Nvidia Pascal Announcement

GP100 - Tesla P100 announced in April 2016..not Geforce Titan.
GP104 - GTX 1080 launched and available in May 2016..not June 2016.

Next up is GP102.

This roadmap is as much fantasy as it is throwing darts on a board with random names, specs and dates.

And yet it has been spot on for GP100, GP104 and GP106.

It is funny that some readers only focus on the product names (Tesla, Titan) and ignore the actual GPU names GP100, GP104, GP106, GP107.

As for the GP102 that was unknown as of January 2016.
 
There is definitely a supply issue for all cards (14nm and 16nm) right now, its can't be over looked as too much demand........

If you have too much demand supply issues happen, if you have low supply, you still have supply issues lol. Either way you are not selling as much as ya can.
 
My point were not about market share, it was about zombie ppl that buys anything that companies tell them to buy, and it happens not just for apple. The main problem in my opinion are ppl in the USA use its credit card or theirs parents and doesnt bother to think if what they are buying is actually a fair deal.


well those are two different points, and we have gone a bit OT lol.
 
Aaargh.. FE bullshit again..
It's pretty much like saying "hey, the price is actually $300 but don't worry because it may go down to $250 eventually in some months' time".

The RX 480 is going through the exact same thing. $230 announced for the 8GB and the cheapest I can find is over 300€ in local stores (and even amazon has raised their prices?).



GP100 - April 2016 - Check
Please share a link to this Pascal "Geforce Titan" that's been available since April and GTX 1070 cards with GDDR5X. And the GTX1080 was supposedly available in May, not June, etc..

You keep showing this diagram and we keep debunking it again and again. It's tiresome.
 
There is definitely a supply issue for all cards (14nm and 16nm) right now, its can't be over looked as too much demand........

If you have too much demand supply issues happen, if you have low supply, you still have supply issues lol. Either way you are not selling as much as ya can.
Yes. Some 480 weren't in stock since lunch day...They even aren't been listen in Amazon yet...
 
If you're sure that the GP106 die is only 192, then the only logical conclusion is that it is ball compatible with GP104. I don't see why that would be a problem? It will allow them to reuse the same PCB with a cut down GP104 and 256 bits for a 1060 Ti or something.

Well I am sure it is 192 bit but was just making a guess on the pin compatibility. I dont see the point in making a PCB for such a purpose though. It wastes a bit of space and also has to be designed for higher power delivery (Even a further cut down GP104 will consume ~140-150W). Dosen't make any sense when the 1060 will sell in far larger quantities than the 1060 Ti.
And what's the size of the substrate? If it's the same as GP104, that would almost settle the question about being ball compatible...

I actually checked this and it appears that you are right..the substrate is the same size. So yea..very well could be.
I'm still sort of confused... I can't see how it wouldn't be faster to just do this in "software". Even quantizing after every instruction should only be ~2-4x slower and even handling details like specials and denorms I'm sure you could do it faster than 1/64. Why even bother with the hardware at all?

TBH it doesn't really bother me that they don't have fast fp16 on desktop, but they should have been a lot more upfront about it when they launched consumer Pascal. Lots of folks in the games industry are still working under the incorrect assumption that fp16 is supported and faster on NVIDIA.

Well there's still no replacement for actual hardware testing is there. So 1/64 rate ensures that at minimal die area penalty.
And yet it has been spot on for GP100, GP104 and GP106.

It is funny that readers only focus on the product names (Tesla, Titan) and ignore the actual GPU names GP100, GP104, GP106, GP107.

As for the GP102 that was unknown as of January 2016.

I just showed you how it was wrong for GP104...

And by what metric has it been spot on for GP100? A Tesla P100 was merely announced in April..do you even know if its shipping yet?

Note that it also says 256 bit for the GP106. (There are more mistakes as well...)

So you think Nvidia designed a GP102 after January 2016?
 
With the persistent rumors about a further cut down GP104, I'm wondering if that was a mitigation strategy in case the 480 turned out to be as formidable as AMD hinted it would be.

Say a GP104 with 12 or 13 SMs and 192 bits. Or 15 SMs with 192 bits. Etc. Lots of possibilities.

It always strikes me how in how many ways Nvidia can sell cut down parts while AMD can not. Even the Nano has a full set of units, while non-Titan top dog GTX 980 Ti has some cuts. (The 1080 is an exception to that.)

It gives them the option to refresh with slightly improved version later on, and the yields just be fantastic.

And now with the 1060, it finds itself again in the comfortable position of having a faster product with a cheaper die and lower component cost. The FE edition is just the cherry on the cake.
 
I think it has to do with how much money Nvidia can put on the table compare to AMD. And a very important thing, if we take Rajas words AMD is struggling to finding talent in the industry willing to work with them, he said that it was the most difficult part.
 
Well that is business when they are behind, they have to do with what they have and when they have the rest of AMD holding them back, they really can't do much about it. They need to focus on what they can do best and focus on that. Internal compittion will help them strive for more in the short term, but getting talent from outside takes much more money and time. These are things RTG doesn't have. Internally they do have capable engineers.

