The gap seems to be :
RX550 2GB : 80%***
Base 1050ti : 100%
1060 3GB : 110%
1650 4GB / 1060 6GB* : 120%
570 / 480 : ~135%***
580 : ~145%***
I *starred* the Polaris, +, and Polaris12 results because they really are uneven. In a few random titles, they will be even faster than this average, but at the same time in some titles they are way off the mark. One example is modern RSS or Fortnite, the 480/570 is actually slower than the 1650 there. And in BFV they're about dead even. In most games the gap is more like 10-20% in favor of 480/570.
It's just that the edge cases make it worth looking at individual titles if you or the customer is one of those diehard fans that plays one title very heavily and not much else (perfect example my buddy Joel and BF series, from 3 all through V now lol), then definitely look up how each option handles that particular game. If it's just to be a wide open 'play a bit of everything' box, then those averages above are a good rule of thumb.
The other thing to consider, and why I placed the 1060 6GB where it is, is AIB quality and range. Some models are mostly in a very tight group performance wise, with no extremely large reason to favor one brand/model vs another. Other GPU models seem wildly variant. 1060 6GB and up in Pascal are definitely one of the latter. Depending on 10/11Gbps memory, power delivery range, stock clocks vs boost potential and of course binning and cooling options, you could see HUGE variance between the options, that could even be a different class from what you might expect.
What do I mean exactly by that? Well, my favorite example is probably because Pascal lasted so dang long that it sort of evolved. The initial GTX 1080 Founders Edition was this :
1607 Base, 1733 Boost, 10Gbps Ram, with a cruddy blower cooler held over from olden days design. 1
I ran a 970 for a long time for my personal box, before getting a steal on a 1060 6GB EVGA OC variant, which was a minor upgrade (mainly in VRAM starved stuff like modded Skyrim/Witcher 3). Then, opportunity struck and I was able to buy a GTX 1080 vs Asus, the Strix model after the 11Gbps revision of the lineup. This was about halfway through the Pascal's extremely long duty before RTX. And this model had :
1700/1835 clocks, with 11Gbps Micron and a WAY better HSF that meant it could basically sit at max clocks all day, and OC even further if you wanted. Already about 15% faster than 1080FE in most situations, it could go easily beyond 2Ghz and open the gap far wider due to the overall build quality and power delivery.
Where this gets relevant is how cards are measured when new models are released. It's obviously beneficial to present the new models in their best possible light, and since the Nvidia FE Maxwell and Pascal cards were the early models with the worst performance, they invariably choose those as examples despite them being rare in actual overall sales. It was more common in the 780/780ti days, but in 900/1000 timeframe, I *very* rarely see FEs in the wild. People widely bought superior AIBs with better performance after they became available shortly after launch. The FE was Nvidia being cheeky and taxing launch day diehards. So when RTX was compared to FE Pascal, it looked better than it was. But in actual fact, a good AIB 1080 was as good as a 2070, and a good AIB 1080ti like the Aorus etc was as good as a 2080 (better arguably with 11GB vs 8GB if RT didn't interest you).
Nvidia changed this up with RTX FEs. Perhaps knowing that they needed to justify the immense prices a bit more, these FEs actually were quite nice compared to AIB levels, often being as good or better, and only the best AIBs exceeding FE specs, mainly due to power delivery and sustained clocks. Eg; the 2060FE is better than the 2060KO and other cheapo 2060 models.
So that adds an extra wrinkle to A/B comparing your options. Say you have an option to get a XYZ OC model Nvidia vs an ABC OC model AMD, then you have the added research to find the gap vs typical models. Some vary little, some vary massively. Whee
Hence this incredibly long response.
And across all of this, the thing still has to meet your desired size, PCIe power connection/consumption levels, and heat/noise concerns, all of which may mean being able to deal with a big hungry bastard or needing something that leans closer to small, quiet, and either single 6-pin or only slot powered.
YMMV widely of course, but more research is always valuable, and when it comes to the true budget extremes, it gets pretty interesting. Like, what is the best $25 or $50 used GPU you can find for gaming? You might be surprised! And this of course means seeing how some old used models play with modern OS and games, which makes research even trickier.
Facebook
Twitter