NVIDIA Pascal Thread

[ShintaiDK]

im not acting nothing its their words not mine they didnt said "oh we threw 3 bill to the pascal arch" they said we threw 3bill into the "hpc"

So Pascal got zero gaming improvements. Because all R&D is spend on HPC. Got it.

 

[ShintaiDK]

If some people still don't understand that Nvidia didn't raise the prices to cover R&D cost, maybe this will help you understand what many people here said

                               Year Ending                       Year Ending
                                 Jan 2013                           Jan 2011
Revenue                        4.28B                               3.54B
Cost of Revenue             1.83B                               1.95B
Gross Operating Profit      2.45B                               1.60B

Year Ending 2011 - few months after Fermi GPU release, Year Ending 2013 - few months after Kepler release where Nvidia doubled or nearly doubled prices. Do you see what happened back in 2012? Their Gross Operating Profit skyrocketed 😀

You still forgot to lookup what gross profit is.

Did AMD milk consumers too? Not really?

 

[airfathaaaaa]

So Pascal got zero gaming improvements. Because all R&D is spend on HPC. Got it.

the hell are you talking about? did i mention anything about gaming perfomance i said that their ceo said that they threw 3 bill to the hpc department and that is only why you twist everything just to suit your needs? get real for once

 

[swilli89]

New estimates are saying 316mm^2 for GP104.

 

[railven]

Oh that's funny,what about all the threads moaning about console ports being shoddy and running at cinematic framerates.

What does that have to do with me?

<rest of post>

Seems like you are trying to paint people with a wide brush. I have no issue paying money for a product I deem I will use. And I have zero issues buying those quirky JRPG's that are finally get PC versions realsed because, I know I will enjoy them even if the one I'm currently playing is essentially a PS Vita port locked at 1080p and 60 FPS.

I'm doing this on a GTX 980 Ti, that downclocks to basically 800mhz and my i7 also howevers around 2600 or so.

I'm buying a product I feel suits my needs. Yet, that seems to bother some folk around here so far as to call me an idiot. Which is where I even got into this conversation.

Have fun!!😀

I always have fun 😉

 

[PPB]

I think what people are missing here is what perf/mm2 might nv be getting from uarch changes. They already had their perf/mm2 jump with maxwell and most people are arguing what big architectural changes will be introduced with pascal, if any. They also misjudge that amd's best perf/mm2 isnt fiji but rather hawaii and that is including a juicy big 512b gddr5 PHY.

Honestly I wouldnt dismiss an scenario where amd's 232mm2 die is 5% within of nvidia's 310mm die, factoring in 14LPP slight density advantage over 16FF+, that AMD this round is doing the maxwell and nvidia is doing the GCN type of changes, which have totally diferent impacts on perf/mm2.



Sent from my XT1040 using Tapatalk

 

[swilli89]

I think what people are missing here is what perf/mm2 might nv be getting from uarch changes. They already had their perf/mm2 jump with maxwell and most people are arguing what big architectural changes will be introduced with pascal, if any. They also misjudge that amd's best perf/mm2 isnt fiji but rather hawaii and that is including a juicy big 512b gddr5 PHY.

Honestly I wouldnt dismiss an scenario where amd's 232mm2 die is 5% within of nvidia's 310mm die, factoring in 14LPP slight density advantage over 16FF+, that AMD this round is doing the maxwell and nvidia is doing the GCN type of changes, which have totally diferent impacts on perf/mm2.

Throw in the fact Pascal still can not do Async compute and other console synergies and I believe you will be shown to be correct once both chips are out.

 

[sontin]

Sure, AMD will come within 5% when they yet need a bigger die with nearly twice the power.

Seriously guys, you need to stop this nonsense. P100 clocks around 1500Mhz with 15 billion transistors. While AMD used only a 850Mhz Polaris 10 for their first demo. nVidia will have a huge clock advantages over AMD. I guess much bigger than it was with Maxwell over Tonga/Hawaii/Fiji.

BTW: Hawaii needs the 512bit interface because it doesnt have any delta compression.

 

[Mopetar]

Seriously guys, you need to stop this nonsense. P100 clocks around 1500Mhz with 15 billion transistors. While AMD used only a 850Mhz Polaris 10 for their first demo. nVidia will have a huge clock advantages over AMD. I guess much bigger than it was with Maxwell over Tonga/Hawaii/Fiji.

While I agree that any comparisons at this point are either nonsense or purely hype driven, your post makes two serious mistakes.

The first is that we don't know what the max clock rate is for AMD (or if it even matters given the alleged SMT-like technology they might be using) only that they've shown it's at least 850 MHz. For all we know they could be sand-bagging.

The bigger mistake though is that clockspeed is meaningless when discussing different architectures. Apple's SoCs have almost always had lower clock speed than other SoCs, yet typically have better performance. Intel's NetBurst (Pentium 4) microarchitecture was also capable of higher clock speeds than their previous generation of chips as well as AMD's offerings, but they typically didn't perform as well even though they could clock significantly higher.

Clock speed increases in GP100 are only useful when comparing it to Maxwell, and even that only holds true as long as the architecture hasn't changed significantly. We can think of it as a lower-bound for performance gains, but it doesn't tell us how it will compare against Polaris without having more details about its performance.

 

[ShintaiDK]

the hell are you talking about? did i mention anything about gaming perfomance i said that their ceo said that they threw 3 bill to the hpc department and that is only why you twist everything just to suit your needs? get real for once

No, they spend 2-3B on Pascal.

 

[sontin]

The first is that we don't know what the max clock rate is for AMD (or if it even matters given the alleged SMT-like technology they might be using) only that they've shown it's at least 850 MHz. For all we know they could be sand-bagging.

