AdoredTV attempts to justify his position on Pascal

[TheoneandonlyMrK]

Always seamed like a stepping revision on a new node to me too, I think people forget Nvidia is mostly a software company,and its their software and drivers that by and large get and keep them in the game hence their largely software and driver async feature.
However you have to be fair about this the node swap alone would likely require major changes to get it right and they clearly did make small refined changes here or there, but their shader arrays weren't broken soo why rock an already risky boat ,all in just makes sense to me.

Also fin fets are an unknown to nvidia ,so the most scientific way to go forward would have been to make the same thing and see how it does then refine your next design to best suit the process.
Sorry but the phone app needs burning and phones need banning from "you" (tpu corrected) nowt wrong with it but the rage saved might be worth it

 

[Frick]

I refuse to watch videos, but wasnt that kinda obvious from the petformance? That both Pascal and Polaris are more die shrinks than anything else?

 

[fourletterfame]

sooo then the 1070 is a overclocked 970 ? i see your point but ... a 1070 that has Titan X performances in general and GDDR5 all the same as the 970 (sans the 3.5+0.5 and with 1gbps more) even on a refined 16nm FinFet node would not give that even with a substantial OC ...

i fail to see Pascal as a "die shrink" of Maxwell nonetheless ... as even a Skylake to Haswell can have the same performances if you downclock it ... i can't recall (nor i want to review the vids right now) did you test both at exact same clock and get that result or did you just lowered the 1080 clocks until it reached the 980Ti level (btw that's wrong to compare the 980Ti to the 1080 as the predecessor is the 980 even if they have better performances, well the 1070 is ... "on par or slightly above" the 980Ti nonetheless )

The notion that pascal is just a die shrink seems to be exactly the conclusion all of the data is suggesting. No one is saying that it's a bad thing, on the contrary adored in his video said it made perfect sense for nvidia to do.

The video is meant to be informative, thought-provoking and entertaining. People inserting their personal fanbias into it is what causes a large portion of the snarky tension you see in every other post in this thread. Anyone who wants to argue against the conclusions of the video are welcomed to make their own, citing their own sources and drawing their own conclusions, shouting to the heavens that you don't agree "because reasons" is intellectually dishonest.

 

[newtekie1]

Why do I get the feeling that you'd be one of the first people claiming that AMD didn't do anything except rebrand for 4 years, even though they added more tessellation every GCN revision, better memory compression and finally HBM?

No, I'm 100% aware of what GPU's AMD and nVidia have done nothing but rebrands with.

Downclocking the GPUs reduces their requirement for more bandwidth, that one shouldn't be hard to figure out. It's why weaker GPUs have less bandwidth to start with, they don't need it.

But it doesn't eliminate the memory bandwidth difference affecting performance. Just like the other differences you just say to ignore because you claim they don't make any difference. Um...no, ROPs actually do make a difference, so does memory bandwidth. I don't care what you say.

 

[Frag_Maniac]

Kudos for doing such tests, and stepping in to clarify things Adored.

That said, although it may sound stuffy and unappreciative to many, it still comes down to performance for price, and there the 1000 series is doing every bit as well as the 900 series. All AMD have in response so far is a mid ranged card that isn't nearly as powerful in the RX 480.

So as accurately stated, it will come down to whether Dx12 actually ends up making much of a difference to most gamers, and what cards do best with it. Even if AMD have an upswing as some are predicting, Nvidia are far enough ahead of them on cash flow to respond as needed.

So at the end of the day, I don't think this revelation is going to mean much to most gamers.

 

[Adored]

It wasn't meant to be a revelation. I just do this stuff for a love of the tech. Nvidia are very far ahead without question.

 

[FordGT90Concept]

Volta is going to prefer DX12/Vulkan over DX11/OGL. Pascal has heavy bias in favor of the latter.

There's actually a lot of changes in Pascal over Maxwell:
-More compute units with fewer stream processors on each. This makes it similar to GCN which boosts performance in VR.
-Dynamic scheduling that reduces idle time.
-Added a dedicated compute queue.

Only Witcher 3, with heavy PhysX testing, would benefit from the dynamic scheduling. The other games didn't use VR nor a technology that can utilize compute queues so expected performance should slow, not increase. Of course that's more than made up for by Pascal's ridiculous clockspeeds.

