low energy use that´s all. By the end of the day the GM107 is a low end chip that´s good enough for HTPCs or very low budget PCs (but therefore it´s really cool that you can just live of the PCIe without an extra power connector).

 

But the real Maxwell start comes with the GM204.

 

BTW nicely explained with the memory controller .

[aduman]

Hate to tell you but that´s not entirely true.

 

First of all yes the GTX 780 Ti is really fast. I´ve got 2 of them in SLI and it rocks from 1080p-1440p all the way all the day . But the 3 GB VRAM is too less for 4K since this resolution is very demanding in terms of memory.

 

  

might of been a good idea to snip my post .

yeah ive had multiple intel and nvidia rigs over the years and moved to the underdog because i was tired of intels monopoly and nvidia bullcrap schemes and have never been happier, i will move back when they stop, and im sure they'll do that once they stop making

I tried to snip your posts this time.

I'm not experienced about the technical side of this and I apologize for any mistakes I made but your posts was very helpful towards understanding the concept, I thank you for this.

I'll bring it to this: What path would you take for 1440p 144Hz gaming? As this is the best option for gaming at the moment imo. Current gpu solutions imo are not meant for 4K, that is why I'm asking this. The only thing I'm sure about my path is that it will be including SLI in it. AMD is unfortunately not an option since 3D Vision is important for me. (I do simracing)

So many "imo"s was used there..

Thank you!

[SlaughterSmurf]

  

I tried to snip your posts this time.

I'm not experienced about the technical side of this and I apologize for any mistakes I made but your posts was very helpful towards understanding the concept, I thank you for this.

I'll bring it to this: What path would you take for 1440p 144Hz gaming? As this is the best option for gaming at the moment imo. Current gpu solutions imo are not meant for 4K, that is why I'm asking this. The only thing I'm sure about my path is that it will be including SLI in it. AMD is unfortunately not an option since 3D Vision is important for me. (I do simracing)

So many "imo"s was used there..

Thank you!

1440p 144hz? Well if you want to actually get 100+ FPS on ultra maybe CF 290x or SLI 780 ti? For high/medium CF 290 or SLI 780.

[Najuno]

  

I tried to snip your posts this time.

I'm not experienced about the technical side of this and I apologize for any mistakes I made but your posts was very helpful towards understanding the concept, I thank you for this.

I'll bring it to this: What path would you take for 1440p 144Hz gaming? As this is the best option for gaming at the moment imo. Current gpu solutions imo are not meant for 4K, that is why I'm asking this. The only thing I'm sure about my path is that it will be including SLI in it. AMD is unfortunately not an option since 3D Vision is important for me. (I do simracing)

So many "imo"s was used there..

Thank you!

 

The question is not what path would I take, it´s more what path am I taking . I´ve made my decision for now a have a GTX 780 Ti SLI with a custom water cooled loop and will buy the new Asus RoG Swift monitor as soon as it is available in the US. Currently I am using a Dell 27" 1440p @ 60Hz and the GTX 780 Ti SLI is awesome for this resolution .

[7850OC]

i still wonder if nvidia will launch the 880 line with 2GB of Vram 

[Najuno]

i still wonder if nvidia will launch the 880 line with 2GB of Vram 

Hopefully not this would be a huge step back...

[aduman]

The question is not what path would I take, it´s more what path am I taking . I´ve made my decision for now a have a GTX 780 Ti SLI with a custom water cooled loop and will buy the new Asus RoG Swift monitor as soon as it is available in the US. Currently I am using a Dell 27" 1440p @ 60Hz and the GTX 780 Ti SLI is awesome for this resolution .

 

Thank you Najuno, there is just a tiny little step remaining for me: Which model for 780Ti? Is the reference cooler better for SLI? My case is a Obsidian 550D and it has two intakes on front, one intake on bottom next to the PSU, one exhaust on the back which is the 140mm fan behind the radiator for the H90 liquid cooler. I can add fans if necessary. Again thank you, I really appreciate your help. This forum was the best one I've been in, no doubt.

[Hieb]

But isn't nvidia much better because of Physx, Game wurx, drivers, quality and performance plus EVGA ?

 

Does anyone even use PhysX anymore?  Also drivers suck on both... there was about a year's worth of drivers that caused me crashing every driver between 314.22 and 332 made my games crash daily. But yeah, EVGA is a pretty good selling point lol

[Tainted]

Does anyone even use PhysX anymore?  Also drivers suck on both... there was about a year's worth of drivers that caused me crashing every driver between 314.22 and 332 made my games crash daily. But yeah, EVGA is a pretty good selling point lol

 

Last time I used PhysX, Batman AA. I wasn't impressed, my next gpu went back to amd.

[Najuno]

Thank you Najuno, there is just a tiny little step remaining for me: Which model for 780Ti? Is the reference cooler better for SLI? My case is a Obsidian 550D and it has two intakes on front, one intake on bottom next to the PSU, one exhaust on the back which is the 140mm fan behind the radiator for the H90 liquid cooler. I can add fans if necessary. Again thank you, I really appreciate your help. This forum was the best one I've been in, no doubt.

 

No problem man, I am always up for helping nice people in the forum .

 

I know the design of the Corsair 550D it is a pretty good case and I like the overall built quality. You are geared up pretty good so far and that 2nd GTX 780 Ti will fit perfect in your built. But it is a matter of truth that more fans will not help you with the cooling of the first GPU. And since you have already a AIO watercooler in there I wouldn´t bother and stack more fans in there.

The best case scenario depends on 2 things IMHO. First you need to look at the overall air flow in your case. You want to make sure that you can maintain a positive air pressure in it. That gives you better cooling for all passive elements and helps you avoiding getting too much dust in your machine. With that being said a 550D has good airflow, and I do not see too many issues here.

Second thing is you should think about the possibilty of adding later down the road a watercooled loop. If you are sure that you won´t do thast you can go with what ever design of the boardpartners you want, but if you have been thinking about it I´d buy EVGA´s ACX design, this is perfect for aircooling and the PCB meets EK´s requirement for the standard full cover waterblock .

Then there´s another point, the first card needs to have a suction cooler design such as the reference design. If not the temps will go really high and I mean a lot higher that necessary. I tested this myself with the EVGA GTX 780 Ti ACX. When this card was in 1st slot I had in Unigine Valley around 83° Celsius. When I put the card in 2nd slot I ended up having 58° Celsius. I mean that´s a huge temp delta right there.

The best way for a 2 way SLI on aircooling for performance is 1st card in reference design (suction) and 2nd card in a design such as EVGA´s ACX (blower). Or for sure if you want it for the looks when you have a windowed side panel 2 ref cards .

 

Sorry I wrote again a lot.. another wall of text. But this might help you I think .

 

EDIT: As a personal remark from me... a GTX 780 Ti SLI is totally kick ASS . BTW haha I like your avatar Frank Drebin ftw .

[Acquire]

My 780 Ti SLI setup involves two blower coolers (MSI's Twin Frozr).  They generate tons of heat.  My HAF X can handle it fine though.  Not sure if I would be comfortable with that setup in a smaller case like the 550D.  The performance is beautiful at 1440p.  Can't wait for that Rog Swift to finally launch here.