TPU: nVidia readies 1660TI

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

EXCellR8

Diamond Member
Sep 1, 2010
3,979
839
136
Yea I'm not liking the sub 2000 naming scheme either but as another pointed out it certainly distinguishes them from the RTX enabled cards.

I did find this funny though...

983c0195b3369832459a078549cdb7af0ee9c8b45eb6042914ed7b6b51b03ae4.gif
 
  • Like
Reactions: Thrashard

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
Well it will be 18% faster than a gtx1060 for $20 less than a AIB gtx 1060 or $50 less than the 1060 FE edition.

If its $250 it will be 10% faster than a rx590 and cost $30 less.

Lets examine the performance tiers.....
gtx1080ti tier 1
1080 tier 2
1070ti tier 3
1070 tier 4
1060 6gb tier 5

2080ti tier 1
2080 tier 2
2070 tier 3
2060 tier 4
1160 tier 5 (faster than a gtx980ti.)

the mining boom warped all our minds. 1060 6GBs were $250 2 and a half years ago. 18% performance improvement at steady price in 2.5 years? garbage.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
No, the reasonable position is to assume all rumors are false.

I don't consider that a reasonable position. It might work if you bet on the veracity or outcome of all rumors, but you still wouldn't be correct 100% of the time.

Even you don't believe it's a reasonable position, because your very first post in this thread states:

This is definitely along the lines of expectation. In general the expectation was that the low end chips would not be RTX.

Shouldn't you have assumed that all of those rumors were false? That's being pedantic, but clearly there are some rumors or speculation that you assumed was more true than others.

That Link is mixing supposition with fact. All that Lisa Su actually confirms is that there are tapeouts. She does not say tapeouts are Navi: "We will be doing tape outs later this year and as we get closer to production will give more insights there. But the idea is to be more competitive throughout the portfolio."

It's not an unreasonable assumption for anyone who had looked at AMD's roadmaps that they'd given earlier:

AMD-GPU-Roadmap-Polaris-Vega-Navi-1480x833.jpg


Companies don't always spell everything out, but it's hardly an unreasonable assumption to make. Once again, we don't know the true answer, because AMD hasn't told us. Companies don't share ever detail about their operations, plans, etc.

Vega 7 was always planned as a consumer card. Lisu Su stated as much in a recent interview. It's again rumor and assumption that got it wrong, so now we get wild theories, when actual facts are out there.

Once again this is just after fact. Once you say that you've announced a product you can always claim that it was always the plan. There probably was some plan about using Vega 20 in a consumer card, but that's not guarantee that it wouldn't be canceled. When you look at the actions of AMD rather than their words, it doesn't seem like they're really behind Radeon VII as a product and that this was a plan of last resort. You can't reasonably claim that AMD had no other plans that might have been canceled, because this plan sucks too much for it to be a top plan.

Even JHH considers it utterly bizarre. In a recent interview that he gave, he said the following regarding Radeon VII:

It’s not even available yet. Our 2080 is all over the world already. RTX is offered by every AIC in the world. I think theirs is going to be available on their website? It’s weird. It’s a weird launch. Maybe they thought of it this morning.

If you look at everything we know about Radeon VII and consider the possibilities, you'd have to conclude that AMD is utterly incompetent as a company if this was what they had intended to do all along.

But this is getting off topic considering this a thread about a new NVidia card, not the Radeon VII. I think we can continue the discussion in the other thread instead.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
I don't consider that a reasonable position. It might work if you bet on the veracity or outcome of all rumors, but you still wouldn't be correct 100% of the time.

We are talking about rumors that have already failed to happen.

So the sequence is:

Rumor says "A":

"A" does not happen.

The most reasonable assumption at that point, is that the rumor stating "A" was incorrect. Simple, obvious, done.

But instead you stacked more assumption to continue believing that a rumor already shown incorrect, was somehow really going to be correct, but yet another rumored/assumed action (failure by the company) prevented the rumor from coming true.

It's like using two wrongs to make a right, instead of just admitting the first rumor/assumption was wrong.

Even JHH considers it utterly bizarre. In a recent interview that he gave, he said the following regarding Radeon VII:

JHH being a jerk isn't evidence of anything except JHH being a jerk. The reality is that AMD has been behind NVidia for years.

