Nvidia Pascal Announcement

They sure as hell weren't on finnish retailers, not even cheapest of the AIBs on launch. Maybe their own shop had them at right price, dunno.

Yeah some countries it has gone a bit strange, same with AMD 480 in say Sweden.
I think in the discussion it is tricky when looking at launch because there were no custom AIBs, nor for AMD 480 8GB.
I do not like the new sales model from Nvidia but I guess part of the reason is because they clock just as well as custom AIBs when adjusting power/temp targets and fan profile, and we are seeing less partners going for MSRP with their initial custom cards.
And this is causing pricing to be all over the place where same model can have a 5-10% difference depending upon the retail store, pricing is really not well controlled because of the price structure of the FE.
But there are models now that are competitively priced for Nvidia and in stock, the downside is lack of consistency between retailers and one can use statistic to show high prices or now some competitive models in stock, albeit overall still high pricing that IMO will come down.

Anyway I think this discussion has two stages of context, reference prices and later custom AIB prices from IHV.
The reason being the most popular cards are the custom AIB models whether that be AMD or Nvidia.
Cheers
 
That's just nonsense, NVIDIA isn't putting "3GB 1060 as 1050", 3GB 1060 is just out of the picture because it would look bad against 4GB RX 480 even if it performed better. GTX 1050 is completely different (read: cut part, 3GB 1060 plans or lack of them has nothing to do with it)

Yep..wouldn't be easy to market 3 GB GTX1060 against a 4 GB Rx480, even if the performance was better. And it would be bad for NV in the long run as the card would age poorly and get bad reviews in newer, more demanding games. Also the price difference between the 3 GB 1050 and 1060 would likely have been too close so maybe they decided to avoid it for greater segmentation. Either ways I'm glad they dropped it. My speculation for GTX 1050 is 2 SMs disabled so 1024 cores. I do wonder whether we might just see it with 4 GB on a 128 bit bus though. (It'll basically be a higher clocked GTX 960)
http://www.purepc.pl/notebooki/geforce_gtx_1060_mobile_pelny_uklad_gp106_trafi_do_laptopow
"Even smaller the difference is with the active GPU Boost 3.0. In the version for gaming notebooks, timing increases since the 1671 MHz, while the desktop variant soars in this mode to 1,709 MHz. We also conducted a test in the popular 3DMark Firestrike where notebook equipped with Intel Core i7-6700K, GeForce 1060 GTX and 16 GB of RAM received close to 10,300 points. The result is almost identical compared to laptops with GeForce GTX 980 on board. Yes, I know, benchmarks do not always reflect the actual gaming performance, but I hope that it will be equally high.
....
As for the temperature or the general mobility equipment equipped with a GeForce GTX 1060, here nothing else I can not tell;) Maybe, except that the person waiting for the new notebooks will be positively surprised"

The high clocks are not much of a surprise surprising considering the Maxwell mobile clocks. In some cases the mobile parts were even clocked higher than the desktop parts (860M and 960M). GP107/108 and Polaris 11 will be the biggest sellers though..that's gonna be a good fight.

Launch of the Pascal mobile parts should be close..With Maxwell they launched the 980m and 970m barely a month after the desktop parts.
The bad news is that there won't be a cheaper version so consumers will lost a opportunity to get better prices is the new and old hardware.

The 3/4 GB GTX 1050 will definitely be cheaper so I dont see what consumers lose here.
 
The 3/4 GB GTX 1050 will definitely be cheaper so I dont see what consumers lose here.
yes it will, with lower performance...ergo there wont be a drop in the price for used/new cards above it. We as customer should ask for the highest performance on the lowest price.
 
Yep..wouldn't be easy to market 3 GB GTX1060 against a 4 GB Rx480, even if the performance was better.
Slap a low price on it, and it will do wonders.

And it would be bad for NV in the long run as the card would age poorly and get bad reviews in newer, more demanding games. Also the price difference between the 3 GB 1050 and 1060 would likely have been too close so maybe they decided to avoid it for greater segmentation. Either ways I'm glad they dropped it.
I don't see it. Just like the 480 is universally reviewed with 8GB, the 1060 will be universally reviewed with 6GB. The 480 4GB results are an afterthought. Nvidia could have done just the same with a 1060 3GB.