Sure or they wont be able to clock higher with their design. Dont forget that AMD said that their Finfet process doesnt allow for much higher clocks over 28nm...

The bigger mistake though is that clockspeed is meaningless when discussing different architectures. Apple's SoCs have almost always had lower clock speed than other SoCs, yet typically have better performance. Intel's NetBurst (Pentium 4) microarchitecture was also capable of higher clock speeds than their previous generation of chips as well as AMD's offerings, but they typically didn't perform as well even though they could clock significantly higher.

Clock speed increases in GP100 are only useful when comparing it to Maxwell, and even that only holds true as long as the architecture hasn't changed significantly. We can think of it as a lower-bound for performance gains, but it doesn't tell us how it will compare against Polaris without having more details about its performance.

Clock rate matters with GPUs. When you can clock your GPU higher while using the same power you will certainly win most benchmarks. GPUs have a lot of "seriell" pipelines.

 

[airfathaaaaa]

No, they spend 2-3B on Pascal.

go see the interview

 

[ShintaiDK]

go see the interview

I saw the event. Spending 3B$ on Pascal HPC only would equal all Nvidia R&D for 2½ years.

 

[airfathaaaaa]

I saw the event. Spending 3B$ on Pascal HPC only would equal all Nvidia R&D for 2½ years.

no you didnt because they never said anything about this on the event it was an interview on a site D:

 

[Sweepr]

Chiphell is saying Geforce GTX 1080 comes with 8GB GDDR5X, 256-bit, 3x DisplayPort, 1x HDMI, 1x DVI-D.

Sure, AMD will come within 5% when they yet need a bigger die with nearly twice the power.

Seriously guys, you need to stop this nonsense. P100 clocks around 1500Mhz with 15 billion transistors. While AMD used only a 850Mhz Polaris 10 for their first demo. nVidia will have a huge clock advantages over AMD. I guess much bigger than it was with Maxwell over Tonga/Hawaii/Fiji.

BTW: Hawaii needs the 512bit interface because it doesnt have any delta compression.

Not to mention AMD itself describes the product as desktop mainstream, while leaks indicate NVIDIA is positioning GP104 as a more expensive GM200 replacement. Really doubt they would charge over $500 for the new VGAs if they weren't very confident about beating the competition.

 

[ShintaiDK]

no you didnt because they never said anything about this on the event it was an interview on a site D:

JHH told what the development of Pascal costed at the event when he announced GP100.

 

[jpiniero]

Clock rate matters with GPUs. When you can clock your GPU higher while using the same power you will certainly win most benchmarks. GPUs have a lot of "seriell" pipelines.

It sounds like Polaris 10's TDP is going to be 120ish, whereas the 1070 and 1080 will be similar to the 970 and 980 (145 and 165). It's not out of the question that a full Polaris 10 could have similar perf/w but is slower overall due to the lower clock speed.

 

[Timmah!]

So at the 317/333 mm2, what do you think are the chances of the core count to be more than 2560?

 

[Aristotelian]

Agreed, once you both also condemn the opposite. High-end hardware buyers belittling low-end purchasers. It happens quite often here.

I highly suggest that you re-read the post I quoted, which implied that one must be an idiot for wanting to spend USD 500 or more on ANY PC component. That is the post that ridicules others. My post, stating that people are free to spend what they want, does not ridicule anyone - it's a defense of people who want to pay for the best of the best.

You and other posters that came far after this post of yours who refer to e-peen, etc., are the ones attempting to shame those willing to pay more in order to prevent them from 'manipulating the market' which apparently makes the products you want to buy more expensive for you.

So the belittling here is only happening in one direction - especially when implying things are for the sake of epeen or that one must be an idiot for wanting to buy something. You don't like the price of something? Don't buy it.

And don't presume to ridicule others for their purchasing choices except in specific performance cases like:

X is twice as fast, more quiet, uses less power, than Y. But Y costs twice as much as X. Buying Y is silly because there's no discernible benefit in doing so, even if it is to support the team. But that's not what we are talking about here.

 

[SteveGrabowski]

Chiphell is saying Geforce GTX 1080 comes with 8GB GDDR5X, 256-bit, 3x DisplayPort, 1x HDMI, 1x DVI-D.



Not to mention AMD itself describes the product as desktop mainstream, while leaks indicate NVIDIA is positioning GP104 as a more expensive GM200 replacement. Really doubt they would charge over $500 for the new VGAs if they weren't very confident about beating the competition.

At least GDDR5X would make their quoted prices a little more palatable.

 

[ShintaiDK]

So at the 317/333 mm2, what do you think are the chances of the core count to be more than 2560?

Small.

 

[airfathaaaaa]

JHH told what the development of Pascal costed at the event when he announced GP100.

link

 

[sontin]

It sounds like Polaris 10's TDP is going to be 120ish, whereas the 1070 and 1080 will be similar to the 970 and 980 (145 and 165). It's not out of the question that a full Polaris 10 could have similar perf/w but is slower overall due to the lower clock speed.

It is out of question. A reference GTX980 uses 165W. A 390X - similar performance - is around >300W.

Dont forget that a GTX980 has two huge disadvantages: Less compute performance and less bandwidth. Both should be fixed with a >300mm^2 die over Polaris 10.

 

[Adored]

It is out of question. A reference GTX980 uses 165W. A 390X - similar performance - is around >300W.

Those are interesting numbers that I don't see repeated mostly anywhere.

Dont forget that a GTX980 has two huge disadvantages: Less compute performance and less bandwidth. Both should be fixed with a >300mm^2 die over Polaris 10.

Right and the main reasons for the 980's lower power draw is....?

 

[airfathaaaaa]

and the complete lack of any hardware scheduler design i might add