Nvidia GeForce GTX 1660 Ti vs. GTX 1060 vs. GTX 960

[HEXUS]

Solid improvements over the ages, right?

Read more.

[philehidiot]

I was kind of hoping this would shut a few people up (including me).

I fear fuel has been added to the flame war....

As you say, a good enough card but without good enough pricing. Given the relatively minor refinements AMD have made to their mid range cards, they'll have saved a fortune on R&D and probably have a lot more pricing flexibility than Nvidia who have just spent a fortune on their (currently disappointing for me) new tech.

[DanceswithUnix]

The numbers reveal the graphics improvement was greater between GTX 960 and GTX 1060, for all types of games, while the gap between introductions was a lot shorter. By rights, then, GTX 1660 Ti ought to have doubled the GTX 1060's performance.

What???

960 was a 28nm design, 20nm didn't happen for GPUs, we went straight to 14nm/16nm.
So 960 to 1060 was no less than two full process shrinks, and a change that we had been waiting *years* for as one 28nm product seemed to roll out after another. Now we drop to 12nm, but that's just a process tweak over 16nm and not actually a shrink at all.

So really, we can't expect the 1660 to be much faster than a 1060 unless it costs more, unless you expect the extra die size to come out of Nvidia's profits.

I think it is a real shame that you didn't use a 4GB 960 card, other than as a warning to people thinking about buying 3GB 1060 cards that when GPUs hit a memory wall they hit it hard.

[EvilCycle]

Two of the Forza Horizon 4 tests (1080p & 4K) state a 1660Ti where the 960 should be.

[philehidiot]

I think it is a real shame that you didn't use a 4GB 960 card, other than as a warning to people thinking about buying 3GB 1060 cards that when GPUs hit a memory wall they hit it hard.

I had a GTX780 which had 3GB and it was definitely the memory which was the first limiting factor. It was a decent amount for the time but.....

I now have a Vega64 and I'm actually a little concerned that 8GB of VRAM just won't be enough in two years. And this card is budgeted for to last way more than that. I work on cost per year for things like this and so it's the cost over time that is important which means you take longevity into account. I do it with everything including motorbikes, razors and guns - realistically, how long is this going to serve my needs for and is it worth the investment? For a motorbike, an initial investment of a grand a year seems reasonable, for a razor.... well a straight edge will last longer than me and for a gun, well that depends on what range I'm shooting and so on. For some indoor ranges an air rifle works just fine and for others you need something with a little more poke but as skill increases you need something that you can take advantage of, which is where the initial investment over time comes in.

The reason I was rambling on about all that is that this article gives you an idea of how long a mid range card is going to be good for. We have a 4 year old mid range card which is simply put, NOT going to be any good in a modern system. Honestly trying to play Metro at 4K on that poor old thing was just mean. Made me feel like they were flogging a trusty old steed. So, it might be worth considering something faster that'll last longer but it might also work out a similar price over time to get a mid range card that suits your resolution and renew it in 2 years rather than 4-5. The rate of progress slowing is interesting and makes life very difficult. These days you might now expect quite a bit more life out of a mid range card as consoles slow progress and we've seen this from AMD where the mid range really doesn't get much more than a tickle and some more make up every year.

Mr Unix, sir. I think you're overthinking the situation from most purchaser's perspective. Imagine I'm a punter. Do I care about how many nanometers my transistors are or the EUV process and the (frankly utterly awesome) advancements in fabs to do this? No. I look at the performance I'm getting for the money I'm spending and if this performance isn't increasing in line with previous performance jumps whilst the price goes up, I'm going to feel screwed and concerned as to whether this performance jump is going to be enough to service future games for the life of my system.... again, remember a lot of purchasers of this kind of card buy the PC as and see the PC as a box that is binned and replaced when the box slows down (I remember when someone asked if, when I mentioned I build my own PCs, I choose the box, monitor, keyboard and mouse and put them together....).

I think there's also a difficulty in pitching these reviews. Me, you and many others on here get the technical challanges and so on because we take the interest. But Hexus is NOT an in depth technical review site. For that you need to look at Anandtech, etc (where I can read it, but it's mostly gibberish to me) and you can expect the technical challanges and details to be accounted for there. Whilst I expect that those working at Hexus do know and understand all of this stuff, I expect they also pitch their articles carefully and to go into such things might alienate a lot of their target audience. That is the tech savvy but interested more in the consumerism / performance side of a product rather than how it looks under an electron microscope.

