I own a Gigabyte 970 G1. So far, i am amazed by how well the card performs. Although the card is massive, i had no problem whatsoever putting it in NZXT H440 case. It looks fantastic. Card is quiet and it runs cool. The highest temperature it reached was 52 while playing crysis 3 on 1440p.

I havent had any experiences with Msi or Asus. But of all these three, Gigabyte's has the highest clocks and you can overclock them further easily.

Let me know if you need to know anything else.

Would the 970 G1 fit into a Corsair 200r? http://www.newegg.com/Product/Product.aspx?Item=N82E16811139018

I would say yes according to Pcpartpicker.

http://uk.pcpartpicker.com/parts/case/?compatible_with=gigabyte-video-card-gvn970g1gaming4gd

I have the EVGA GTX 970 SC 2.0 because it's one of the few that fits in a Silverstone SG05 case. Has a lot of coil whine unfortunately, tested that in 2 different PCs. It is quiet and doesn't run hot, during 2D stuff the fans don't even spin. Other than the coil whine, I'm happy with it.

I don't own it, yet at least, but i've been reading of them a lot lately. As you said the Gigabyte is prolly the biggest, but it's also one of the fastest, and coolest/less noisy, not to mention that apparently it's the most overclockable one (you probably don't care anyway)
The MSI is a good compromise, it's pretty much as fast as the Gigabyte (from the benchmark i saw, it's a little slower than the gigabyte), good cooling ofc not as good as the gigabyte's one, but still, and a lower price compared to the Gigabyte, but not as overclockable. The EVGA is probably the most expensive but still it's not the top, maybe the fastest but im not sure it's worth the downsides. Asus, mmh i haven't really read a lot about that, i reckon it's pretty similar to the MSI, don't know about price and performance tho.

why don't you guys just check the dimensions for yourselves, instead of trusting that stupid website? it's easier than you think...

i went through the same shit as you, spent a couple days trying to figure out which one I wanted (and waiting for those to come in stock), but then I realized something - none of it matters haha. they're almost all identical in power, speed, cooling and sound, the only thing you need to worry about is whether or not it fits in your case and has a third party fan.

I ended up getting the EVGA. I do not have coil whine and the only time I can hear the fan is if I force it to about 50% speed, but it never reaches that by itself when gaming.

I despise my GTX 970. It's EVGA, for what it's worth.

In just about any game I play, my FPS goes from ~250 to ~60. It lasts for about 5 seconds, and it happens roughly every 15 seconds. It makes it a hell of a lot more annoying if you're playing on a 120/144hz monitor, too.

I've seen a lot of people have this issue, but Nvidia seems really stumped about it. The part that's even better about this that I bought the 970 a couple months back. After realizing how shitty it is (and it was past the refund window at this point), I upgraded every single piece of hardware in my computer except my SSD. I formatted, got everything set up, install a few games, and I have the exact same problem with an entirely different computer.

The Nvidia customer support has been less than helpful, with stupid fucks saying shit to me like "oh, well 60 FPS is actually quite acceptable."

lol?????????

I didn't have this shitty problem with any of the older video cards I've tried. I could just use my old video card in the meantime, but my old video card doesn't run Payday 2 that well.

And when I mean any game, I really mean just about any game in any scenario. CS:GO, TF2, Terraria, Payday 2, etc. I could be on the main menu and I would still get these FPS drops. It's pretty fucking bad. The only game that I've seen it not affect is Quake Live. And that's really worthless because a video card from 2001 could make Quake 3 / Live run at 125 FPS.

You could buy it and see if it affects you too, then just return it if it does. I was pretty busy when I first bought it, so I didn't really recognize it until it was too late. So, this fucking sucks for me.

Take the G1.

Performance doenst really matter, you won't really notice a difference anyway.Go for the one that actually fits in your case and has the best cooling/is the most silent one.
I had pretty much every gtx970 in my hands and the Asus Strix seems to be the best choice.

I have an Asus strix and I've been really satisfied with it so far. It doesn't run as cool as my MSI gtx670 PE but it's still runs pretty cool. Can't speak for how well it will oc yet because I havent oced. I don't have any of the coil whine issues that seem to be common with the 970s.

Fragile if you're still covered under the evga step up program you could try to upgrade to a slightly higher priced model and try your luck again. Sounds like a super shitty problem, that sucks man. Just wondering, does this only happen on steam games? Try playing with overlay off.

Only looking at noise levels, load temperatures and overclocks my money is still on the MSI 4G and the Gigabyte G1.

I must admit that I haven't really kept up with everything lately.
From memory, might be outdated or simply incorrect:
EVGA had a lot of issues, also a bit low memory OC iirc.
Gainward and Galax temps a bit too high, also on the loud side and coil whine.
Palit temps a bit high and coil whine.
Zotac temps a bit high and coil whine save for the really expensive ones.
Asus Strix average across the board, no flaws, coil whine levels acceptable, on par with the 4G and G1, doesn't OC as high though and higher temps and noise in some cases (case as in housing). Ok, but didn't impress me and gets beat by the 4G and G1 in every aspect, granted it's close in some aspects but it's still "only" 3rd place after the "i don't know which is better" joint 1st place.