1080 vs Vega 64, looks very similar to 2080 vs Vega 7, so the gap hasn't really changed much.

AMD really needs a Ryzen Moment with a big architecture upgrade to catch NVidia.
 
Last edited:

linkgoron

Platinum Member
Mar 9, 2005
2,286
810
136
Well it will be 18% faster than a gtx1060 for $20 less than a AIB gtx 1060 or $50 less than the 1060 FE edition.

If its $250 it will be 10% faster than a rx590 and cost $30 less.

Obviously price/perf should be selected against one of the worst cards around. According to TPU, the RX 580 can be found for $185 and is essentially equivalent to the 1060. So, for $65 more (i.e. 35% more) the 1660 will only be 20% faster.

Lets examine the performance tiers.....
gtx1080ti tier 1
1080 tier 2
1070ti tier 3
1070 tier 4
1060 6gb tier 5

2080ti tier 1
2080 tier 2
2070 tier 3
2060 tier 4
1160 tier 5 (faster than a gtx980ti.)

At 4k even the stock 980ti is 20% faster than the 1060, so we'll need to see if the 1660 is actually faster than the three and a half years old 980ti. In reality, any decent after market 980ti that you could buy 3+ years ago was at least 15% faster at 1440p or 4k.
1548191005989.png
https://www.computerbase.de/2015-06/geforce-gtx-980-ti-custom-test-partnerkarten-vergleich/
https://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_G1_Gaming/30.html
https://www.techpowerup.com/reviews/Gigabyte/GTX_980_Ti_XtremeGaming/23.html
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
  • Like
Reactions: cbn

Ranulf

Platinum Member
Jul 18, 2001
2,331
1,139
136
$279 would definitely pressure AMD and possibly make them cut the price of the RX 590.

The 590 is already down to $259.99 on NE for a Saphhire model with 3 free games. But Nvidia lives in their own world on pricing and sadly can afford to. They sure haven't offered any real discounts on the 1060 cards in the past 2-3 months.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Yes $300 does seem logical in which case it still not a midrange card. Mid range is 250 so 2050Ti will be mid range.

On further thought, $300 seems to high, especially if you look at shaders / dollar. The 1660 Ti is rumored to have 1536 CUDA cores, and at $300, that's 5.12 per dollar. The 2060 with 1920 CUDA cores is just a bit under 5.5 per dollar.

I looked at some past generations of NVidia cards and typically as the price of the card increases, you see a downward trend of shaders / dollar. This isn't a perfect measure because it doesn't account for memory differences, and once you get towards the bottom of a product line, the cost of the PCB, etc. start to account for a greater part of the cost so you tend to see a drop-off there.

However, we're comparing two different cores, with those on the 2060 being more capable in terms of features. It doesn't make a lot of sense to sell a card that has a worse shader / dollar ratio that's also less feature rich. At $280 it has the same ratio as the 2060.

Again, it's not perfect because it's rumored to have the same amount of memory (so at a lower price, the memory eats away more of the margins), but on the other hand the die is going to be a lot smaller without ray-tracing tech.

Put another way, if this card is $300, why wouldn't you just scrape together another $50 to get a 2060 instead?
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
On further thought, $300 seems to high, especially if you look at shaders / dollar. The 1660 Ti is rumored to have 1536 CUDA cores, and at $300, that's 5.12 per dollar. The 2060 with 1920 CUDA cores is just a bit under 5.5 per dollar.

I looked at some past generations of NVidia cards and typically as the price of the card increases, you see a downward trend of shaders / dollar. This isn't a perfect measure because it doesn't account for memory differences, and once you get towards the bottom of a product line, the cost of the PCB, etc. start to account for a greater part of the cost so you tend to see a drop-off there.

However, we're comparing two different cores, with those on the 2060 being more capable in terms of features. It doesn't make a lot of sense to sell a card that has a worse shader / dollar ratio that's also less feature rich. At $280 it has the same ratio as the 2060.

Again, it's not perfect because it's rumored to have the same amount of memory (so at a lower price, the memory eats away more of the margins), but on the other hand the die is going to be a lot smaller without ray-tracing tech.

Put another way, if this card is $300, why wouldn't you just scrape together another $50 to get a 2060 instead?
Ok but what incentive is there for Nvidia to price it below 300? Amd haa 590 which would be slower so pricing based on current market situation and then lower it when situation changes.
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Ok but what incentive is there for Nvidia to price it below 300? Amd haa 590 which would be slower so pricing based on current market situation and then lower it when situation changes.