The irony is, I probably spend more time reading about the technical details of GPUs than I do playing games whereas the 1660 is probably aimed at those people who actually play the damned games rather than reading what runs them.....

</ramble>

[Iota]

I look at the performance I'm getting for the money I'm spending and if this performance isn't increasing in line with previous performance jumps whilst the price goes up, I'm going to feel screwed and concerned as to whether this performance jump is going to be enough to service future games for the life of my system....

My last Nvidia card was the MSI GTX570, which cost about £300. My new one is the RTX 2080 FE which cost £750. Without looking at benchmarks I'm pretty certain I'm getting quite a bit more than a doubling of the performance, so although the price seems expensive, I'm not overly concerned. Plus when I had the 570 I was gaming at 1680x1050, now I'm gaming at 2560x1440. To me, it's a justifiable cost when I look back at what my old card could do.

I just hope the 2080 doesn't die like my 570 did, which got replaced with an R9 290 (was a great little card for the cost).

[DanceswithUnix]

Mr Unix, sir. I think you're overthinking the situation from most purchaser's perspective.

But that is why tech sites need to give clear advice, and I don't think that was present here. It's basically a graph of performance over time with just three points, complaining that the current increase isn't as good as the previous increase. That's bad science, you need more points than that to see a trend and whether this card is poor increase or the 960 to 1060 was an above average increase. But then the cards aren't comparable, the 960 was available in 2GB and 4GB variants, the 1060 in 3GB and 6GB variants. So this jumps from the crippled 960 to the top end 1060. So is the average punter going to make of the conclusion:

The numbers reveal the graphics improvement was greater between GTX 960 and GTX 1060, for all types of games, while the gap between introductions was a lot shorter. By rights, then, GTX 1660 Ti ought to have doubled the GTX 1060's performance.

well I think they will be confused as to what to buy. You can still get 1060 cards for a decent price, maybe they will get one of those, and perhaps as the memory size angle was brushed over they will get one of the 3GB ones. That would be a fail in my eyes.

I think for this article to start getting meaningful results it needs to add a 960 4GB card, and a 1060 3GB card. To come to the conclusion about how much performance to expect per generation, you need to include 760 and 660 cards at least. In the same way that this article seems to be just picking a few select results from the Palit 1660 ti review, you could probably do this by using different reviews for each generation of card.

There is no doubt that overall performance increases are slowing. Moore's Law is pretty much dead at this point, and the days when I expected to buy a new mid range card (for about £130) every 18 months with double the performance of the previous card are long gone. That stability should make purchase decisions easier, but with the street price of these mid range cards climbing (my 8600 GT was about £70 new) and the expected life to be longer it feels like more of a big deal to choose.

Your question about memory sizes is an interesting one, and potentially hard to write an article about as some people will just turn the graphics settings down one notch and not think more about it, others will demand to stick to super ultra settings, so there are multiple points where a card could be said to hit a wall. Unless you really cling to those high settings, you should be OK with 8GB though, given how many 6GB cards Nvidia are selling games devs are going to have to tune for that for years to come.

So although my reply might have been overly technical, I don't think I'm overthinking this. If people want a clear steer on which card to get, this isn't it. It doesn't choose a clear champion, it doesn't point out pitfalls.

[DanceswithUnix]

Interesting data points, using the Anandtech Bench:

260 to 460: https://www.anandtech.com/bench/product/664?vs=542
460 to 560: https://www.anandtech.com/bench/product/542?vs=543
560 to 660: https://www.anandtech.com/bench/product/1661?vs=1597
660 to 760: https://www.anandtech.com/bench/product/1778?vs=1745
760 to 960: https://www.anandtech.com/bench/product/1745?vs=1744
960 to 1060: https://www.anandtech.com/bench/product/1744?vs=1771
1060 to 1660 ti: https://www.anandtech.com/bench/product/2301?vs=2429

Plenty of small increments in performance there, showing the 1060 was indeed one of the larger jumps.

Their results for 960 vs 1060 seem to agree with the article here, so it looks like they too are using a 2GB card, so not much to learn there.

Bench was awesome, Anandtech seem to have dropped updating it in recent years though. I'm sure a more storyboard driven version would be even better, so you can select mid budget cards and 1080p and it digs out what is relevant to you rather than having to already know which cards to select.

[scaryjim]

tbh what I really took from the article was that the 1060 was more expensive than the 960 because it was essentially a different class of card.

