News GeForce RTX 3060 12GB Review: Hope Springs Eternal

[renz496]

Except the actual successor to the 1060 cards would be the 1660 models, not the 2060, making the gains for the last generation smaller. It only really makes sense to compare performance based on what's available at similar price points, not arbitrary model numbers. The GTX 970 was a $330 card, and the 1070 was a $380 card, so the $350 2060 and the $330 3060 are more successors to those. They don't keep the price points exactly the same between generations though, so to get meaningful results one would need to calculate the change in performance-per-dollar, not just performance.

Nvidia hasn't yet released a successor to the 960/1060/1660 this generation, something roughly around that $200-$250 price range with a TDP around 120 watts. It's also likely that the shift in model numbers that the higher-end cards saw last generation will work its way down the stack in place of something like the "16-series" they used before. So that card may be marketed as an RTX 3050 or 3050 Ti, despite not bearing much resemblance to the sub-$200, sub-100 watt cards Nvidia has traditionally associated with the "x50" model number. If I had to guess, a "3050 Ti" would probably perform a lot like a 2060, with 6GB of VRAM and an MSRP around $250-$280. Probably closer to the lower end of that price range considering there were already some $300 2060s last year, though it's hard to say in the current market, and I wouldn't expect the cards to be available for anywhere close to those prices for at least a number of months.

If there's a 3050 Ti, then there may also be a 3050 priced closer to $200, with performance more like that of the 1660 SUPER, only now with RTX functionality. I'm not sure how usable something like that would be for enabling raytraced lighting effects in games though, since the 2060 is already barely usable for that purpose.

2060 is the true successor to 1060 since it's performance is similar to 1080. (1060 is equal to 980). But thing is a bit complicated during 20 series era. First nvidia did not go with die shrink instead they go with enhance 16nm node (12nmFFN). And before the existence of 7nm node it was probably the most expensive node at TSMC. hence nvidia increasing their price structure with with 20 series. 16 series exist because AMD largely still competing with polaris based GPU in sub $300 market back then. But still we see some improvement on sub $300 segment because in 2016 GTX1070 performance will cost us $400. In first half of 2019 that level of performance finally entering sub $300 mark with $280 1660Ti. So what happen in 2018/2019 end up messing up product succession. Even before nvidia release their 30 series i was thinking a lot how Nvidia will going to position their 3060. But in any case 3060 by far is the worse x60 we had in a decade.

 

[Phaaze88]

Except the actual successor to the 1060 cards would be the 1660 models, not the 2060, making the gains for the last generation smaller. It only really makes sense to compare performance based on what's available at similar price points, not arbitrary model numbers. The GTX 970 was a $330 card, and the 1070 was a $380 card, so the $350 2060 and the $330 3060 are more successors to those. They don't keep the price points exactly the same between generations though, so to get meaningful results one would need to calculate the change in performance-per-dollar, not just performance.

Nvidia hasn't yet released a successor to the 960/1060/1660 this generation, something roughly around that $200-$250 price range with a TDP around 120 watts. It's also likely that the shift in model numbers that the higher-end cards saw last generation will work its way down the stack in place of something like the "16-series" they used before. So that card may be marketed as an RTX 3050 or 3050 Ti, despite not bearing much resemblance to the sub-$200, sub-100 watt cards Nvidia has traditionally associated with the "x50" model number. If I had to guess, a "3050 Ti" would probably perform a lot like a 2060, with 6GB of VRAM and an MSRP around $250-$280. Probably closer to the lower end of that price range considering there were already some $300 2060s last year, though it's hard to say in the current market, and I wouldn't expect the cards to be available for anywhere close to those prices for at least a number of months.

If there's a 3050 Ti, then there may also be a 3050 priced closer to $200, with performance more like that of the 1660 SUPER, only now with RTX functionality. I'm not sure how usable something like that would be for enabling raytraced lighting effects in games though, since the 2060 is already barely usable for that purpose.

