Nvidia GeForce GTX 1000 Series (Pascal) MegaThread: FAQ and Resources

[WilkinzMicawber]

If a single gtx 1080 consumes 400 watts of power under load, what kind of a psu is needed for two in SLI, overclocked, with an overclocked cpu?

Also, weren't all the overclock runs of the Founder's Edition with the fan speed unaltered from the 50% run speed? I would guess the throttling would be eliminated at 100% fan speed.

 

[Deleted member 217926]

I don't know where you're getting 400 watts from. The Founders edition uses about 190w at stock speeds and a highly factory overclocked model uses around 250w at the absolute maximum load using synthetics. In a very demanding game you are looking at ~220w with an overclocked card.

https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X/22.html

You could push that a little more with a more aggressive overclock but nowhere near 400 watts.

I'd go with a quality 850w unit for 2 in Sli with heavy overclocking on everything. 750 watts is probably sufficient though.

 

[JackNaylorPE]

I'm just not sure an article about how 970s overclock is all that relevant to the 1070 and 1080. Several 1080 reviews have come out and all of them mention that they are able to achieve overclocks in the 2.0-2.1GHz range. This is true in the founders cards and it's also true in the Gigabyte and MSI reviews I have read.

I don't doubt that all the parts you talk about are better in the more expensive models but if they have the same overclocking limit as the cheaper designs does that really matter? The FE cards won't perform as well because they will hit that thermal throttle point and lower their clock speeds more often but assuming that the cheaper cards' coolers are sufficient they shouldn't run into thermal throttling much, if ever. If they aren't thermal throttling they should theoretically be able to hold the same overclocks as the better models.

Again, all those parts you talk about probably are better but if Nvidia is artificially capping the amount of overclock they can achieve with their limiters I'm not sure why I should pay for them.

1. Past performance is certainly relevant. Was careful to point out the it's not conclusive, but it is certainly worth considering when no definitive info is available. Is it not relevant for example that recent generation AMD cards typically overclock only in single digits while NVidia cards go up and past 30% ?

2. The FE cards are all experiencing thermal throttling so lets not brag to much how well they OC.

3. The 1080s are showing 13 - 19% OCs (from the limited data available) ... considerably **less** than the 25% we saw on the 980 and the 30% we saw on the980 Ti. It is however comparable to the 17% we say on the 970 which, interestingly enough, was artificially nerfed by gimping the throttling point.

The 1070 ... aka the new 980 .. is showing a 25% overclocking ability on the non-reference cards ... interestingly enough exactly matching the OC of the 980.

All that aside... and somewhat off topic. The point being can reference cards match the performance of non-reference cards with superior components ... this is undeniably relevant, because all that has changed is the GPU.....so if we look at the 970 and we see that when AIBs improved these components, we consistently saw a improvement in performance ... if it was true for the 970, why wouldn't it be true for the 980 ... if it's true for the 980, why wouldn't it be true for the 1070 ?

Historically, the reference cards have never been able to match the over clocking ability of the non-reference cards so yes, the relevance is certainly there....

4. I don't know how you missed it, but yes they clearly do not "have the same overclocking limit" the referenced review clearly shows this and the result is consistent across almost every review I have read This was true for the 9xx, 7xx, 6xx, 5xx and so on. It was most glaring with the 5xx series where the 570 cards were well noted for toasting their VRMs, the reference card was well noted for doing this, the EVGA SC was well noted for doing this but again, EVGA rarely changes anything on the PCB for their SC series.

5. Will the 10xx series change all of this ? Too early to tell, a) can't make a judgment as yet as two few samples to choose from b) Overclocking has entered a whole new world with Boost 3.0 the adjustments that are available and c) higher OCs don't always translate into higher fps.

At techpowerup, in BF3....