A 1060 with 3GB will still be miles ahead of, say, a GTX 750 Ti or a 950, both with 2GB, for which there exists a market. It's a very easy case to make with a sufficiently low price.
 
yes it will, with lower performance...ergo there wont be a drop in the price for used/new cards above it. We as customer should ask for the highest performance on the lowest price.

We as a consumer will always ask for that. Heck I'm sure we'd all love GTX 1080s for $199. However market forces will determine what prices we pay.
Slap a low price on it, and it will do wonders.

How low though? Given that the 6GB would be priced at $249..there's no way they would have priced it lower than $199 (and even that seems too low to me for just 3 GB less RAM. $219 seems more likely), and in that range..a 4 GB Rx480 is more attractive to the general consumer. Even for the 6 GB 1060..an 8 GB RX480 for $10 will certainly make a case for some people.
I don't see it. Just like the 480 is universally reviewed with 8GB, the 1060 will be universally reviewed with 6GB. The 480 4GB results are an afterthought. Nvidia could have done just the same with a 1060 3GB.

A 1060 with 3GB will still be miles ahead of, say, a GTX 750 Ti or a 950, both with 2GB, for which there exists a market. It's a very easy case to make with a sufficiently low price.

The 1060 isnt aimed at that market though..it is a GTX 960 successor. Its the GTX 1050 which will compete in the 750/950 segment..and will be priced accordingly.
 
We as a consumer will always ask for that. Heck I'm sure we'd all love GTX 1080s for $199. However market forces will determine what prices we pay.

What determine the prices in last place is what people are willing to pay for that product. Intel wouldn't be able to sell a 4790k for 350 if people weren't buying then in first place. The 250 dollars of the 1060 is just Nvidia saying that they can sell a mid-rage GPU for 250 dollars and people will still be happy to buy it. For reference I bough my Asus DirectCU 7850 1GB for 120 dollars. even if the 1060 direct comparative is the 7870 there is still 130 dollars difference in price, or more than 2 times what I pay.
 
The 250 dollars of the 1060 is just Nvidia saying that they can sell a mid-rage GPU for 250 dollars and people will still be happy to buy it.
You are assuming that Nvidia even *could* sell it cheaper if they wanted to. There is also the possibility that Nvidia can't price the 1060 as an entry class GPU, because it was miscalculated in some way. We don't have confirmation that Nvidia actually managed to drive the production cost down for a given performance class. And the 1060 certainly is an upgrade over the former 960 in terms of performance.
 
You are assuming that Nvidia even *could* sell it cheaper if they wanted to. There is also the possibility that Nvidia can't price the 1060 as an entry class GPU, because it was miscalculated in some way. We don't have confirmation that Nvidia actually managed to drive the production cost down for a given performance class. And the 1060 certainly is an upgrade over the former 960 in terms of performance.
In that regard AMD chip is bigger and the PCB is way more expensive.
 
In that regard AMD chip is bigger and the PCB is way more expensive.
And die size alone tells you what? How many chips fit on a single wafer, and to a limited extent also how many dies could fit at most within the rectangular limit. But not how many masks had to be applied successively or what the yields are like. That easily multiplies your costs by a factor which outweighs the cost for an additional wafer. Just because a chip is larger, doesn't mean it's necessarily more expensive to produce.

And a "way more expensive PCB"? Seriously? Even had a look at what the assembly parts cost? Hint: The PCB itself costs probably more than the assembly parts on it - leaving RAM and the GPU itself aside. That's no more than a $10 difference for the assembly parts.
 
What determine the prices in last place is what people are willing to pay for that product. Intel wouldn't be able to sell a 4790k for 350 if people weren't buying then in first place. The 250 dollars of the 1060 is just Nvidia saying that they can sell a mid-rage GPU for 250 dollars and people will still be happy to buy it. For reference I bough my Asus DirectCU 7850 1GB for 120 dollars. even if the 1060 direct comparative is the 7870 there is still 130 dollars difference in price, or more than 2 times what I pay.