Pascal is definitely an incremental update to Maxwell and not an entirely new architecture. Volta might be a new architecture.

 

[GreiverBlade]

The notion that pascal is just a die shrink seems to be exactly the conclusion all of the data is suggesting. No one is saying that it's a bad thing, on the contrary adored in his video said it made perfect sense for nvidia to do.

The video is meant to be informative, thought-provoking and entertaining. People inserting their personal fanbias into it is what causes a large portion of the snarky tension you see in every other post in this thread. Anyone who wants to argue against the conclusions of the video are welcomed to make their own, citing their own sources and drawing their own conclusions, shouting to the heavens that you don't agree "because reasons" is intellectually dishonest.

well for one, i have no fanbias ... and the 1070 is not a die shrinked overclocked 970 (the 980 to 1080 maybe ... altho it compare to the 980Ti ... this is where it's wrong to me )

again i ask ... does the "oh lookies the 1080 has same performance as the 980Ti once i lower the clock" have been tested at "exact" same clock or not? because again the 1080 is not an evolution of the 980Ti but of the 980

because to me (970 to 1070 in example ) what i see here is not a die shrink

the clock difference is not that huge and yet the performance gap is ... it take more than a die shrink to do that imho.

1080 to 980 (and ... 980Ti ... )

altho those 3 pics show the 1080 is a "in-between" the 980 and 980Ti in terme of transistor count/shader/ROP/TMU(yet it still outperform the later, just as the 1070 does the same well rather "equal" it) similarity between 2 GPU is 1080 to 980 ... so if you OC a 980, can you reach 980Ti performances? not sure about that, tho if you lower the clock of a 1080 to a 980 level, which one would be the fastest? probably the 1080, the only conclusion to that is: a 1080 is not an upgrade for a 980Ti but for a 980 yes (well ... i upgraded forcibly to a 1070 from a 980 because of pricing ... and because of a dying GPU )


the shader arrangement might be the same but does that mean that the GPU is "exactly" the same?

you can bring down a i7-6950X to a Core 2 Duo E8500 level if you want

Volta is going to prefer DX12/Vulkan over DX11/OGL. Pascal has heavy bias in favor of the latter.

There's actually a lot of changes in Pascal over Maxwell:
-More compute units with fewer stream processors on each. This makes it similar to GCN which boosts performance in VR.
-Dynamic scheduling that reduces idle time.
-Added a dedicated compute queue.

Only Witcher 3, with heavy PhysX testing, would benefit from the dynamic scheduling. The other games didn't use VR nor a technology that can utilize compute queues so expected performance should slow, not increase. Of course that's more than made up for by Pascal's ridiculous clockspeeds.

actually ... that, indeed.

 

[rtwjunkie]

well for one, i have no fanbias ... and the 1070 is not a die shrinked overclocked 970 (the 980 to 1080 maybe ... altho it compare to the 980Ti ... this is where it's wrong to me )

again i ask ... does the "oh lookies the 1080 has same performance as the 980Ti once i lower the clock" have been tested at "exact" same clock or not? because again the 1080 is not an evolution of the 980Ti but of the 980

because to me (970 to 1070 in example ) what i see here is not a die shrink
View attachment 77361
the clock difference is not that huge and yet the performance gap is ... it take more than a die shrink to do that imho.

1080 to 980 (and ... 980Ti ... )
View attachment 77364 View attachment 77362 View attachment 77363
altho those 3 pics show the 1080 is a "in-between" the 980 and 980Ti in terme of transistor count/shader/ROP/TMU(yet it still outperform the later, just as the 1070 does the same well rather "equal" it)


the shader arrangement might be the same but does that mean that the GPU is "exactly" the same?

you can bring down a i7-6950X to a Core 2 Duo E8500 level if you want



actually ... that, indeed.

anyway ... off, i am.

have fun.

You seem particularly upset by a site reviewer who took alot of time to research this and put forth his own assessment.

And look, he actually took the time to come and explain it further, knowing some people were in complete disagreement. I say it's ballsy, and commend him for doing it, whether I agree or not.

