Gaming is NVIDIA's largest market by far & one of its fastest growing, if you think NV is taking its eye off the ball and giving AMD an opening I don't know what to tell you.
It's obvious that NVIDIA is building GPUs tailored for each application because it has the luxury to do so. They have the revenue in these major segments to justify doing an HPC/professional oriented GPU as well as a set of gaming-oriented versions which I'm sure we'll learn about closer to launch.
I can tell you that I sell way more Quadros than anything. Geforce is not Nvidia's primary focus. Not even close.
Your understanding of NVIDIA's business needs some work. Tesla is a very, very small portion of NV's overall revenues; GeForce GTX is far more important to the company in terms of total revenue and gross profit dollars.
AMD is hyping Polaris 10/11 because their current GPU sales are in the toilet and by hyping them up in the press, they would create an image of "being ahead" of NV (no doubt to help boost its stock price). NV, which actually makes a lot of $ from selling GPUs obviously doesn't want to signal to gamers that "hey, the stuff we're trying to sell you is crappy and obsolete, wait for the new stuff!"
Anyway, the point is that there were some really comical posts across the web claiming that NVIDIA hadn't taped out Pascal when in fact they had not only taped out a monster of a GPU using state-of-the-art process tech & packaging technology but have now gone into volume production.
Full GP100 with Zero DP units will have 5760SP.My first reaction when I saw Nvidia's blog post on Pascal was: are you kidding me? Only 3584 shaders from a 610mm^2 chip on a new node shrink? That's pathetic. Maxwell has 3072 shaders, so for a similarly sized chip, we should be seeing 6144. Instead we get a measly ~17% increase (okay, 25% if you count the full die, which apparently even Tesla buyers won't be getting at first).
After thinking it over, I realized that this is strong evidence that the "GP102" rumors are true - and that Nvidia's double-precision HPC chips will be completely separate from their graphics chips from now on. We saw the first steps in this direction with GK210, which was a completely new revision of Kepler's big die that never saw a consumer-grade release. As others have pointed out, the blog post on GP100 contains no mention of ROPs at all. It could just be an oversight, but I think they aren't mentioned because they are not included on the chip. GP100 isn't designed to support video output at all, it's purely a HPC computing chip.
The apparent reason why Nvidia is doing this is that their method of DP computing takes up a tremendous amount of die space. To get the 1/2 DP/SP ratio, they need to have one DP CUDA core for every two SP CUDA cores. Assuming the DP cores need more hardware because they're wider, that means as much as half the chip's shader power is for DP support only, and completely useless both to gaming and to ordinary single-precision GPGPU apps. If a chip of this same size was done Maxwell-style with minimal DP support, we could be looking at a whopping 7680 shaders! Now that's more like it. That said, though, I don't think GP102 will be quite this massive. A stronger possibility is 6144 shaders, double GM200's count. Whether Nvidia will use HBM2 on GP102 is another question, but given their traditional conservatism when it comes to new memory types, I think a 384-bit GDDR5X bus is more likely. GP102 would then be the highest actual GPU in the Pascal lineup, powering the most expensive Quadro card and the next-generation Titan when it arrives.
We might well see a situation where even GP104 has more single-precision power than GP100. That would be unprecedented, but today's announcement is a spanner in the works and is different from everything that Nvidia has done up until now. If they really plan on having the GP100 being the best in their lineup all-around, and everything else inferior to it, then AMD is going to win this generation without a doubt.
My first reaction when I saw Nvidia's blog post on Pascal was: are you kidding me? Only 3584 shaders from a 610mm^2 chip on a new node shrink? That's pathetic.
Each SMX have 32x DP units.Full GP100 have 60x SMX.32x60 is 1920DP units.3840+1920 is 5760SP
Full GP100 with Zero DP units will have 5760SP.