If no one thinks it has good value relative to the 2060, it won't sell at $300 even if AMD doesn't have anything of value near that price. Even in absence of an AMD product, NVidia still has to compete against its other offerings. Between this at $300 and the 2070 at ~$550, the 2060 is kind of a no brainer.

Also, the 8GB 580 is finally dropping below $200, so even though the 590 isn't a good buy AMD can still pick up sales just because the value / dollar is so good. Even though the performance of the 1660 Ti would put it in another performance tier, that 580 is still a decent value if you're using 1080p, or don't mind not running the highest settings at 1440p. Maybe a little less so now that NVidia cards will work with freesync.

I suppose the other thing to consider is what the salvage part (presumably just a non-Ti 1660) ends up looking like and what they try to sell that for as that helps to set the price as much as the 2060 above this. Supposedly there's also a TU117 that's supposed to be the lowest part. No idea when that comes out, but the rumor/leak put it ~15% less than a 1060.
 
  • Like
Reactions: VirtualLarry

beginner99

Diamond Member
Jun 2, 2009
5,208
1,580
136
Put another way, if this card is $300, why wouldn't you just scrape together another $50 to get a 2060 instead?

Some products simply exist to make others seem more attractive.

Like in grocery stores the luxury brand is never profitable alone. What makes it profitable is that instead of buying the budget brand people then choose the standard brand. Without luxury brand, most go for the budget version.

So how do you push people to adapt RTX? Make it look better than it is.
 

ZGR

Platinum Member
Oct 26, 2012
2,052
656
136
If this does sell at $300, I feel it is just a way for Nvidia to upsell someone to the 2060. Now I wonder, if this card is real how many will they have stockpiled vs the 2060?

The 2060 has so many different models it's overwhelming.
 
  • Like
Reactions: SteveGrabowski

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Put another way, if this card is $300, why wouldn't you just scrape together another $50 to get a 2060 instead?

Yes but that may be further incentive for NVidia to price it at $300; To convince more people to buy 2060 (and thus RTX cards).

If it performs around 1070 level for $300 it will still advance Perf/$ a tiny bit.

Also if the rumor is true, it is releasing with the 1660 (Non Ti), at the same time, which say does a bit better than the 1060 at it's old $250 price point.

Here is my speculation:

1660 3GB? $200 (not in rumor, but would fill in the stack).
1660: $250 ( Slightly above 1060 perf)
1660 Ti: $300 (~1070 perf)
2060 $350 (slightly above 1070Ti perf)

Or you can knock $20 off the 1660s and that works too, but I just think NVidia will go for a cash grab on release, with lots of room for future price cuts.
 

Guru

Senior member
May 5, 2017
830
361
106
They can't price this anything more than $250 because Nvidia has destroyed their own market for anything $300. The RTX 2060 for $350(and a free game) is way too good of an option(faster than a 1070ti) for someone to go GTX 1070 speeds with the 1660ti for $300. Might as well pay $50 more and get close to 1080 speeds in some cases.

Again, they've priced themselves out of the $300 range and NO ONE would buy this card at $300. Heck even the RTX 2060 is a bad upgrade from a GTX 1060 or RX 580, about 50% performance increase is not that great of a value, so the 1660ti is not going to be an upgrade path for 1060 and RX 580 owners, especially not at $300.

It might be a decent option for $220, for those looking to upgrade from Maxwell mid level like the GTX 950, 960, maybe even the 970 or AMD's previous gen like the R7 380, etc...
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Some products simply exist to make others seem more attractive.

Like in grocery stores the luxury brand is never profitable alone. What makes it profitable is that instead of buying the budget brand people then choose the standard brand. Without luxury brand, most go for the budget version.

So how do you push people to adapt RTX? Make it look better than it is.

That's true to a certain extent, but I don't think it works quite as good in this particular case.

If this were like the iPhone where Apple sells different tiers, I'd agree with this since the entry tier is just their to feel inferior enough to up-sell you to buy the model with more memory which costs you and extra $100 or more, but maybe only a tenth of that for Apple.