 

[GreiverBlade]

You seem particularly upset by a site reviewer who took alot of time to research this and put forth his own assessment.

And look, he actually took the time to come and explain it further, knowing some people were in complete disagreement. I say it's ballsy, and commend him for doing it, whether I agree or not.

nope i am not upset... if i would be ... i would not take time to do that come'on, you know me better

all in all the video and idea behind is interesting, nonetheless.

p.s.: my 1070 still beat your 980Ti buddy (i jest i jest )

i realized something ...

if you clock them at the same speed ... a card that has less transistor/core/ROP/TMU perform same as one that has more... and that only because of a "die shrink"? well that's a feat then if it's the same architecture with just a die shrink, it's impossible, no? if other improvement are involved (which is more than likely) then it's not a rebranded GPU with a shrink and higher clock? right?

 

[Steevo]

nope i am not upset... if i would be ... i would not take time to do that com'on, you know me better

all in all the video and idea behind is interesting, nonetheless.

p.s.: my 1070 still beat your 980Ti buddy (i jest i jest )

Only by a few percent, even that is here and there depending on game, and when accounting for the cost and availability, the 980Ti's with discounts are price king right now and look to remain so until stocks of 1070's manage to shore up the price gouging.

I have seen some other videos from awhile back, he seemed to hit the nail on the head with AMD's positioning in the market with consoles, and where they want to go, the only exception is the politics of game studios and even inside AMD right now, still failing at being truly profitable, the only reasons their books look "better" is due to a lot of financial wrangling of assets and debts on future notes.

 

[GreiverBlade]

Only by a few percent, even that is here and there depending on game, and when accounting for the cost and availability, the 980Ti's with discounts are price king right now and look to remain so until stocks of 1070's manage to shore up the price gouging.

no luck for me ... i was given the choice between a 1070 at 531chf or a 980Ti at .... 957chf ... although the 1080 where 690 to 750chf which still make it cheaper than the 980Ti, where i live ofc ... situational indeed. (also ... "i jest" i know a 1070 is only a 980Ti equivalent, and still preferable nonetheless as it has other advantages than "only a few percent" )


noticed something else ... the 1070 clocking is not far from the 980Ti and sports the same GDDR5 but has way less transistor/core/TMU/ROP/bandwidth ... only a shrink and +300mhz/-68mhz core/mem make that happen? i don't think so. for a underclocked 1080 i understand it come near a 980Ti ... but if even the 1070 does it at stock...

 

[Frag_Maniac]

It wasn't meant to be a revelation. I just do this stuff for a love of the tech. Nvidia are very far ahead without question.

It was stated as a fact I observed, not an intention of yours I assumed. Whether it makes a difference or not, many weren't aware of this, so it is a revelation of sorts anyway, whether intended to be or not.

I think a more telling tale will be whether AMD can eventually offer up anything on the high end that can dethrone the 1070 and 1080 on price and performance, and whether we'll see a lot of devs making use of Dx12 in games, which the AMD cards seem to use quite well.

I basically just want some healthy competition back in the industry. It's sorely needed.

 

[R-T-B]

Why do I get the feeling that you'd be one of the first people claiming that AMD didn't do anything except rebrand for 4 years, even though they added more tessellation every GCN revision, better memory compression and finally HBM?

Downclocking the GPUs reduces their requirement for more bandwidth, that one shouldn't be hard to figure out. It's why weaker GPUs have less bandwidth to start with, they don't need it.

The point was that better graphics compression IS a core improvement. So it's not exactly a "die-shrunk" Maxwell, albeit it may be close.

 

[cadaveca]

The notion that pascal is just a die shrink seems to be exactly the conclusion all of the data is suggesting. No one is saying that it's a bad thing, on the contrary adored in his video said it made perfect sense for nvidia to do.

The video is meant to be informative, thought-provoking and entertaining. People inserting their personal fanbias into it is what causes a large portion of the snarky tension you see in every other post in this thread. Anyone who wants to argue against the conclusions of the video are welcomed to make their own, citing their own sources and drawing their own conclusions, shouting to the heavens that you don't agree "because reasons" is intellectually dishonest.