The 1080 FE managed a 9.1 rating by the reviewer
The 1080 Gaming managed 9.8 rating

I'll be the first to admit .. nVidia is making it harder and harder for its partners to distinguish themselves both legally and with design limitations. The differences between reference and non-reference cards are shrinking such that spending an extra $100 for a Classified, Lightning and Matrix has long not been justifiable. But shrinking is not the same as vanishing.

Non-reference cards still have more too offer. So, to answer your question as to why you should pay more.... let's compare based upon techpowerups analysis of the MSI Gaming 1080 and FE 1080

The 1080 FE managed a 9.1 rating by the reviewer
The 1080 Gaming managed 9.8 rating

The 1080 FE managed a idle sound rating of 28 dbA
The 1080 Gaming managed a idle sound rating of 0 dbA

The 1080 FE managed a idle sound rating of 31 dbA
The 1080 Gaming managed a idle sound rating of 37 dbA

The 1080 FE managed a Load Temp of 83C
The 1080 Gaming managed a Load Temp of 72C

The 1080 FE managed a OC'd Load Temp of 83C
The 1080 Gaming managed a OC'd Load Temp of 74C

There it is ... the big shazul ... the FE card is throttling at stock settings and its temp cant go up when OC'ing as it's already throttling. Why does this matter ?

During gaming, the card goes above 82°C, which results in lower clocks due to Boost 3.0; see our Boost 3.0 Analysis for more details.

Idle temperatures [on the MSI Gaming 4G] are excellent even with the fans turning off in idle. During gaming, the card also runs much cooler than the reference design, which avoids clock throttling above 82°C.

This means that for the first time in GPU history, lower temperatures directly translate into more performance - at any temperature point and not only in the high 80s. I just hope that this will not tempt custom board manufacturers to go for ultra-low temperatures while ignoring fan noise.

That my friend is what I like to call the "crux of the biscuit". Notice the wide gap that ya get along the curve now w/ Boost 3. See links for images

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/images/boost1.jpg
http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1080/images/boost2_small.jpg

Note how clock frequency doesn't change with heat variations. Now let's look at what happened with the FE card when both cards are subject to a throttling test (most sites don't do this)

The MSI Gaming delivers pool table flat 1910 cock, while the FE is all over the place from 1680 to 1795 .... witht he result being:

It probably wouldn’t be that bad if not the frequency spikes. While average clock is somewhere around officially stated boost clock, those spikes cause micro-stuttering, which negatively affects gaming experience.

Finally ...

The 1080 FE managed 154.9 fps.
The 1080 Gaming managed 159.9 fps.

So is their a difference... yes there certainly is. Overclock the FE all ya want, it's still gonna throttle regardless of what GPU_z calculates what the boost clock **should** be.

And, as expected, the cards are running $20 more than the FE Edition ....

-Would I pay $20 more for the 28 dbA idle / 6 dbA load sound advantage .. in a heartbeat
-Would I pay $20 more for the 9C stock / 11C OC temp advantage .. in a heartbeat
-Would I pay $20 more for a card that doesn't ever throttle .. I'd pay a lot more

Quick look at the 1070 versus MSI Gaming 1070

76C FE load temp drops to 69C w/ MSI Gaming
55C FE VRM temp drops to 40C w/ MSI Gaming
34 dbA idle noise drops to 33 dbA w/ MSI Gaming
40 dbA idle noise drops to 38 dbA w/ MSI Gaming
1975~2050MHz unsteady boost clock jumps to solid 2088 w/ MSI Gaming

The 2088 vs 2050 isn't all that impressive, the steadiness of it is.

I understand your point about cheaper aftermarket cards but that's a big assumption ... are they in fact going to be cheaper ? The 4 cards that generally compete usually come in at the same price (Gigabyte G1 / Asus Strix / MSI Gaming / EVGA SC) ... the SC shouldn't be on that list and should certainly be a cheaper card but historically hasn't been. Ranked by number of newegg reviews, the top 4 970 finishers are ... MSI Gaming 4G 610, Gigabyte G1 499, EVGA SC 472, Asus Strix 458 which indicates, at least to my eyes, that these were the cards that everyone wanted.