Which is exactly what I said..market forces will determine the price. I cant make any sense of the rest of your post or what you're trying to say. Either ways..this continued discussion of pricing is getting tiring and redundant.

Back to actual Pascal related stuff..I'm beginning to think a 4 GB 128 bit GTX 1050 is quite possible. Like I said earlier..it would basically basically a higher clocked GTX 960, which had a 128 bit bus with 7 Gbps GDDR5. The clocks are higher by about ~40% (1.66 Ghz vs 1.18 Ghz). With 8 Gbps memory and approx 20% better compression on Pascal..the increase in memory bandwidth is ~37%, so it just about matches
 
How low though? Given that the 6GB would be priced at $249..there's no way they would have priced it lower than $199 (and even that seems too low to me for just 3 GB less RAM.
$200 should be ok, given the Nvidia markup that people are willing to pay.

The 1060 isnt aimed at that market though..it is a GTX 960 successor.
The 960 was introduced 18 months ago without a 4GB option. If that was possible then, 3GB should have been possible to today. Anyway, since it has been canceled or renamed (if it existed at all), it's water under the bridge...
 
And die size alone tells you what? How many chips fit on a single wafer, and to a limited extent also how many dies could fit at most within the rectangular limit. But not how many masks had to be applied successively or what the yields are like. That easily multiplies your costs by a factor which outweighs the cost for an additional wafer. Just because a chip is larger, doesn't mean it's necessarily more expensive to produce.
There's no reason to believe that AMD chips require fewer production steps than NV chips, and certainly no actual evidence - why would there be? You're making a strawman argument, please stop with that.

And a "way more expensive PCB"? Seriously? Even had a look at what the assembly parts cost? Hint: The PCB itself costs probably more than the assembly parts on it
Citation Needed, I think...

Anyway, $10 multiplied by millions of product sold = big money. So you seem to be missing the point somewhat. Successful cost-cutting is an important part of how you make the difference between being financially succesful versus the opposite.
 
And die size alone tells you what? How many chips fit on a single wafer, and to a limited extent also how many dies could fit at most within the rectangular limit. But not how many masks had to be applied successively or what the yields are like. That easily multiplies your costs by a factor which outweighs the cost for an additional wafer. Just because a chip is larger, doesn't mean it's necessarily more expensive to produce.

And a "way more expensive PCB"? Seriously? Even had a look at what the assembly parts cost? Hint: The PCB itself costs probably more than the assembly parts on it - leaving RAM and the GPU itself aside. That's no more than a $10 difference for the assembly parts.
Just look at all the PCBs tear-down to see AMDs board has not only more but better components on it. And the PCB itself is more expensive since its 256bits vs 192...
 
Just because a chip is larger, doesn't mean it's necessarily more expensive to produce.
I agree that there's a remote possibility that a 15% smaller die on a less dense process with higher volumes will be more expensive... Hope springs eternal!

And a "way more expensive PCB"? Seriously?
On one hand a bunch of people are calling Nvidia for cheaping out on their FE PCB, and praising AMD for using lots of high quality VRMs (out of necessity) but on the other, cheaping out doesn't actually mean its cheaper. Did I get that right?

The non-GPU BOM of the 1060 is cheaper in terms of power regulators and DRAM. What more needs to be said?

Even had a look at what the assembly parts cost? Hint: The PCB itself costs probably more than the assembly parts on it - leaving RAM and the GPU itself aside.
Leaving out the DRAM is very cute indeed.

If AMD or Nvidia are paying more than $3 for the PCB itself, they're being robbed. The components are much higher than that.
 
And the PCB itself is more expensive since its 256bits vs 192...

The GTX 1060's PCB has 2 empty RAM areas, so it might be made for 256bit GPUs (perhaps a cut-down GP104 GTX 1065 down the road).
 
The cost of a PCB depends on the area of the PCB and the number of special features (layers, hidden vias, minimum feature size, PCB material.)

The number of DRAM sites has zero impact.
And now you're suggesting those DRAM zones are not connected to anything?

:rolleyes:
 
Back
Top