FUDGE, I screwed up again - forgot the 1660...

I know the price points aren't exactly the same. The problem I have with that is the performance per currency value can swing wildly based on:
A)Which models people buy, be it a FE, or a FTW3.
B)How much they're willing to spend - if they buy from a 3rd party seller, official retailer, scalper, miner, etc.
C)Pricing between regions. The US tends to have the best prices for these things. The UK, India, or New Zealand? Not so much.
Trying to figure out the PPC after that would make my head explode. Performance alone is way easier to follow.

 

[cryoburner]

2060 is the true successor to 1060 since it's performance is similar to 1080. (1060 is equal to 980).

In terms of chip size, power draw and price, the 2060 doesn't bear much resemblance to a typical "x60" card. The chip it shares with the 2070 is actually larger than those used for the 970/980 and 1070/1080. And as previously pointed out, the 1660 cards fit that role more closely, and even they use a chip nearly as large as what the 1070 and 1080 had used in the previous generation...

GTX 960 = 227 mm2 - 120 watts - $200
GTX 1060 = 200 mm2 - 120 watts - $200/$250
GTX 1660/SUPER/Ti = 284 mm2 - 120/125 watts - $220/$230/$280

GTX 970 = 398 mm2 - 145 watts - $330
GTX 1070 = 314 mm2 - 150 watts - $380
GTX 2060 = 445 mm2 - 160 watts - $350

GTX 3060 = 276 mm2 - 170 watts - $330
GTX 3060 Ti = 392 mm2 - 200 watts - $400

Now, the 3060 uses a chip more comparable in size to the one used for the 1660 cards (and different from the much larger one the 3060 Ti shares with the 3070), though even the 1660 cards used an atypically large chip for an "x60" card, and there was a process shrink this generation. But the TDP actually increases to 170 watts to get the most out of that silicon, well above what is typically seen for an "x60" card, and near the level of even some "x80" cards from the past.

What ultimately determines a card's successor is what price bracket it is targeting though. Those in the market for a $200-$250 card are not likely to spend $350 just to follow some completely arbitrary model numbering scheme that the manufacturer changes on a whim. AMD similarly shifted around model numbers from RX 550/560/570/580 to RX 5500/5600/5700 last generation. The model numbers might follow a similar trend, but that doesn't mean the $400 RX 5700 XT is in any way a successor to the $170 RX 570. So the only logical way to compare cards from one generation to the next is to look at how much performance you get at a similar price level. The 1060 is targeting an entirely different price bracket from the 2060 and 3060, so it doesn't make much sense to compare performance gains to that card.

So, to go by a somewhat more logical comparison, using MSRPs along with techpowerup's reference 3060 12GB review numbers at 1440p (and the 2060 review numbers for the 970)...

970 -> 1070 = ~60% more performance @ 15% higher price = ~ 39% better performance per dollar
1070 -> 2060 = ~19% more performance @ ~8% lower price = ~29% better performance per dollar
2060 -> 3060 = ~21% more performance @ ~6% lower price = ~28% better performance per dollar
...and...
2060 -> 3060 Ti = ~58% more performance @ ~14% higher price = 38% better performance per dollar

So, the improvements in value for the money at launch of the generation are almost exactly the same between this generation and the last for the 3060 and the 2060. The 1070 offered a bit more of an improvement in terms of raw rasterized performance for the money, as did the 3060 Ti, though at somewhat higher prices (ignoring current availability issues). If we consider the time between each generation though, the 1070 came out around 21 months after the 970, the 2060 around 31 months after the 1070, and the 3060 Ti and 3060 around 23 and 25 months after the 2060. That means the 1070 provided an improvement in performance per dollar of about 22% per year over its predecessor, while the 2060 only improved that by around 11% per year, the 3060 by around 13% per year, and the 3060 Ti by around 20% per year. So actually, the 2060 provided the least improvement over time of these generations, with the 3060 being slightly better, but still well below the gains seen by the 1070 and 3060 Ti.