Right now, EVGA has two non-reference 1070 cards on newegg ... neither designated as SC. One is $419 (1506 clock), one is $439 (1594 clock).. Is it worth the $2... who knows, you'd have to tear them down and do an analysis and see if the differences pan out. .

 

[bubbleboy04]

So when will the other models be released? I am very interested in the MSI card that is actually going to sell at $599.

 

[Math Geek]

If a single gtx 1080 consumes 400 watts of power under load, what kind of a psu is needed for two in SLI, overclocked, with an overclocked cpu?

Also, weren't all the overclock runs of the Founder's Edition with the fan speed unaltered from the 50% run speed? I would guess the throttling would be eliminated at 100% fan speed.

you clearly need to go back to the first posts and re-read some reviews of the reference and custom cards. 200w was the norm for the custom cards out of the box and higher for some tweaking. founder's editions were about that 180 stock and about 215w or so heavily oc'ed. simply upping fan speeds is not the answer to the FE thermal throttling as it was done during the review.

 

[Math Geek]

So when will the other models be released? I am very interested in the MSI card that is actually going to sell at $599.

my crystal ball is still in the shop getting polished so can't really say 🙁

all we can (and have been) saying is to keep an eye on the stores for when models show. some of the brands sell cards directly and checking there would be a good idea as well. if it's for sale on the site, then it should be for sell elsewhere. pcpartpicker would also be a decent source to check. they end to list what's available and so far little is listed. i'm checking there for info as well in case something slips by.

 

[TehPenguin]

Regarding SLI once again: 25 out of 27 seems like a really good ratio. Alas it's just for the games that were tested. I agree that most AAA titles get multi-gpu support but the world of games has oh-so much more to offer. AAA doesn't necessarily mean graphics. It means big budget.
Now I don't wish to spend too much time analyzing this, but just for a quick overview: my steam library consists of roughly 220 games. I'd wager about 30 of them can be listed as triple-A with all the rest being Indie developed or just by a small studio. Not all 220 are graphically demanding but I am confident in the assumption that at least 100 of them are. Even if all of the big budget games I own had spotless SLI-support, it would still leave us with a 30 to 100 ratio which in my opinion is a handful.

Now regarding thermal throttling on the FE 1080: the 1080 never ever, ever throttles below it's base clock. It should be self-evident that if you want to overclock your GPU to the highest possible clock you absolutely have to consider an alternate cooling option. There hasn't been a card that would overclock as nicely as the 1080 does with a reference cooler. The whole thermal throttling thing is a bias.

 

[king3pj]

Spoiler

I'm just not sure an article about how 970s overclock is all that relevant to the 1070 and 1080. Several 1080 reviews have come out and all of them mention that they are able to achieve overclocks in the 2.0-2.1GHz range. This is true in the founders cards and it's also true in the Gigabyte and MSI reviews I have read.

I don't doubt that all the parts you talk about are better in the more expensive models but if they have the same overclocking limit as the cheaper designs does that really matter? The FE cards won't perform as well because they will hit that thermal throttle point and lower their clock speeds more often but assuming that the cheaper cards' coolers are sufficient they shouldn't run into thermal throttling much, if ever. If they aren't thermal throttling they should theoretically be able to hold the same overclocks as the better models.

Again, all those parts you talk about probably are better but if Nvidia is artificially capping the amount of overclock they can achieve with their limiters I'm not sure why I should pay for them.

1. Past performance is certainly relevant. Was careful to point out the it's not conclusive, but it is certainly worth considering when no definitive info is available. Is it not relevant for example that recent generation AMD cards typically overclock only in single digits while NVidia cards go up and past 30% ?