The fault with the video is in how TFLOPs are matched. Of course GPU designs with similar TFLOP performance will be nearly equal.. what do you think TFLOP measures? Then, where 1080 gets a bit ahead is entirely due to the memory compression. Seems like just a bit of false science to me. If someone didn't know that a FLOP is a way of measuring computational performance, the video could mislead them. Seems like that was the point in the video... to mislead people so that the poster's opinion is justified.

"Hey's let's match VGAs to the same computational performance, and then tell people that they are the same because of that!"

SMH.




So, me being the reviewer, I'm gonna call BS on this video, because that's just my character. I became a reviewer because I wanted to see an end to BS like this, in fact.

The number of shaders can't be matched by matching TFLOP performance. That simply matches performance in general, not shader counts. OMFG! I can't believe you're posting on here like this is anything worthwhile for anyone to watch or pay attention to. It's outright BS.

But as a point of discussion, sure, it's interesting to see how many people didn't catch this part of it. I purposely waited a bit before posting this to give some time for people to see it and form their own opinions, and to see if anyone else would notice this obvious under-handed way of trying to prove a point that amounts to spreading shit all over the internet.

 

[OneMoar]

The fault with the video is in how TFLOPs are matched. Of course GPU designs with similar TFLOP performance will be nearly equal.. what do you think TFLOP measures? Then, where 1080 gets a bit ahead is entirely due to the memory compression. Seems like just a bit of false science to me. If someone didn't know that a FLOP is a way of measuring computational performance, the video could mislead them. Seems like that was the point in the video... to mislead people so that the poster's opinion is justified.

"Hey's let's match VGAs to the same computational performance, and then tell people that they are the same because of that!"

SMH.




So, me being the reviewer, I'm gonna call BS on this video, because that's just my character. I became a reviewer because I wanted to see an end to BS like this, in fact.

The number of shaders can't be matched by matching TFLOP performance. That simply matches performance in general, not shader counts. OMFG! I can't believe you're posting on here like this is anything worthwhile for anyone to watch or pay attention to. It's outright BS.

But as a point of discussion, sure, it's interesting to see how many people didn't catch this part of it. I purposely waited a bit before posting this to give some time for people to see it and form their own opinions, and to see if anyone else would notice this obvious under-handed way of trying to prove a point that amounts to spreading shit all over the internet.

pretty much this
also I suspect nvidia will be-able to get async working a little better or find some extra performance hidden in a dark corner of the ASIC,its not there style to sit and let amd have a 20% advantage in anything

 

[CAPSLOCKSTUCK]

Now ive read all that im gonna finally watch the vid.

 

[TheoneandonlyMrK]

I think many don't realise just how much a node swap let alone complete change in transistor design and shape have to change a design regardless of improvement in IP.

 

[Lionheart]

Thanks for the video Adored, appreciate it

Can't believe people are even bitching about this lol, Maxwell on steroids with some new features isn't a bad thing.

 

[OneMoar]

yea and like AMD haven't been guilty of the same thing
except with AMD its 3 generations of re-brands with no improvement what so ever

 

[GreiverBlade]

The fault with the video is in how TFLOPs are matched. Of course GPU designs with similar TFLOP performance will be nearly equal.. what do you think TFLOP measures? Then, where 1080 gets a bit ahead is entirely due to the memory compression. Seems like just a bit of false science to me. If someone didn't know that a FLOP is a way of measuring computational performance, the video could mislead them. Seems like that was the point in the video... to mislead people so that the poster's opinion is justified.

"Hey's let's match VGAs to the same computational performance, and then tell people that they are the same because of that!"

SMH.




So, me being the reviewer, I'm gonna call BS on this video, because that's just my character. I became a reviewer because I wanted to see an end to BS like this, in fact.

The number of shaders can't be matched by matching TFLOP performance. That simply matches performance in general, not shader counts. OMFG! I can't believe you're posting on here like this is anything worthwhile for anyone to watch or pay attention to. It's outright BS.