The 2060 did add raytracing acceleration, though it's questionable how useful it will be on hardware with its capabilities. Especially since it has relatively low VRAM compared to other recent cards in this price range, and at least CyberPunk has already proven to be too much for raytracing on the 2060's 6GB, seeing a disproportionately large performance hit and only getting around half the frame rates of the 8GB 2060 SUPER when enabled, when one would normally expect it to perform within about 15% of that card. The 3060 12GB again does better here, though I can't help but think that designing it for 8GB could have provided a better balance between price and performance (again, ignoring the current market). There are rumors of a 6GB 3060 as well, though it will also likely suffer from a VRAM deficiency when enabling RT in Cyberpunk and other demanding future games, though perhaps not quite as bad when installed in a PCIe 4.0 capable system. If they don't cut it down in other ways it might offer better value for games not utilizing RT though, and could possibly even slip down near the 1660 TI's launch price, allowing it to compare a bit more favorably with that card.

 

[Exploding PSU]

I found one in my local store, said "in stock". But it sells for 3x the MSRP, you seriously can buy 2x 2070s with that money.
I like EVGA models though, they always look cute

 

[renz496]

In terms of chip size, power draw and price, the 2060 doesn't bear much resemblance to a typical "x60" card

nvidia push the price structure up with 20 series. if nvidia retain kepler/maxwell/pascal price structure then RTX2060 will be $250 card. doesn't matter what kind of chip nvidia use for x60 because historically x60 does not always being limited to chip with dies size of 200-220mm2 nor 120w-130w power.

GTX460/GTX560Ti die size is 332mm2. power consumption is around 160-180w.
GTX260 even use GT200 chip which is have a massive die size at 576mm2. power consumption rated around 180w.

also with turing nvidia going back with full compute architecture. add the fact nvidia did not use die shrink but instead just an enhance 16nm so it really affect die size and power. remember there are reasons why die size are quite small for nvidia previous three generation. kepler still considered as compute architecture but nvidia adopt very drastic approach with kepler design to lower their power consumption (the issues they had with fermi). they ditch some hardware on the GPU to be replaced fully by software. and because of that back then nvidia able to pull something that some people thought it is impossible for nvidia to do against AMD, having much smaller die size and use comparable or less power for the same performance. maxwell and pascal are pure gaming architecture. that's why GA102 die size did not break 500mm2 despite being nvidia flagship.

 

[cgilley]

Nvidia's GeForce RTX 3060 12GB targets the mainstream market and promises reduced mining performance. Packing more VRAM than the 3060 Ti, it's a good card, but we suspect it won't hit its $329 MSRP for a long time.

GeForce RTX 3060 12GB Review: Hope Springs Eternal : Read more

Silly question time - if " we suspect it won't hit its $329 MSRP for a long time" why even bother to write an article on it?

 

[atomicWAR]

For the last few years upgrading every generation becoming less and less enticing. With 3060 you can see that nvidia purposely limit it's performance to 2070 only by disabling 2SM.

I agree with you whole heartily. I still think the performance figures for the RTX 3060 are a little too gimped. Nvidia really pushed the performance gains on their high end SKUs like the RTX 3080 compared to the RTX 2080....with a 69% performance increase. Whereas the RTX 3060 is only gaining 20% vs the RTX 2060. Truly I think gamers would have been much more excited about this card had they increased the performance to something closer to Tom's OC results as the base performance. The article even brings up these bigger drop offs in performance as you move down the product stack and how it is generally a bad thing. Nvidia hasn't always been this bad <cough> greedy <cough> about it in the past so it is disappointing to see them wade further and further in to these anti-consumer waters. Personally I think they should separate each SKU by the same 30-40% and call it a day.