2. The FE cards are all experiencing thermal throttling so lets not brag to much how well they OC.

3. The 1080s are showing 13 - 19% OCs (from the limited data available) ... considerably **less** than the 25% we saw on the 980 and the 30% we saw on the980 Ti. It is however comparable to the 17% we say on the 970 which, interestingly enough, was artificially nerfed by gimping the throttling point.

The 1070 ... aka the new 980 .. is showing a 25% overclocking ability on the non-reference cards ... interestingly enough exactly matching the OC of the 980.

All that aside... and somewhat off topic. The point being can reference cards match the performance of non-reference cards with superior components ... this is undeniably relevant, because all that has changed is the GPU.....so if we look at the 970 and we see that when AIBs improved these components, we consistently saw a improvement in performance ... if it was true for the 970, why wouldn't it be true for the 980 ... if it's true for the 980, why wouldn't it be true for the 1070 ?

Historically, the reference cards have never been able to match the over clocking ability of the non-reference cards so yes, the relevance is certainly there....

4. I don't know how you missed it, but yes they clearly do not "have the same overclocking limit" the referenced review clearly shows this and the result is consistent across almost every review I have read This was true for the 9xx, 7xx, 6xx, 5xx and so on. It was most glaring with the 5xx series where the 570 cards were well noted for toasting their VRMs, the reference card was well noted for doing this, the EVGA SC was well noted for doing this but again, EVGA rarely changes anything on the PCB for their SC series.

5. Will the 10xx series change all of this ? Too early to tell, a) can't make a judgment as yet as two few samples to choose from b) Overclocking has entered a whole new world with Boost 3.0 the adjustments that are available and c) higher OCs don't always translate into higher fps.

At techpowerup, in BF3....

The 1080 FE managed a 9.1 rating by the reviewer
The 1080 Gaming managed 9.8 rating

I'll be the first to admit .. nVidia is making it harder and harder for its partners to distinguish themselves both legally and with design limitations. The differences between reference and non-reference cards are shrinking such that spending an extra $100 for a Classified, Lightning and Matrix has long not been justifiable. But shrinking is not the same as vanishing.

Non-reference cards still have more too offer. So, to answer your question as to why you should pay more.... let's compare based upon techpowerups analysis of the MSI Gaming 1080 and FE 1080

The 1080 FE managed a 9.1 rating by the reviewer
The 1080 Gaming managed 9.8 rating

The 1080 FE managed a idle sound rating of 28 dbA
The 1080 Gaming managed a idle sound rating of 0 dbA

The 1080 FE managed a idle sound rating of 31 dbA
The 1080 Gaming managed a idle sound rating of 37 dbA

The 1080 FE managed a Load Temp of 83C
The 1080 Gaming managed a Load Temp of 72C

The 1080 FE managed a OC'd Load Temp of 83C
The 1080 Gaming managed a OC'd Load Temp of 74C

There it is ... the big shazul ... the FE card is throttling at stock settings and its temp cant go up when OC'ing as it's already throttling. Why does this matter ?

During gaming, the card goes above 82°C, which results in lower clocks due to Boost 3.0; see our Boost 3.0 Analysis for more details.

Idle temperatures [on the MSI Gaming 4G] are excellent even with the fans turning off in idle. During gaming, the card also runs much cooler than the reference design, which avoids clock throttling above 82°C.

This means that for the first time in GPU history, lower temperatures directly translate into more performance - at any temperature point and not only in the high 80s. I just hope that this will not tempt custom board manufacturers to go for ultra-low temperatures while ignoring fan noise.

That my friend is what I like to call the "crux of the biscuit". Notice the wide gap that ya get along the curve now w/ Boost 3. See links for images

http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/images/boost1.jpg
http://tpucdn.com/reviews/NVIDIA/GeForce_GTX_1080/images/boost2_small.jpg

Note how clock frequency doesn't change with heat variations. Now let's look at what happened with the FE card when both cards are subject to a throttling test (most sites don't do this)

The MSI Gaming delivers pool table flat 1910 cock, while the FE is all over the place from 1680 to 1795 .... witht he result being:

It probably wouldn’t be that bad if not the frequency spikes. While average clock is somewhere around officially stated boost clock, those spikes cause micro-stuttering, which negatively affects gaming experience.