If that is true and NV making Gaming GP100 with zero DP units then GP104 should have 3500-3800SP
But Gp100 didnt any Rops.Atleast on GPU diagram.There are no Rops.It looks like 100% compute SKU.Despite NV's dominance in both HPC and Gaming, their profits aren't that big. Look at the numbers.
Do you think they will waste more money making a gaming focused GP100 in parallel? On what production capability? TSMC 16nm FF is tapped out between all the mobile demands.
What do they make with the few 16nm wafers available at TSMC?
GP100 for Tesla that sells for over $10,000 (just look at the prices of their custom Pascal rig).
Or
GP100 for Gaming that sells for $1,000? With HBM2 adding to the expense.
Let's just play a very simple thought game for those who don't understand this concept.
NV sells the chip for ~$500 to go into a $1000 GTX SKU. With 50% margin, the profit each chip is $250.
NV sells the same chip on a Tesla SKU for $9,500. They earn so much more profit for the same chip.
When production is limited and yields are low, it does not make any sense to rush to market a consumer GTX SKU for anything big.
There is no gaming-focused high FP32 GP100. It's illogical because it competes with the same 16nm wafers from TSMC. A big gaming chip is a recipe for failure unless the node is very good with excellent yields.
What they can do and what they are doing, is make a HPC big chip, that has insane profits so it can offset any yield issues, basically paying for the wafer and more. Then the harvested chips later will go to consumer GTX SKUs. Making the most of products and profit per wafer.
Do you think they will waste more money making a gaming focused GP100 in parallel? On what production capability? TSMC 16nm FF is tapped out between all the mobile demands.
What they can do and what they are doing, is make a HPC big chip, that has insane profits so it can offset any yield issues, basically paying for the wafer and more. Then the harvested chips later will go to consumer GTX SKUs. Making the most of products and profit per wafer.
But Gp100 didnt any Rops.Atleast on GPU diagram.There are no Rops.It looks like 100% compute SKU.
Source? iPhone sales have been tapering off, with Apple facing its first year-over-year decline in iPhone sales since the first model debuted. That doesn't indicate to me that TSMC is "tapped out". To the contrary, it seems likely they're going to need more GPU orders to fill up their capacity.
That makes all of your other arguments moot, since they rely on TSMC capacity constraints that we have no reason to believe exist at this point.
Even if GP100 does have ROPs (and nothing so far indicates that), it would not make a good gaming chip. Yes, it could do better than Titan X, but it will definitely lose to Vega. Again, we're talking about a meager 25% increase in shader count from GM200. Unless Nvidia wants to write off the high-end gaming market, they will need a GP102 chip that focuses on shader performance and sacrifices double precision. Quadro sales will make it worthwhile (look at the price on the Maxwell-based Quadro M6000). Most Quadro users don't care about DP support any more than gamers do. DP support is now a niche category that gets its own chip. And in retrospect it already was - the HPC market got the exclusive GK210, and the non-HPC market got GM200.
It is noticeable that every Streaming Multiprocessor thus has only about half as many shader units like Maxwell - with 64 sets in the same number as AMD's GCN architecture.
I can tell you that I sell way more Quadros than anything. Geforce is not Nvidia's primary focus. Not even close.
The fact remains that AMD has demonstrated working FinFET GPU silicon, and Nvidia has not.
And yet they couldn't seem to spare an actual GP100 unit for the presentation. Instead all we saw is a 3D render.
Not only Apple taps into TSMC's FF node. Come on. TSMC was last reported as saying 16nm FF is still only a small share of it's projected revenue for 2016. It's ramping up, definitely not what we could consider as a mature node.
Do you not believe Nvidia has working finfet gpus? With P100 in mass production and shipping in June in the DGX-1?
Do you not believe Nvidia has working finfet gpus? With P100 in mass production and shipping in June in the DGX-1? When did it become a thing that companies need to demo future products for them to be real? Someone needs to tell Intel, Apple etc. they need to get with the program and start demoing unreleased products.