Does this card have any known issues with coil whine? The noise from my 560Ti under load drives me absolutely batty.

Yes just like almost all cards today

gtx 900 serie is infamous for their loud coil whine. my msi gtx 970 had too much that i had to RMA. However the retail store gave me money back instead of giving me replacement because they said they had too many RMA with same coil whine issue. they ended up losing alot of money when they sent customer replacement cards as replacement cards ends up with coilwhine again. so they changed their rma policy regarding gtx 900 series

they mention that this is the first time they had to amend their rma policy since gtx 900 is specifically loud/alot of cards wth noise compare to older gpu cards

What kind of cooling are you using for your case?

I've heard good things about MSI's cards being solid and reliable but I could be mistaken.

I wouldn't even take the overclocking in to account when picking a card because overclocking gpu's without custom cooling wont get you far regardless. It's really not worth it.

I have this on my Radeon HD 7970 too so it can't be an nvidya specific issue.
Although it's less frequent than every 15 seconds, maybe once every one or two minutes.

http://www.lazygamer.net/general-news/nvidias-gtx970-has-a-rather-serious-memory-allocation-bug/

Apparently it was the benchmark that was bugged, no problem with the 970. I would wait a couple days before buying one just to be sure, but if you have one already as is I wouldn't worry.

http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

Also a couple of my friends have Gigabyte 970's and love them. Highly recommended although I know MSI makes some great cards too.

got my brother a Gigabyte GTX 970 G1 recently and it is crazy. he loves it.
runs very cool (doesnt go anywhere above ~50 degrees in the australian summer)
pretty damn quiet (suprised me, but his case is kinda quiet, large fans with noctua cpu cooler)
no coil whine at all

Wrong.

http://hardware.slashdot.org/story/15/01/24/200256/nvidia-responds-to-gtx-970-memory-bug

This is a real issue and if you attempt to address the last 0.5GB of memory you will get terrible frames.

I have the evga version, so far better frame rates, quiet as a mouse. I had the gtx 760 before that. So far, straight improvement

fdgdfggdfgdjhhjum

the VRAM issue with the 970 is real, and it's really troubling. I'm guessing GTX and the card partners are already talking about recalls because the issue's difficult for average consumers to fix, The card itself still might be good, but you will be paying a bit too much for a defective product.

DON'T BUY 970s RIGHT NOW, WAIT AND SEE!

coil whining is pretty much due to bad assembling of the components on the video card pcb, you'll rarely find it if you buy a good branded videocard

It's not a defect, it's the result of a conscious design decision, and thus an intentional bait-and-switch by Nvidia. They won't fix this.

You all need to chill out.

I think I'll have to clear up a few things. Maybe I'll make a flowchart.

1. It's not a bait-and-switch. http://en.wikipedia.org/wiki/Bait-and-switch
The other options would have been to actually just put 3GB on it and cripple the bandwidth throughout the whole memory or not to use colour compression and cripple the bandwidth or not to sell the GTX 970 at all. Clearly a better deal for everyone involved. Technical explanation further down.

2. All those game benchmarks didn't suddenly change. The GTX 970 still performs the same.

3. Who of you would have actually used exactly 3.5-4GB VRAM? The driver is actively trying to keep VRAM usage below 3.5GB so once you manage to go over that you'll probably go over 4GB aswell. That means swapping with the RAM and a lot more performance issues.

Technical stuff:
DISCLAIMER: This might not be correct. I haven't confirmed it myself and I won't dissect a cut down GM204 just to please some people on the internet. So take this with a grain of salt, some of it might be wrong. However that's also true for most of the accusations.

What everyone thinks is the issue won't be fixed because they can't fix it.
The issue is real and I hope they'll fix it on the GM200 and maybe later versions of the GM204, but it's more like a minor inconvenience rather than the absolutely game-breaking, performance-destroying and possibly life-threatening bug everyone makes it out to be.

Most of the benchmarks were also run incorrectly. People didn't bother to read the instructions. Some caught onto that and are now trying to blame the author, because a program that was coded in 30 minutes isn't fool proof. If the benchmark is run incorrectly it can show lower bandwidth for 1GB. It'll also show lower bandwidth or infinite bandwidth towards the end of the memory on every single nvidia card in existence. What it's actually showing is the RAM/swapping bandwidth. The whole point of that benchmark was to find out if the last 0.5GB of VRAM is that much slower than the rest to be able to cause these issues people have reported when using >3.5GB VRAM. What it's showing when run correctly* is that for some reason the swapping starts with 0.5GB left. It's not actually the VRAM being slow, it's the DRAM (via PCIe) being used instead of the VRAM which is incredibly slow.
*headless, the gpu can't be used for a display, use the iGPU, or the OS will reserve VRAM and the weird CUDA memory swapping will show up earlier or before you actually run out of memory on other cards that don't even have that issue