Finally ...

The 1080 FE managed 154.9 fps.
The 1080 Gaming managed 159.9 fps.

So is their a difference... yes there certainly is. Overclock the FE all ya want, it's still gonna throttle regardless of what GPU_z calculates what the boost clock **should** be.

And, as expected, the cards are running $20 more than the FE Edition ....

-Would I pay $20 more for the 28 dbA idle / 6 dbA load sound advantage .. in a heartbeat
-Would I pay $20 more for the 9C stock / 11C OC temp advantage .. in a heartbeat
-Would I pay $20 more for a card that doesn't ever throttle .. I'd pay a lot more

Quick look at the 1070 versus MSI Gaming 1070

76C FE load temp drops to 69C w/ MSI Gaming
55C FE VRM temp drops to 40C w/ MSI Gaming
34 dbA idle noise drops to 33 dbA w/ MSI Gaming
40 dbA idle noise drops to 38 dbA w/ MSI Gaming
1975~2050MHz unsteady boost clock jumps to solid 2088 w/ MSI Gaming

The 2088 vs 2050 isn't all that impressive, the steadiness of it is.

I understand your point about cheaper aftermarket cards but that's a big assumption ... are they in fact going to be cheaper ? The 4 cards that generally compete usually come in at the same price (Gigabyte G1 / Asus Strix / MSI Gaming / EVGA SC) ... the SC shouldn't be on that list and should certainly be a cheaper card but historically hasn't been. Ranked by number of newegg reviews, the top 4 970 finishers are ... MSI Gaming 4G 610, Gigabyte G1 499, EVGA SC 472, Asus Strix 458 which indicates, at least to my eyes, that these were the cards that everyone wanted.


Right now, EVGA has two non-reference 1070 cards on newegg ... neither designated as SC. One is $419 (1506 clock), one is $439 (1594 clock).. Is it worth the $2... who knows, you'd have to tear them down and do an analysis and see if the differences pan out. .

I pretty clearly said that the FE cards don't hold their overclock as well because of thermal throttling so I'm not sure why the majority of your post compared the FE to cards you like better. You say there is an overclocking difference between better and lower end cards but every review I have read has been able to overclock to the same 2.0-2.1GHz range.

As long as cards like the $620 EVGA ACX 3.0 are able to keep themselves from thermal throttling they should be able to hold that 2.0-2.1GHz overclock just as well as more expensive cards like the $680 FTW model with the same cooler.

The reason I said that the 970 articles you kept posting weren't relevant is because I don't remember reading reviews about the 970 saying that they all overclocked to the same clock speed because Nvidia was limiting it there. I seem to remember some 970s reaching higher speeds than others. That doesn't seem to be the case with Pascal cards. Even the reference models can reach the same speeds. The cooler just isn't sufficient to keep them from thermal throttling. If you put an ACX 3.0 (or other custom cooler) on that reference card that probably won't be a problem anymore.

 

[opio]

I just wished I looked around better and picked up the 4gb VRAM versions of the GTX 760 instead of the 2gb ones. But then again I'm not sure that msi had released the 4gb version when I bought my cards

and to people saying they want to sli two 1070's instead of a 1080... just get the 1080 guys, seriously, sli is just a headache, I currently have an sli setup and it is just a big suckfest

 

[U6b36ef]

I just wished I looked around better and picked up the 4gb VRAM versions of the GTX 760 instead of the 2gb ones. But then again I'm not sure that msi had released the 4gb version when I bought my cards

and to people saying they want to sli two 1070's instead of a 1080... just get the 1080 guys, seriously, sli is just a headache, I currently have an sli setup and it is just a big suckfest

You probably saved a fortune getting a 2GB one. Leaving more to spend now when it counts more. Since Pascal make a bigger leap in performance over the previous gen; I think anyway.