But the main thing here is they need to stop making obvious mistakes, just by doing that it will improve their corporate image and branding regardless if their products aren't as performant as their competition if they are priced right. But by messing up things like the pci-e spec, fury x launch mistakes, all the AMD cpu screw ups, just makes them look like a bunch of buffoons trying to piece things together after the fact. And they aren't buffoons not from an engineering point of view, from a PR and marketing point of view that is where the problems are coming from.

If this was a different situation like if they were in Intel's or nV's position and they make a slip up on a rare occasion, that's ok (relatively), they can recover.....
 
I think it has to do with how much money Nvidia can put on the table compare to AMD. And a very important thing, if we take Rajas words AMD is struggling to finding talent in the industry willing to work with them, he said that it was the most difficult part.

Is there a link to the interview where that was stated? I either did not see that interview or missed that part of one I did.
 
Is there a link to the interview where that was stated? I either did not see that interview or missed that part of one I did.
Let me look through my YT history. he was under the effect of good Tennessee whiskey so it may have been secret info :runaway:. He said that his college told him that he was crazy for accepting going back to AMD.

Edit:

you will have to forgive me can't find the time mark but! its a very interesting interview so it is recommended to watch it. I will watch it again lol.
 
Last edited:
Not true. At least in GER and for the 1080, AIB models started at 680 (or 660?) EUR compared to the 789 EUR for the FE.
You're seriously saying that even the initial batches were priced lower than FE in Germany? Either you got majorly screwed by "higher-than-rest-of-the-world" with FE, or you were the only country with cheaper AIBs when they first arrived, US, UK, Finland, Sweden etc etc all had first AIBs priced cheaper than FE's at the time
 
Please share a link to this Pascal "Geforce Titan" that's been available since April and GTX 1070 cards with GDDR5X. And the GTX1080 was supposedly available in May, not June, etc..

You keep showing this diagram and we keep debunking it again and again. It's tiresome.

What is tiresome is those who see things through blinders as again you only mention Titan, GTX 1080/1070. Those are NOT GPUs but cards based on GPUs.

I guess tiresome for you means accurate as the chart has been proven right for the GP100, GP104 and GP106 dates.

Expect the GP107 in September.
 
Last edited:
I just showed you how it was wrong for GP104...

No you really did not.

And by what metric has it been spot on for GP100? A Tesla P100 was merely announced in April..do you even know if its shipping yet?

Well you can go and order a DGX-1 and find out.

http://www.nvidia.com/object/deep-learning-system.html

Note that it also says 256 bit for the GP106.

That may be true (256bit) as the GTX 1060 card shows two missing rams which if filled would result in 8GB on a 256bit interface.

So you think Nvidia designed a GP102 after January 2016?

That is not what I said. I said that whoever put the roadmap together in January might not have had information about the GP102.
 
Last edited:
Quick measurement of the GP106 die comes up to 202mm²

1fsk36.png


Non rounded numbers:

Width: 11.431578947368421052631578947368 mm
Height: 17.747368421052631578947368421053 mm
 
a faster product with a cheaper die

Is the die actually cheaper? There may be some 10-15% more candidates per wafer, but TSMC 16 seems a lot more booked than GF 14, I doubt the wafer price is the same (ofcourse with the WSA AMD's wafer cost is hard to define at all, also depending on zen and maybe neo production/schedule).

Expect the GP107 in September.

Can you this time tell us beforehand if you mean the pre-announcement, the announcement, the review date, the card availability or the laptop availability? :D
 
Is the die actually cheaper? There may be some 10-15% more candidates per wafer, but TSMC 16 seems a lot more booked than GF 14, I doubt the wafer price is the same (ofcourse with the WSA AMD's wafer cost is hard to define at all, also depending on zen and maybe neo production/schedule).
Nvidia has already listed in some SEC documents that they have some chips in production at Samsung. That should be enough of a motivation for TSMC to offer competitive wafer pricing.

Nvidia has much higher volumes overall, so that's a factor also.

And the smaller die makes for better yields. It's obviously possible that GF has better detect density than TSMC.

But even if the die price is the same, Nvidia still has the benefit of having less DRAM chips, at much higher volume. And a cheaper power solution.

In a price war, AMD would lose.
 
Is the die actually cheaper? There may be some 10-15% more candidates per wafer, but TSMC 16 seems a lot more booked than GF 14, I doubt the wafer price is the same (ofcourse with the WSA AMD's wafer cost is hard to define at all, also depending on zen and maybe neo production/schedule).

You bring up a very valid point..we dont know what the actual wafer costs are. And with the WSA, if the wafers were not going to be taken up by the CPU division anyway..then you could even say the wafers are "free"! But given that the GF..nee Samsung process is denser..I wouldn't put too much money that it is substantially cheaper, if at all.
Can you this time tell us beforehand if you mean the pre-announcement, the announcement, the review date, the card availability or the laptop availability? :D

You forgot to add GPU silicon availability :LOL:
 
Back
Top