Nvidia Pascal GPU has 17 billion transistors

[ocre]

I think you all must be struggling here.

The Radeon Pro Duo is only a mere $1500 and barely sips energy at a miniscule 350 watts, you really need to keep that in mind.

Here we have nVidia violating us with a board that is barely any faster, uses a STAGGERING 180 watts and costs an OBSCENE $599- how can you people be so confused.......?



On a more serious note- I absolutely understand why on a technical, competitive and financial basis nVidia is using this strategy, but I'm really not a fan of getting a x80 part based on a x04 GPU. Completely understand all of the utterly valid reasoning behind it, and I also totally get with how shockingly inept AMD is there is no reason to do it differently(they absolutely *could*, but it would hurt the margins pretty badly).

Lol. That's funny.
I want to hit on a few things.
1st is the fact that AMD themselves found it important to come out and make clear that Polaris was a mid-range mainstream card. The points they are selling it on clearly indicate it's not gonna compete in the high end. Perf per watt increase- 2.5x compared to who knows which last gen card. But we can be sure that it won't be last gen nvidia cards, the 2.5x will be based on probably the 290x which I expect polaris 10 to perform like. Which is supported by AMD'S other selling point they have been promoting- expanding the available market with lower priced cards. They have pretty mum on performance.

2nd-
Fury pro 2 duo X or whatever they call it. I guess they had meetings to try to come up with the awesomest over the top name ever and somehow they come up with the that name by merging Radeon and core 2 duo. All jokes aside, the fact the card just launched...it is telling. This is amd trying to compete in the high end. Polaris simply will not.

3rd-
Cpu technology has run up to a wall and scaling over it has proved very difficult. As much as anyone wants to try to call nvidia greedy, there is also the real reality. We are not seeing the gp100 today because it just isn't possible. The technology is moving and advancing slower. It is taking more time and more money, more than ever before. So as these launches are staggered, it would be a big blow if cash flow dried up at the same time. You see, the gp104 will out perform the high end. Releasing it could be like cutting off their head, their high end made irrelevant. If the price of the 1080 was 350 bucks with gp100 so far out, it is a big issue. Not only are investors gonna be concerned, the cash flow is needed. I know you might be specifically talking about the 50$ increase up from the 980 but before I talk about that it is important to look at the way things have been moving these past few generations. Let's start with 28nm. The 680 launched at a time when the gk 100 could not be viable for the mass gaming market. We heard a lot about how nvidia had low yields but there was also the even bigger issue of wafer availability. This is always overlooked. It also contributed to the high launch price of the original titan. The intent was to price it high as it was restricted to be a low volume part. You would have to imagine the frustration if you were nvidia, to have designed and spent all this money for a monster chip but stuck with this reality that wafer limitations would keep it from mass market for years. But that was the reality. If you look back, you can find signs of that frustration in articles before kepler. The fact that TSMC would not be making large wafers anymore. That hit hard for nvidia, whatever happened at TSMC to abandon their larger wafers really put a damper on NVIDIA's large die strategy.

The combination of many factors led to the way things have unfolded. I believe this became an issue mainly because the naming scheme nvidia uses, see no one sees that AMD has actually followed a similar path. Tahiti was not all that large and turned out to be "mid range" for 550 bucks. Hawaii was the big die, the larger GCN with the bigger bus. AMD may have put off producing it but chip design to fab takes years. AMD knew nvidia was going with a large die flagship and knew that capacity was not there. They knew realisticly, nvidia could not produce many and they jumped on the opportunity by dropping Tahiti when they did and at the price they did. AMD thought for sure they would have the high end, all to themselves. But, they done this with their mid sized GCN chip. Hawaii was not possible at the time but was absolutely their planned big boy. Hawaii was the big GCN. AMD knew way before 28nm launch that nvidia was making a mammoth kepler chip for their high end and amd tried to out maneuver. It was a well planned strategy for amd, they had a beefed up mid sized die they thought would own the top end while nvidia fell flat engeenering a mammoth flagship that couldn't be produced in quantity for years. AMD planned their large die hawaii to come in for the 2nd punch, a 1-2 tko.

Looking at the specs of gk104 it sure didn't look like much compared to what amd was gonna drop with Tahiti. 256bit bus and the original clock speeds for the 670ti. Heck there were even performance indicators for the early gk104, we had benchmarks.

It's just that nvidia was able to extract way more out of the gk104. A part of me thinks that the whole 670ti leak was intentional. Nvidia knew about how big the Tahiti was and they knew they were in a tough position. But I believe they planted the 670ti leaks by giving these cards and bios to sources they knew were not reliable..for keeping secrets.
I don't think they were planning on those clocks at all, either that or they hesitated to launch. But....the clocks were so low for the 670ti. Nvidia was low balling it and I think it was on purpose.
I believe this intended leak set the stage for what played out. I believe this put the fire on amd and they rushed the 7970 out at the clock speeds that still gave them somewhat efficiency.

AMD did not think they had to worry. They believed they had a path, that the 7970 was gonna be the fastest chip for a long time. Hence the 550 price.

No matter if you can agree that nvidia planted the 670ti leak, clearly the 670ti benchmarks had an effect on AMD. It gave them confidence. I believe it played a role and ultimately led to nvidia edging out the 7970 while also being able to undercut the price. That was the only way to bring excitement to an otherwise unimpressive transition to 28nm.

The movement forward has been slow these past few generations. It is taking more and more money, more time and engineering to get these smaller gains. The fact that the low end is dropping out for dgpu and the node shrinks don't come with lower prices anymore....its all adding up to higher prices.

And lastly, I believe the gddr5x is contributing to the extra 50 bucks for the 1080. I think it will take some time before that memory is cheap and widely available