The discussion on SLI is really well divided. Those saying its the bees knees, (odd expression). Others saying stay away. The general rule is buy the most powerful single card that you can afford. That's what I read anyway.

 

[detroitwillfall]

so any 10xx model besides founders is the best deal lol

 

[melenie]

I've been running galax EOC gtx 970's in sli (sold last week waiting for 1080 availability) since week of release and I did not have a single problem..EVER..period (not taking into account games which had crappy broken releases which were....broken for everyone). I game at 2k and both cards were peaking max overclocks and running smoothe as anything on a 750w power supply with a heavily overclocked i7 3770k.

Now my question...do those who preach sli to be a headache actually own cards in sli? Or are you just a sheep who pulls words from others mouths? If you don't own sli cards...don't comment on sli. Likewise, if your last sli'd cards were dinosaur 7xxx 6xxx 5xxxx etc series cards...don't comment.

GTX 970's in sli....gave a better dollar/performance value then the 980. This is all anyone should be looking at and again, what people should be looking at with pascal. The discussion of "games not working in sli" is grossly exaggerated.

Anyway, still eagerly awaiting more bench marking on sli setups with both 1070's and 1080's. I've seriously never really remembered there to be such a delay before. Pascal is giving me a headache.

 

[melenie]

so any 10xx model besides founders is the best deal lol

You are getting a card which overclocks smoother, cools more efficiently, is quieter and (imo) looks better....all for a cheaper price (supposedly). Quite sure no one will agree that a founders edition is a more logical choice.

 

[Toothless010]

Also, has anybody found anything online as to when an EVGA 1080 Hybrid or msi GTX 1080 Seahawk will be available on newegg or tigerdirect? Or any other non-reference cards for that matter. It's been over 2 weeks since release and I haven't been able to find anything, except on ebay, but for ridiculous prices.

I have been following the EVGA forum specifically for the EVGA 1080 Hybrid. EVGA is telling folks the Hybrid is late June. Today is the 14th, so in the next 1 or 2. I also have my eye on getting the 1080 Hybrid from EVGA. Mainly for the cooling.

 

[TehPenguin]

I've been running galax EOC gtx 970's in sli (sold last week waiting for 1080 availability) since week of release and I did not have a single problem..EVER..period (not taking into account games which had crappy broken releases which were....broken for everyone). I game at 2k and both cards were peaking max overclocks and running smoothe as anything on a 750w power supply with a heavily overclocked i7 3770k.

Now my question...do those who preach sli to be a headache actually own cards in sli? Or are you just a sheep who pulls words from others mouths? If you don't own sli cards...don't comment on sli. Likewise, if your last sli'd cards were dinosaur 7xxx 6xxx 5xxxx etc series cards...don't comment.

GTX 970's in sli....gave a better dollar/performance value then the 980. This is all anyone should be looking at and again, what people should be looking at with pascal. The discussion of "games not working in sli" is grossly exaggerated.

Anyway, still eagerly awaiting more bench marking on sli setups with both 1070's and 1080's. I've seriously never really remembered there to be such a delay before. Pascal is giving me a headache.

While your feedback on 970's is really helpful, please try reducing your affronts to zero. Calling people sheep or telling them not to comment might blow off your steam a little but this is a forum where it is meant for people to share their opinions.
Like with everything, mileage may vary. You had a wonderful experience with your SLI, which is great, but you can't possibly have played all the games on the market. There still are popular games that are poorly optimized for SLI and the risk of bumping into one is for many people enough to go with a single card. Even if there is a plethora of games that function beautifully, those few games that don't might make some people regret going SLI and it is important that they are informed of such risks otherwise it's false advertising. This whole conversation started because reviewers tested poorly optimized games.