Indeed, if you dig back through the archives to the 960 launch that kind of confirms that the 960 is an aberration in that class of card - the 760 launched at $249, just like the 1060. The 960 was cheaper, and it showed in the initial reviews where a Strix 960 was barely any faster in real world performance than the comparison 760.

The 1060 -> 1660 Ti generational change isn't the aberration here. The 960 was just a slow card for this market segment, which was reflected in its launch pricing.

[Iota]

Bench was awesome, Anandtech seem to have dropped updating it in recent years though. I'm sure a more storyboard driven version would be even better, so you can select mid budget cards and 1080p and it digs out what is relevant to you rather than having to already know which cards to select.

It's a useful tool, up until you're comparing older GPUs with newer GPUs, probably in part due to a change in the benchmark software being used.

[007]

My last Nvidia card was the MSI GTX570, which cost about £300. My new one is the RTX 2080 FE which cost £750. Without looking at benchmarks I'm pretty certain I'm getting quite a bit more than a doubling of the performance, so although the price seems expensive...

Here is the issue, looks like you are one of the price insensitive customers but many (maybe even most of them) are, and would consider adding 150% to the cost of a single components a lot.

I am very occasional gamer, thus not very demanding and used to buy x70 class cards or equivalents when the price dropped. I am now priced out of the market and forced to buy either used or inferior cards. Not that I cannot afford it but I simply can't justify spending such significant amounts on something I use for several hours a month just because someone, having limited competition wishes to earn for the shareholders more.

I am not angry at Nvidia, I just won't buy their card and if necessary, switch to consoles completely.

[LemmingOverlord]

The BFV 1920x1080 and 2560x1440 values for the GTX 960 seem to be wrong. It looks like the card performs better in QHD than in FHD... and that's impossible.

The 2GB GTX 960 should be (at best) compared to the 3GB GTX 1060, for price-point similarity. If you wanted an honest apples-to-old-apples comparison you'd need to test the 4GB version of the GTX 960... otherwise it is handicapped.

Having the GTX 960 strapped to a 128-bit memory bus with 2GB is a waste of editorial, tbh.

Price-point wise it isn't a fair comparison either: The 2GB 960 launched at $199, the 6GB 1060 at $299 and the 6GB 1660 Ti at $279. Pricing seems to be in line with what we expect the mainstream to be...

Of course we Europeans get rightly fooked in the currency exchange, and that is something I'll never understand.

The testing is all ok (apart from what looks to be a SNAFU with the BFV numbers) and it's a good article, but I wouldn't say that the generational improvement is so much more from GTX 960 to GTX 1060 because you hobbled the 960 severely and didn't start from similar price-points.

[Ozaron]

Okay, but why 1660Ti and not 1660 non-TI or RTX 2060? Feel like I'm missing something.

[philehidiot]

Lots and lots of typing....

Bench was awesome, Anandtech seem to have dropped updating it in recent years though. I'm sure a more storyboard driven version would be even better, so you can select mid budget cards and 1080p and it digs out what is relevant to you rather than having to already know which cards to select.

You're right about the poor method, I have to admit. As for comparing relative performance, I use gpu.userbenchmark.com.

I think if they were to prove anything useful from this then you're right, they'd want multiple performance measurements using the same point in the product range (although hard to determine sometimes) over time and then we'd see a real pattern. I guess it's a problem of time investment and availablility of the older cards to test...

[illusio13]

Excellent article; without any thought except generational leaps......... :-)

More like this please guys.

Have a RX 580 8G......; see no reason to give the 1660ti even a look in; forget about any AMD bias i sometimes carry; the V56 is my next natural upgrade, they ( Nvidia ) can rubbishrubbishrubbishrubbishOFF with that limiting DDR pool........; Ohhhh can it run current games, what @ 1440p OK with said 6GB pool........., rubbishrubbishrubbishrubbish; tomorrow it will stutterrrrrrrr and Lag because of that pool..................... :-( WTF Nvidia....., atleast give me a reason to upgrade........, not down/side-grade... numptys :-()

[DanceswithUnix]

It's a useful tool, up until you're comparing older GPUs with newer GPUs, probably in part due to a change in the benchmark software being used.

That's just hard to do. I have an 8600GT in the garage if anyone wants to compare that 256MB card to a 1660ti for a mid-range comparison, but I'm really not sure what game you would run on both of them, DooM 3 perhaps?