 

[cryoburner]

nvidia push the price structure up with 20 series. if nvidia retain kepler/maxwell/pascal price structure then RTX2060 will be $250 card. doesn't matter what kind of chip nvidia use for x60 because historically x60 does not always being limited to chip with dies size of 200-220mm2 nor 120w-130w power.

If the 2060 were a 1060 successor, then what does that make the 1660 cards? They are in a similar price bracket to the 1060, have similar power requirements to recent "x60" cards, and the 1660 SUPER/Ti improved on the performance of the 1060 6GB by close to 40%. That's around 25% better performance-per-dollar when comparing the 1660 Ti against the 1060 6GB at their MSRPs, over 35% better when comparing the 1660 at its lower price, and nearly 50% better when comparing the 1660 SUPER, albeit around 8 months later. Those are relatively decent generational improvements (though not exactly great after 31+ months) while keeping within the same price range, so they are clearly intended as the successors to the 1060 cards.

Again, performance at a given price level is what matters most when making these comparisons. The model numbers themselves are just a means of marketing them that may or may not remain meaningful from one generation to the next. With the 20-series, Nvidia didn't raise prices so much as they shifted around model numbers to disguise less-than-stellar performance gains when comparing cards against their similarly-priced predecessors at some of the higher price levels. As it was, reviews tended to be lukewarm for the 20-series at launch, but the performance gains looked better for anyone making the mistake of comparing cards primarily based on similar model numbers rather than what you are actually getting for the money.

 

[renz496]

Silly question time - if " we suspect it won't hit its $329 MSRP for a long time" why even bother to write an article on it?

Because it is their bread and butter? Also some people only want to see the performance even if they are not buying it.

 

[Phaaze88]

GTX 760 -> 960 = +10%
GTX 960 -> 1060 = +72%
GTX 1060-> 1660 = +17%
GTX 1660-> 2060 = +35%
RTX 2060 ->3060 = +19%
[Left out the 1660 by mistake previously, also went back and used TPU's Relative Performance charts - it's far from perfect though.]
Well, that's the past 5 x60s...
Perf per currency? I'm staying out of that one - unless one narrows it down to only MSRP or something, which would be wrong, because not everyone pays those kinds of prices or even close to them at times.

Dang, Pascal was good: The 1060's pricing was fair back then, and at one point, a bunch of them became available through the 2nd hand market.

 

[cryoburner]

GTX 760 -> 960 = +10%
GTX 960 -> 1060 = +72%
GTX 1060-> 1660 = +17%
GTX 1660-> 2060 = +35%
RTX 2060 ->3060 = +19%
[Left out the 1660 by mistake previously, also went back and used TPU's Relative Performance charts - it's far from perfect though.]
Well, that's the past 5 x60s...

The 1660 and 2060 were both from the same Turing generation of cards though. In fact, the 1660 and 1660 Ti both came out shortly after the 2060, so the 2060 isn't in any way a successor to them. The 16-series cards are just the "lower-end" 20-series models that leave out the Tensor/RT cores. Since the 1660 and 1660 Ti were the cards roughly filling the 1060's place in the market, that leaves the 2060 to be filling the role of the 1070. The 2060's gains in rasterized performance over the 1070 weren't particularly large, but it did eventually add RT acceleration (in at least some capacity) and DLSS in some titles, along with a bit lower suggested price.

And while relative pricing might not always be completely consistent between regions, it's kind of necessary to consider pricing in some way. The 1060 6GB, for example, may have provided some good performance gains over the 960, but it also had a higher MSRP. And some generations offered more than one option at different performance levels, like the 1660 Ti, that cost more than a 1060 6GB, and the 1660 that cost less. Comparing the base 1660 against the 1060 6GB might look a bit underwhelming, though it's not as bad once the lower MSRP is factored in.

 

[spongiemaster]

Putting a damper on Ethereum wasn't enough - the card is still profitable, miners are still gonna buy it.

At current rates, it would take about a year to break even with a 3060 if you paid $900 for it, so I would argue it isn't profitable.