 

[melenie]

I've been running galax EOC gtx 970's in sli (sold last week waiting for 1080 availability) since week of release and I did not have a single problem..EVER..period (not taking into account games which had crappy broken releases which were....broken for everyone). I game at 2k and both cards were peaking max overclocks and running smoothe as anything on a 750w power supply with a heavily overclocked i7 3770k.

Now my question...do those who preach sli to be a headache actually own cards in sli? Or are you just a sheep who pulls words from others mouths? If you don't own sli cards...don't comment on sli. Likewise, if your last sli'd cards were dinosaur 7xxx 6xxx 5xxxx etc series cards...don't comment.

GTX 970's in sli....gave a better dollar/performance value then the 980. This is all anyone should be looking at and again, what people should be looking at with pascal. The discussion of "games not working in sli" is grossly exaggerated.

Anyway, still eagerly awaiting more bench marking on sli setups with both 1070's and 1080's. I've seriously never really remembered there to be such a delay before. Pascal is giving me a headache.

While your feedback on 970's is really helpful, please try reducing your affronts to zero. Calling people sheep or telling them not to comment might blow off your steam a little but this is a forum where it is meant for people to share their opinions.
Like with everything, mileage may vary. You had a wonderful experience with your SLI, which is great, but you can't possibly have played all the games on the market. There still are popular games that are poorly optimized for SLI and the risk of bumping into one is for many people enough to go with a single card. Even if there is a plethora of games that function beautifully, those few games that don't might make some people regret going SLI and it is important that they are informed of such risks otherwise it's false advertising. This whole conversation started because reviewers tested poorly optimized games.

Lets begin. I've not called anyone a sheep...it's more or less a figure of speech. It was not targeted at any individual.
Secondly, If you have a deep enough pocket to consider a sli configuration, I'm sure you won't be too disappointed with the odd occasion where your single card out performs your sli configuration. Furthermore, these problems generally only occur for a short period of time and are rectified with driver update and game patches.

 

[TehPenguin]

http://www.gamersnexus.net/guides/2246-gtx-970-sli-vs-gtx-980-ti

970 SLI vs 980Ti vs 980

Tell me what you think

 

[Mousemonkey]

I think that I will stick with SLi'd 970's for a bit longer and let the early adopters out there deal with the high prices and any bugs that crop up.

 

[TehPenguin]

I think that I will stick with SLi'd 970's for a bit longer and let the early adopters out there deal with the high prices and any bugs that crop up.

No reason not to. It's beyond the point, though. Point being: when SLI scales well, it does it really well(40% quicker than 980ti in best case scenario!). When it doesn't it's unplayable(look min. fps).

 

[Mousemonkey]

Now that Nvidia are dropping support for 3 and 4 way SLi things might improve. I don't have any issues with 2 way SLi tbh and so I don't see why people get so worked up over it.

 

[-Fran-]

I also have anecdotal evidence from the 8800GT era. A friend of mine got a couple and SLI'ed them. He still has them SLI'ed and hasn't complained about them. He is not a full fledged gamer now, but at the time everything ran decently enough.

And, another friend got a couple of 6870s for XFire, following this other friend's example. XFire was also kind to him and one of the 6870's died recently. He, outside of the usual "update your drivers" issue, did not have many issues with XFire either.

I don't have any other close experiences with SLI nor XFire to report on.

Cheers!

 

[TehPenguin]

Now that Nvidia are dropping support for 3 and 4 way SLi things might improve. I don't have any issues with 2 way SLi tbh and so I don't see why people get so worked up over it.

It all depends on the games you play. SLI can make wonders happen(look Witcher 3), I don't wish to argue that. What I try to convey is that you have to be wary of the fluctuating performance between games with an SLI setup.

 

[opio]

I've been running galax EOC gtx 970's in sli (sold last week waiting for 1080 availability) since week of release and I did not have a single problem..EVER..period (not taking into account games which had crappy broken releases which were....broken for everyone). I game at 2k and both cards were peaking max overclocks and running smoothe as anything on a 750w power supply with a heavily overclocked i7 3770k.

