Question NVIDIA GeForce RTX 2060 REVIEWS, – Faster Than GTX 1070 Ti For $349 US

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Mopetar

Diamond Member
Jan 31, 2011
4,270
199
126
These are NOT scavenged dies. 2060 is the high volume part, full yields on mature process are typically 90+ percent.
The AnandTech review lists both the 2060 and the 2070 as TU106, so the 2060 would be a cutdown 2070. It wouldn't make sense for them to have yet another separate die as the 2080 (TU104) and 2080 Ti (TU102) are already separate dies with the 2080 Ti being a cutdown Titan.
 

Qwertilot

Golden Member
Nov 28, 2013
1,346
22
106
It is an x60 card. If it was x70 then it would perform similarly to the 1080Ti.
The 2070 is a direct equivalent to where the 9/1080 were - their full 'medium' die high end chip - and the 2060 is a mildly cut down chip sitting precisely where the 9/1070 stuff did. Price, performance gap, die size, time to launch everything.

It doesn't perform as per a 1080ti because between the mild die shrink and the switch to RTX they couldn't gain enough 'normal' performance to do that.

Complain about that if you want. Complaints based purely on the fact they've reorganised what their numbering system means - also seen with the move of the Titan to compute and the xx80ti into the price bracket - are simply a little bit silly.
 

PeterScott

Platinum Member
Jul 7, 2017
2,486
63
96
Actually this is the exact reason GPU makers exist, to make us happy. What did you think, that they exist to remind us of the real work we need to perform every day?!

Gaming GPUs are toys. Period.
Sorry, but no. Toys or not, companies don't exist to give you pricing that makes you happy.

Often "toys" are highly priced because they are optional luxuries.
 

PeterScott

Platinum Member
Jul 7, 2017
2,486
63
96
The AnandTech review lists both the 2060 and the 2070 as TU106, so the 2060 would be a cutdown 2070. It wouldn't make sense for them to have yet another separate die as the 2080 (TU104) and 2080 Ti (TU102) are already separate dies with the 2080 Ti being a cutdown Titan.
You are missing the point. Of course this is the same die cut down.

It is not scavenged, in the sense that they are not recovered dies, that were otherwise headed to trash if not for the 2060. Some trivial small number might be, but for the most part, 2060 dies are fully functional dies, disabled for product segmentation. This is not the same thing as "Scavenged dies".
 

Mopetar

Diamond Member
Jan 31, 2011
4,270
199
126
You are missing the point. Of course this is the same die cut down.

It is not scavenged, in the sense that they are not recovered dies that have bad units. Some trivial small number might be, but for the most part, 2060 dies are fully functional dies, disabled for product segmentation.
I'm not sure what you're arguing about. Binning has been a standard practice for years. People use terms interchangeable.

It's pretty likely that even with yields being good that there are defective dies that are perfectly serviceable as a 2060, but would not work as a 2070.

It makes no sense to sell a perfectly functional die as a 2060 for $350 when you can sell it for $600 with a 2070, especially when the product is new and the market isn't saturated yet.
 

Charlie22911

Senior member
Mar 19, 2005
519
26
116
Sorry, but no. Toys or not, companies don't exist to give you pricing that makes you happy.

Often "toys" are highly priced because they are optional luxuries.
If it walks like a duck...

Wikipedia said:
Price gouging is a term referring to when a seller spikes the prices of goods, services or commodities to a level much higher than is considered reasonable or fair, and is considered exploitative, potentially to an unethical extent. Usually this event occurs after a demand or supply shock.
https://en.wikipedia.org/wiki/Price_gouging
 
Aug 25, 2001
42,441
220
126
Now again if you are on a RX 580 or 1060 or 1070, etc... there is ZERO reason to upgrade, it's too expensive for the performance increase it brings to make it a good upgrade value from those cards, but if you are looking to upgrade from Maxwell cards like the 960 or 970 its a decent choice.

One thing is for sure if AMD doesn't lower the prices of its Vega cards, they are not going to sell any. Again the Vega 56 would need to come down to around $350 to counter the RTX 2060 and this would make the Vega 56 about 3-5% slower for the same price. AMD is in a peculiar position with this as they would need to lower Vega 56 to $350, possible even lower to be able to compete and the Vega 64 to $400 in order to be a consideration.
For sure.
 

Leadbox

Senior member
Oct 25, 2010
694
2
81
You are missing the point. Of course this is the same die cut down.

It is not scavenged, in the sense that they are not recovered dies, that were otherwise headed to trash if not for the 2060. Some trivial small number might be, but for the most part, 2060 dies are fully functional dies, disabled for product segmentation. This is not the same thing as "Scavenged dies".
You don't know that so please do not state it as fact.
 

