1060 suggestion

[John Naylor]

@John Naylor You will find all of my advice on rigs is tailored to be as cost effective as can be while not losing performance you'd want at time of purchase or within a reasonable time frame afterwards (1,5 - 2 years). I really like your open mind and analytic take on this though, so here's some more of my 2$

Its always difficult to gauge performance 'in a year or two' because you have to decide something today, but history so far has proven me right. VRAM is always *ALWAYS* the limiting factor with an Nvidia GPU. I could have still used my old GTX 780ti (!) if it weren't for its 3GB VRAM.

The proof is in the pudding as the saying goes. When alienbabeltech they did their investigation at 5760 x 1080. If VRAM was a limiting factor, the game would have performed better with 4 GB instead of 2GB. TPUs testing very clearly shows that performance does not improve with more RAM up thru 1440p. The 1060 6GB with more shaders has a 6-7% edge over the 3GB version at 1080p. So again, if VRAM matters at 1440p, then the 3GB should fall further behind at 1440p. It does not. When we move to 2160p, then we finally see the 3GB fall further back ... 14% as I recall.

If we are going to say that 3GB is not enough for 1080p, then we must accept the fact that 11 GB 1080 Ti is not enough for a 2 x 2 stacking of four 1080p screens ... in other words, a 4k Screen. I have yet to see a single test where any 3 GB card fails at 1080p or 6Gb fails at 1440p. And I'm not talking about some youtube yahoo who claimed well if ya do this and change that, then do this ... the kinds thing that was used for the 970 3.5 GB hulabaloo. I could not understand what the huff was about as any set of special actions / settings you could create a problem with for the 3.5 GB 970 always had the same effect on the 4GB 980.

VRAM is mostly about moving pixels. However in today's world we have the impact of poor console ports which for reasons I don't fully understand eat up lots of VRAM. But outside that instance, I have yet to see any negative impact at 1080p / 144 Hz) on the two 3 GB 780s here. The next room has 1440p, 144 hz with 8GB and it's handled all that is thrown at it .... and the 3440 x 1440s having no problem with 8GB (1080) or 11 GB (1080 Ti). I could see the 11 Gb as coming up a little short and i really didn't get why they opted for 11 vs 12.



Ca

I'd only get the 3GB if you wanted to play 3yr (or more) old games. Like you just wanted to catch up on the last several years. Which is still great, if that's what you want. You can run plenty of modern stuff fine @1080p with 6GB though.

That's not what TPUs test results are showing though, nor anyone else for that matter. At 1080p, 3GB cards are doing just as well as 6GB in TPUs **current** game test suite (one used on 10xx series cards) . Not exactly a bunch of 3 year old games.

https://www.techpowerup.com/reviews/MSI/GTX_1060_Gaming_X_3_GB/26.html

Looking at TPUs test suite results .... the article explains 3GB and 6GB models are different in that the 6GB gas about 10% more shaders ... so the 6GB with more shaders will be faster than the 3GB regardless of VRAM. Now at 1080p, the speed advantage of the 6GB with the extra shaders is 6% faster on average than the 3GB. So if VRAM is in fact an issue... then when we look at 1440p, we should see the 3GB fall significantly further behind ... but it doesn't. After allocating for the shaders, we don't see any performance impact between 6 GB and 3 GB until 4k resolution. Only games in the test suite where the 6 GB did more than 1-3 fps better was Hitman and Tomb Raider. For Hitmand it would appear that VRAM may be a factor but RoTR actually does better at 1440p by comparison than 1080p so something else going on there.

This was first shown by alienbabeltech years ago with two otherwise identical 770s ...2 GB and 4 GB. The site no longer exists but you can see the test results here:

The only time they saw a difference when the game was at unplayable fps regardless of which one was used.

Ya have to wonder why we don't see the same card offered with different amounts of VRAM. I have to think, after Pugent Sounds 6xx series test, alienbabeltceh's 77x test and Guru3D 99x test that showed no significant impact at in performance at 1080p / 1440p (and even 5760 x 1080) when doubling RAM on the same card, that's just not something they want to put out there. Yes you can always find an exception or two, poor console ports being the most obvious, but as yet I have yet to see anything that 3GB @ 1080P isn't enough for the great majority of current games. The question that's gotta be asked is. If the $390 MSI 3GB gets ya 57 fps ... is it worth paying $160 - $20 more for the MSI 6 GB model **with the 10% more shaders** that manages 59.6 fps ? There's 18 current games in that test suite and 16 of them have similar results.

 

[Vayra86]

There's 18 current games in that test suite and 16 of them have similar results.

And its the two that fall apart where it matters - that is what consistency means

Game engines evolve and you still speak of crappy console ports but these days the issues are different ones:
- Dynamic resolution scaling. Game engines do this themselves nowadays. Go ahead and fire up Resident Evil 7 in its basic settings and its a blurfest. You NEED to scale resolution to 2.0x to even have remotely sharp image quality. In other words, you're rendering 2x 1080p internally. Dynamic resolution scaling is also a highly effective way to get rid of jaggies when combined with temporal AA. But it sure does like its VRAM. Already you can see this technique is winning terrain fast and it makes sense because it caters to the resolutions offered by the different console versions, with PS4 and X1 upgrades.

- Again, consoles define the new mainstream: 6GB is that mainstream. Do you buy your GPU for nice gaming today, or would you like it to do well a year from now?

- A performance summary diminishes the exceptions (both positive and negative) and creates a nice hierarchy, but it fails to show where cards shine and where they do not. They provide a simple, overall look at things. Not in-depth.

- Check out these gaps, that's an awful lot of FPS (and: we're still only looking at averages here!) for 10% shaders
https://www.techpowerup.com/reviews/MSI/GTX_1060_Gaming_X_3_GB/17.html
https://www.techpowerup.com/reviews/MSI/GTX_1060_Gaming_X_3_GB/21.html

That's a 3GB card demolished at 1080p by a gen older, 4GB counterpart. 4K? almost 50% gap. And this is also at a performance level where the FPS gaps do matter, even with tweaked settings you are edging around playable or unplayable here. That is a bit of a different perspective than saying 'oh but combined across all these games (of which a vast majority is built for an older GPU gen that maxed out at 4GB realistically!) that 10% shader count only provides 6% perf'... isn't it?

As for the 770 - useless comparison because it lacks the core power for 4GB, which is proven by the massive performance gap that the 970 provided over it. The 1060, with or without 6GB does not lack core grunt in comparison.