Now my question...do those who preach sli to be a headache actually own cards in sli? Or are you just a sheep who pulls words from others mouths? If you don't own sli cards...don't comment on sli. Likewise, if your last sli'd cards were dinosaur 7xxx 6xxx 5xxxx etc series cards...don't comment.

GTX 970's in sli....gave a better dollar/performance value then the 980. This is all anyone should be looking at and again, what people should be looking at with pascal. The discussion of "games not working in sli" is grossly exaggerated.

Anyway, still eagerly awaiting more bench marking on sli setups with both 1070's and 1080's. I've seriously never really remembered there to be such a delay before. Pascal is giving me a headache.

I was the one who said sli is a headache, I have two GTX 760's, so yes, I'd say I have a pretty good idea of what sli is like, and a lot of games are not optimized for it such at Titanfall and Doom 4, I'd imagine that Maxwell in sli probably gets better optimization nowadays than Kepler cards do, but it won't be like that for long. I've made my decision, I'd rather get a 1080 now, then sell my 1080 when the 1080 ti comes out. Plus I'm running a micro ATX mobo so pci-e slots come at a premium for me, I need more storage so I'm going to be getting an Intel pci-e SSD.

Maybe now that nvidia is phasing out 3 and 4 way sli, things might bet better for a two card configuration. Believe me, I understand where you are coming from with dollar amount per performance, I bought my cards when the 700 series was new, so AAA games were optimized for it. But if I could go back I would have at least gotten cards with 4gb VRAM, such at msi's 760 4gb version, but it wasn't on the market when I built my rig and it would've only been about 60 dollars more per card than the 760's I bought and I didn't have enough to get two 770's. When my setup was brand new I was in the top 92% on 1080p firestrike, but benchmarking software is always fully optimized for sli.

If I were to sli again, I'd get a 1080 and then another 1080 when I could afford it instead of selling my 1080 for a 1080 ti, but like I said, I only have so many pci-e slots and I'm going to try just a 1 card config this time around then completely overhaul my setup in about 2, maybe even 3 years only upgrading my GPU and my CPU's AIO liquid cooler with an NZXT X41 in the interim plus I'll be able to get a windowed side panel as I won't need the side panel fan I have now anymore, I need it now because I had to cut the NZXT liquid cooling bracket in half (my cards are itx editions) leaving the VRAM chips exposed so I had to buy copper heatsinks for them. Gonna try a custom loop as well when I do my overhaul.

It all depends on the games you play. SLI can make wonders happen(look Witcher 3), I don't wish to argue that. What I try to convey is that you have to be wary of the fluctuating performance between games with an SLI setup.

Couldn't have said it better myself.

 

[gamerk316]

Now that Nvidia are dropping support for 3 and 4 way SLi things might improve. I don't have any issues with 2 way SLi tbh and so I don't see why people get so worked up over it.

It all depends on the games you play. SLI can make wonders happen(look Witcher 3), I don't wish to argue that. What I try to convey is that you have to be wary of the fluctuating performance between games with an SLI setup.

Look at it this way: SLI is always going to be sensitive to latency. As long as both cards are "fast enough", you won't have any major headaches. The problems occur once you fall below a certain threshold, and the inherent latency limitations start to show up.

Let's face it: A single 970 GTX can run pretty much everything. So will two of them. But remember, SLI is run as AFR: Each GPU alternates frames. Today, not a problem. But what happens when a single 970 starts to bottleneck on certain frames? That's when the problems with SLI/CF become apparent. While lower tier SLI'd cards will produce higher FPS then a single faster card, the single faster card will almost certainly age better in the future.

 

[RobCrezz]

But what happens when a single 970 starts to bottleneck on certain frames? That's when the problems with SLI/CF become apparent.

I dont understand what you mean by that?

When using AFR its only having to render half the frames a single 970. Are you referring to frame time variance?