Mopetar

Diamond Member
Jan 31, 2011
4,270
199
126
You don't know that so please do not state it as fact.
It's almost guaranteed not to be the case. Even with a reasonably mature process, you'll still have a few (say something good like .1 defects per cm^2) defects sprinkled throughout the wafer. TU106 is still a large die, so you only expect to get around 120 of them per wafer. Because you have such a large chip, even with a very low defect density, you an easily end up with a third or more of the dies having defects of some form. However, with GPUs there's a good probability that they aren't in a region that makes the chip completely useless.

The vast majority of those chips can be salvaged. You won't start artificially binning chips until the market is completely saturated and you've got parts that can't be sold any other way. If the yields were really good, you'd just lower the price on the 2070 to sell more full dies as a 2070 instead of needlessly disabling them to sell them for hundreds of dollars less.
 

PeterScott

Platinum Member
Jul 7, 2017
2,486
63
96
It's almost guaranteed not to be the case. Even with a reasonably mature process, you'll still have a few (say something good like .1 defects per cm^2) defects sprinkled throughout the wafer. TU106 is still a large die, so you only expect to get around 120 of them per wafer. Because you have such a large chip, even with a very low defect density, you an easily end up with a third or more of the dies having defects of some form. However, with GPUs there's a good probability that they aren't in a region that makes the chip completely useless.

The vast majority of those chips can be salvaged. You won't start artificially binning chips until the market is completely saturated and you've got parts that can't be sold any other way. If the yields were really good, you'd just lower the price on the 2070 to sell more full dies as a 2070 instead of needlessly disabling them to sell them for hundreds of dollars less.
Quite the opposite really. This is a mature process, which typically has 90%+ yield.

Which means you are left with ~10% of the parts to try a scavenge recovery parts from.

2060 will be the volume part. So the majority of dies are going to 2060.

Which means MOST of them will be fully functional dies simply disabled for market segmentation.

This is the case with almost every silicon segmented product. As that majority of dies go into low end parts, and the majority of dies are fully functional, so most of the low end parts are simply disabled for market segmentation nothing more.

There is almost never the case where the low end volume part is dominated by scavenged, recovered dies with failures. That is largely an message board myth.

It is product segmentation, nothing more. The scavenged recovery parts are an incidental bonus.
 

Leadbox

Senior member
Oct 25, 2010
694
2
81
Quite the opposite really. This is a mature process, which typically has 90%+ yield.

Which means you are left with ~10% of the parts to try a scavenge recovery parts from.

2060 will be the volume part. So the majority of dies are going to 2060.

Which means MOST of them will be fully functional dies simply disabled for market segmentation.

This is the case with almost every silicon segmented product. As that majority of dies go into low end parts, and the majority of dies are fully functional, so most of the low end parts are simply disabled for market segmentation nothing more.

There is almost never the case where the low end volume part is dominated by scavenged, recovered dies with failures. That is largely an message board myth.

It is product segmentation, nothing more. The scavenged recovery parts are an incidental bonus.
If there is any truth to your numbers, then the answer to your previous question is definitely YES. Yes nvidia should absorb the costs. Their margins might not be at 'record levels', but they'd be still pretty damn good.
Think about it, if they're castrating perfectly good 445mm^2 dies to make these 2060 cards, imagine the goodwill they'd gain and all the money they'd make if they sold them as ~$400 2070 cards instead.
 

PeterScott

Platinum Member
Jul 7, 2017
2,486
63
96
If there is any truth to your numbers, then the answer to your previous question is definitely YES. Yes nvidia should absorb the costs. Their margins might not be at 'record levels', but they'd be still pretty damn good.
Think about it, if they're castrating perfectly good 445mm^2 dies to make these 2060 cards, imagine the goodwill they'd gain and all the money they'd make if they sold them as ~$400 2070 cards instead.
Market segmentation exists for a reason. You still need the lower tier high volume parts, and you still want higher prices for you upper end parts.

The same applies to AMD, think of the good will they would get if they sold all the good Ryzen 2600s as 2700s at 2600 pricing, and the good good 570s as fully enabled 580s, but at 570 pricing.

You segment lower end parts because you NEED those parts. Selling them fully enabled for cheaper prices, means you don't have a range of parts and you get less money for your high end parts.

This kind of segmentation has been done since the beginning, and there are many cases where the parts were not disabled in a foolproof manner, where customers were able to re-enable the disabled cores/units and get the full part working proving that they were indeed fully functional parts only disabled for market segmentation.
 

Hans Gruber

Senior member
Dec 23, 2006
452
15
91
Wasn't the 1080 released in 2016? Please justify why the 2060 should be priced @ $350. When the 1060 was released it came out at $200 for 6GB. Then mining hit and the price was all over the place. The RX 480 was $199 originally. Prices in a flat market trend downward. No more mining craze and this will bring both AMD and Nvidia back down to reality.

It's like a little cult here. You folks do not see the big picture.
 