My guess on what's happening in as simple as I can describe it, this is highly speculative and might not even come close to the truth:
1. Nvidia "hardwired" the VRAM adresses to the L2 cache adresses for the colour compression. Because of that, intentionally or unintenionally, data in the VRAM can't be moved/swapped. So once you run out of VRAM and data gets put in the DRAM (the normal RAM) it's stuck there. Normally in case of a page fault (data is DRAM instead of VRAM) the data would get swapped and the one you're using gets put in the VRAM. This doesn't work now, so everytime you need that data it's going to be send from the DRAM via PCIe (which is incredibly slow compared to VRAM). Now that would only be a problem when you're using a bit over 4GB (compressed size). In fact it's only a problem under specific circumstances, namely when the total memory consumption is >4GB but the actively used memory for the current application/task is <4GB. Once you start using >4GB for one application you have to swap anyway. So unless you've got some pretty big stuff in the background/minimized or windows is reserving stupid amounts of VRAM, this isn't an issue.

2. Because of the way they cut down the L2 cache (1.75MB instead of 2MB)/crossbars on the GTX 970 they can't access the last 0.5GB of the VRAM via the normal "hardwired" colour compression way. However they knew about this and made 2 "partitions". The first 3.5GB would be accessed in the "normal" way, like the GTX 980 does for all 4GB. The last 0.5GB would be accessed without colour compression. That way you'd only lose the extra 30% bandwidth from the compression. It's not ideal but acceptable. For that reason the driver tries to keep the VRAM usage below 3.5GB. The only question is did they have to cut down the L2 cache? Iirc they didn't cut it down on the GTX 780, so I'm leaning towards yes.

3. Now it's time to get to the actual issue. The whole thing started when people noticed that the GTX 970 wouldn't used more than 3.5GB VRAM unless it was forced to. The behaviour itself is normal but the limit should have been 4GB like on the GTX 980. Also a 30% drop in bandwidth shouldn't be able to cause the drastic performance problems that were reported. Nai's benchmark indicates that the GTX 970 can't access the last 0.5GB or starts swapping to DRAM even though it can access them. That's not supposed to happen. Everything else is.

Conclusions:
1. There is an issue. But it's not what people think is the issue, it's only related.
2. Unless Nvidia sent cards with the full 2MB L2 cache to reviewers and then sold the cards with 1.75MB while knowing about the problem it would cause it's not bait-and-switch. A reviewer with press and retail versions of the GTX 970 could confirm this.
3. Benchmarks didn't show it because most benchmarks still only use 3GB.*
4. It's mostly an issue for 4K and/or very high settings that the 970 might not be able to handle anyway. It's a bummer in those cases and in SLI because the 970 might not scale as well as expected since it's running out of VRAM. If were really lucky it's just a driver glitch and fixable without physical changes.

*We've been there before, people claimed you need Titan Blacks because the 3GB on the 780 Ti isn't enough for 4K, triple 1080p, 1440p once you go to 256xSSAA (it's a hyperbole). Turns out that neither of those GPUs can get 60fps with 256xSSAA anyway and that people prefer 60fps 4xSSAA to 0.2fps 256xSSAA. Apparently some of the people working on the drivers actually have a clue about what they're doing. In fact, some of them are so good they even get paid for it. They know what happens when a GPU runs out of VRAM so unless there is absolutely no way to avoid it, it won't happen.

from nvidia:

The GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section. When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rdparty applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we use both segments.

We understand there have been some questions about how the GTX 970 will perform when it accesses the 0.5GB memory segment. The best way to test that is to look at game performance. Compare a GTX 980 to a 970 on a game that uses less than 3.5GB. Then turn up the settings so the game needs more than 3.5GB and compare 980 and 970 performance again.

http://i.imgur.com/M31a21k.png

Ok, the germans that started this are bringing out the pitchforks again, because they don't believe nvidias benchmarks. I don't feel like dealing with them but I can at least slow the witchhunt here.

I'll make this as clear as I can:
Most of the time the GTX 970 will use over 3.5GB if it has to.
The last 0.5GB have 30% lower bandwidth and lead to fps drops of <10%.
The driver will avoid this if it can.

However, 3rd party programs might not show VRAM usage correctly, only displaying 3.5GB max. Therefore when people are trying to force >3.5GB they're actually using >4GB. What they are complaining about are the huge performance issues associated with going over 4GB VRAM usage. Those are the same on the GTX 980 and are present on every GPU ever made.

The only legitimate issue is that in Nai's benchmark those 0.5GB don't seem to get used at all. The reason for this might be that CUDA's memory management isn't automatic. To access the 0.5GB 2nd "partition" different/new commands might have to be used and those might not be implemented in CUDA yet. This issue should not occurr in games, since games do not have the same control over the memory CUDA does.