However, in this case the 1660 Ti is using a different die which means that they want to be able to sell this part at some point. Apple can just offer a bad value entry model that it doesn't have to sell since the rest of the phone is the same and can easily be sold with more flash memory.

NVidia can't turn a 1660 Ti that's priced to make people want to buy an RTX card instead into that card. Given that volume increases as you move down in cost, NVidia will want to make a lot of these because if you look at history, the $150 - $300 market segments sell a lot more than everything above that. If they make it too unattractive, they risk another glut of cards and having to halt production.

$300 could work if the 2060 were trending well above MSRP, but it appears that there's no problem getting one at that price. Sure there are some AIB offerings that are higher at $380 (or even some above $400) and that creates space for a $300 card, but I'm not sure how much NVidia wants to drive people towards the 2060.

Unlike Pascal where the 1060 was a full die GP106, the 2060 is actually the salvage part (cut down 2070) so unless yields aren't as good as expected and NVidia has a nice natural supply of chips that can't work as a 2070, pushing customers towards the 2060 just means that NVidia would need to artificially disable dies that could have been sold as a 2070 to increase the supply of 2060 parts, unless they're fine with prices increasing above MSRP.

Or you can knock $20 off the 1660s and that works too, but I just think NVidia will go for a cash grab on release, with lots of room for future price cuts.

That's certainly a possibility, but if it were me I'd rather have an almost guaranteed sale now at $280 instead of risking make people wait, potentially long enough for AMD to have an actual product that can compete above $250.

We don't know what Navi will be like, and it's a little hard to get on the GPU hype train for AMD after the last few years so I wouldn't assign a high degree of probability to AMD having a knockout product, but any competition is going to force NVidia to move more on pricing than the current lack of it that they're facing right now.
 
Last edited:

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
However, in this case the 1660 Ti is using a different die which means that they want to be able to sell this part at some point. Apple can just offer a bad value entry model that it doesn't have to sell since the rest of the phone is the same and can easily be sold with more flash memory.

It isn't that you want a situation where no one buys a 1660 Ti, but a situation where you up-sell a significant amount of would be 1660 Ti buyers (most would still buy 1660Ti). The up-sell has additional importance this time around. Growing an RTX install base to encourage more game developers into using the RTX feature set.

IMO, the ideal product stack is one where it makes every purchases consider moving up a tier. To get them thinking: "It's only x dollars for this much more performance/features..." Though $280 might still work that way for the 1660Ti->2060. $70 for improved general performance and the new feature set might make that case nearly as well. The problem NVidia has now is that RTX 2060 is so close to RTX 2070 in performance while pricing gulf is large, that it probably has the opposite effect. If I was thinking about a 2070, I would probably get a 2060 now...

That's certainly a possibility, but if it were me I'd rather have an almost guaranteed sale now at $280 instead of risking make people wait, potentially long enough for AMD to have an actual product that can compete above $250.

I was fairly split between $280 and $300, but the more I think about, I think you might have it right. Really in a normal cycle the 1660Ti should match the 1060 price of $250, and if this die lacks both RT and Tensor cores, it will be quite small, so there is only the more expensive VRAM to push up the price, and $280 (Ti with GDDR6) and $230 (GDDR5) will be more competitive in the market.
 
  • Like
Reactions: Ranulf

linkgoron

Platinum Member
Mar 9, 2005
2,286
810
136
Latest rumours are both the 1660 and 1660Ti use the same underlying chip, so neither are based on the 2070 chip. I suspect $300 if they kept the tensor cores for DLSS (which if anything will be more useful in lower end chips and are a good way of staying ahead of Navi), but a little less if they are missing.

If the 1660ti really is 80% 2060 (1536 CUDA cores, without RTX), it will have worse price/perf than the 2060 at $300 (and will also be a bit slower than the 1070), and the value proposition is already worse without RTX. I don't see how anyone would buy this for more than $275, and IMO even that is pushing it. Considering the core count and no RTX the 1660ti should also be significantly smaller than the 2060 (assuming it's a full chip).

Regarding performance, a 80% 2060 would will end up around 5% slower than the 1070, according to TPU. They'll need to increase clocks a bit to beat a 1070. A 1280 CUDA core 1660 (non ti), would end up around ~8% faster than the 1060, assuming 2060 clocks. Anything more than $200 for that kind of performance is really pushing it, as a $185 RX 580 already achieves 1060 equivalent performance .