Hans Gruber

Senior member
Dec 23, 2006
452
15
91
Giant die and expensive GDDR6.
I am talking price/performance ratio. 3 year old performance for $350? Complete nonsense. The new Nvidia cards are duds in performance gains when you consider how long the 1080's have been on the market. Typically they bring out new video cards every 18 months.
 

jpiniero

Diamond Member
Oct 1, 2010
5,917
81
126
I am talking price/performance ratio. 3 year old performance for $350? Complete nonsense. The new Nvidia cards are duds in performance gains when you consider how long the 1080's have been on the market. Typically they bring out new video cards every 18 months.
Not disagreeing, but that's just the reality of the situation.
 

PeterScott

Platinum Member
Jul 7, 2017
2,486
63
96
I am talking price/performance ratio. 3 year old performance for $350? Complete nonsense. The new Nvidia cards are duds in performance gains when you consider how long the 1080's have been on the market. Typically they bring out new video cards every 18 months.
Much like how Vega 7 gives performance and pricing of 2 year old 1080Ti? The reality is that these cards are expensive to make and we have had a stalling out on GPU perf/$ from everyone.

It sucks for consumers, but it has to do with increased productions costs and unwillingness (should be no surprise) of manufacturers, to eat those costs.
 

Hans Gruber

Senior member
Dec 23, 2006
452
15
91
Much like how Vega 7 gives performance and pricing of 2 year old 1080Ti? The reality is that these cards are expensive to make and we have had a stalling out on GPU perf/$ from everyone.

It sucks for consumers, but it has to do with increased productions costs and unwillingness (should be no surprise) of manufacturers, to eat those costs.
So when cost gets out of hand what do you do? You lie, cheat and steal. Then you asked what any currency is backed by. Nothing.
 
Jun 8, 2003
13,984
42
126
Wikipedia lists both the 1060 and 1070 as TU106 with the exact same die size and number of transistors. Characterizing the 2060 is up to interpretation, but it is a die that has been cut down from the 2070, unless wikipedia is wrong.

https://en.wikipedia.org/wiki/GeForce_20_series
The 1070 was a 104 card the 1060 was a 106 I believe. Not the same die, the 1070 was a cut down 1080.
The 2060 and 2070 are from the same die.
I think the 1160 is made on the 116 die.
 
Jun 8, 2003
13,984
42
126
It is an x60 card. If it was x70 then it would perform similarly to the 1080Ti.
But only when you go from 28nm to 14nm not a simple refresh from 14nm to 12nm.
Remember we were supposed to have a 20nm card .Pascal was a 2 node jump, you shouldnt expect to see that kind of performance jump again.
 
Jun 8, 2003
13,984
42
126
Silicon cost/area have gone up each generation as well. So you can really only compare a couple of generations back and:


960 die: 220 mm2
1060 die: 200 mm2
2060 die: 445 mm2

2060 is clearly a MASSIVE increase and a far outlier. Margins would shrink significantly if NVidia ate the cost increase.

I know in a nice fantasy world where GPU makers exist to make you happy, that would happen, but the real world doesn't work that way.

The production costs have increased by a large amount, and it's being passed on to the consumers, as expected.
Yes and lets not forget the increase in performance.
The 1060 to 1070 performance difference was 37%. The 2060 to 2070 is 19%. That's why its more expensive.

Remember the gtx960 performance vs the 970 was like a 50% difference.
It seems they are moving thwe 1060 up the ladder to fill that big hole between the 1060 and 1070 type cards.
 
Aug 25, 2001
42,441
220
126
Geeze, I mean, maybe these prices would be justified, if the GPUs were indeed 7nm products, but they're not. It's just a simple partial-node-shrink, with a new arch., on a relatively mature process. It shouldn't cost that much to actually produce, the costs I assume are mostly paying back the R&D. And seeing as how the "pipeline" for RTX, DXRT, was written by MS (I assume?), then how much R&D did NVidia put in? Maybe all of that supercomputer time for the DLSS (which I regard as mostly voodoo computer "science"), cost the big bucks?

Or maybe, like most here think, this is just a big "money grab" by NVidia, because the writing is on the wall with their stock price decline... they are on the way out of relevancy.
 
Jun 8, 2003
13,984
42
126
Geeze, I mean, maybe these prices would be justified, if the GPUs were indeed 7nm products, but they're not. It's just a simple partial-node-shrink, with a new arch., on a relatively mature process. It shouldn't cost that much to actually produce, the costs I assume are mostly paying back the R&D. And seeing as how the "pipeline" for RTX, DXRT, was written by MS (I assume?), then how much R&D did NVidia put in? Maybe all of that supercomputer time for the DLSS (which I regard as mostly voodoo computer "science"), cost the big bucks?

Or maybe, like most here think, this is just a big "money grab" by NVidia, because the writing is on the wall with their stock price decline... they are on the way out of relevancy.
Normal video cards are less than 5 years these have been part of the R and D for 10 years.
 


ASK THE COMMUNITY