But as a point of discussion, sure, it's interesting to see how many people didn't catch this part of it. I purposely waited a bit before posting this to give some time for people to see it and form their own opinions, and to see if anyone else would notice this obvious under-handed way of trying to prove a point that amounts to spreading shit all over the internet.

well that was the Tflop part that ticked me off to refute the idea of a Die Shrink .... glad to see i was not mistaken

Thanks for the video Adored, appreciate it

Can't believe people are even bitching about this lol, Maxwell on steroids with some new features isn't a bad thing.

not bitching about the idea ... but the methodology(or not bitching at all ... just writing ... (and Maxwell on steroid "with some new feature" on a 16nm FinFet node is not maxwell nor a rebrand it's Pascal ... Maxwell on steroid with no other change than a Die Shrink ... is ... well ... a OC'ed Maxwell on a 16nm FinFet node ... but it does not exist... eh? )

The point was that better graphics compression IS a core improvement. So it's not exactly a "die-shrunk" Maxwell, albeit it may be close.

if it was the only improvement, indeed alone it's valide and luckily for Nv there is some other improvement after

 

[Recon-UK]

@Adored
This is not a new thing with Nvidia or AMD, they have been shrinking already used architectures for generations, the only 1 i can think of that was not die shrunk is Fermi and GT200, hell they used G80 for 3 gens, 9800 series and the GTS 200 series.
Still thanks for your video's, i am one of your existing subscribers. Sadly a lot of people nowadays don't have the passion that you and many enthusiasts have and like to speculate off their own back and come out with utter BS.

 

[OneMoar]

https://www.reddit.com/r/nvidia/comments/4uv30h/vulkan_geforce_cards_suffer_from_major_input_lag/
welp that window is getting smaller

 

[GreiverBlade]

@Adored
This is not a new thing with Nvidia or AMD, they have been shrinking already used architectures for generations, the only 1 i can think of that was not die shrunk is Fermi and GT200, hell they used G80 for 3 gens, 9800 series and the GTS 200 series.
Still thanks for your video's, i am one of your existing subscribers. Sadly a lot of people nowadays don't have the passion that you and many enthusiasts have and like to speculate off their own back and come out with utter BS.

still it's not a die shrink ... and i rather have faith in cadaveca than any YouTube reviewer (honestly, never heard of that one ... and the part "I have seen some other videos from awhile back, he seemed to hit the nail on the head with AMD's positioning in the market with consoles" does not convince me much than that) ofc Pascal was added as a inbetween Maxwell and Volta ... because Nvidia was unable to get HBM2 soon enough, yet if it was a higher clocked rebrand on a smaller process node there would be no other improvement, but .... there is (i wouldn't be so insistant if the change was only a few mhz more and a die shrink )

a rebrand is I.E.: 680 to 770 or Hawaii to Grenada (but! same core/transistor/TMU/ROP count but no die shrink .... in the actual case? nope no similarity whatsoever but with a 16nm FinFet node shrink. )

https://www.reddit.com/r/nvidia/comments/4uv30h/vulkan_geforce_cards_suffer_from_major_input_lag/
welp that window is getting smaller

from what i've seen, seems it only affect 1080 ... strangely enough the 1070 and 1060 seems to be less prone to those kind off issues (not that i am unhappy of that since i have none of the issues they mention and that since the 1st "pascal" WHQL driver) oh well ... Vulkan (and probably DX12, for other games) make the RX480 a better card than a 1060 for DOOM (well ... the 1060 is not really interesting versus the RX480 in the end ... ) i wonder ... who is 3 step beyond who ...

yea and like AMD haven't been guilty of the same thing
except with AMD its 3 generations of re-brands with no improvement what so ever

well G80 was not bad either as a rebrand material, altho i would hardly say "with no improvement" because each rebrand carried several improvement and not only in perf (talking for 7970 to 280X or Hawaii to Grenada ) and the GK104 ... 680 to 770 ... well there was some improvement there too ... (GPU Boost 2.0 and adaptative temperature controller inherent to the 7XX line )

 

[ViperXTR]

Ive always thought that Pascal is an improved Maxwell, a die shrink with added refinements within (but not a completely redesigned architecture ala G70 to G80 or Fermi to kepler). Also, ROPs and TMUs are different for 1080 and 980Ti and these 2 components are pretty critical even now as it was before for standard rendering and rasterization/anti-aliasing at high resolutions.

Anandtech's (late but great and detailed review of the GTX 1080 and 1070) discusses refinements added to Pascal vs Maxwell
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review