Pascal offers no IPC gain?

[Drep]

I'm not a math person and will leave this here for you all to hammer through. Yes, I know it's SA but it looks like he did a bit of homework.

Pascal, A New King?? - Clarence Spurr

According to this article, Pascal offers no IPC gain at all and is just basically utilizing the speed increase gains from a new node. Making it appear that it's just really a slightly beefier updated Maxwell on a new node. King of speed in MHz only

 

[primetime]

yes we already know lol card is designed for really high clocks

 

[Armenius]

In all fairness IPC is getting harder to gain these days with current methodologies in computing and manufacturing processes. The rate has been slowly teetering off for a long time and we're getting closer to the floor.

 

[cageymaru]

In all fairness IPC is getting harder to gain these days with current methodologies in computing and manufacturing processes. The rate has been slowly teetering off for a long time and we're getting closer to the floor.

I agree with our current understanding of design. I'm quite sure that the guy who invented the wheel thought it was the end all of tech discovery. Then someone tossed an axle and some boards across it later on.

 

[razor1]

IPC is not a good measurement for GPU's since scalar architectures do 2 ops per clock, per ALU this hasn't changed for a while. The reason for this is to keep the shader compiler simpler. Also the article writer is trying to explain IPC by frequency overclocking. yeah that doesn't work at all, You aren't going to get perfectly linear increase in performance frequency increases in just the GPU, I'm sure the bottlenecks are sifting.

Now yes Pascal will have lower IPC, since its got less ALU's lol, that is pretty much understandable.

Its like trying to compare a 4 core CPU vs a 6 core CPU both of same manufacturer and gens, which one will have higher IPC? I sure hope the 6 core one does.

 

[Armenius]

I agree with our current understanding of design. I'm quite sure that the guy who invented the wheel thought it was the end all of tech discovery. Then someone tossed an axle and some boards across it later on.

Exactly! Quantum computing is continuing to be developed, though, and ternary logic is supposedly making a comeback. I'm just not sure if we'll ever see practical consumer applications in my lifetime.

 

[cageymaru]

Exactly! Quantum computing is continuing to be developed, though, and ternary logic is supposedly making a comeback. I'm just not sure if we'll ever see practical consumer applications in my lifetime.

I think that people are smarter than what you give them credit for. If there is a honest need for something, and the inventor isn't suppressed, changes come pretty rapidly.

 

[cageymaru]

I'm not a math person and will leave this here for you all to hammer through. Yes, I know it's SA but it looks like he did a bit of homework.

Pascal, A New King?? - Clarence Spurr

According to this article, Pascal offers no IPC gain at all and is just basically utilizing the speed increase gains from a new node. Making it appear that it's just really a slightly beefier updated Maxwell on a new node. King of speed in MHz only

That is one awesome article. Thanks for the read! If you think of the GPU system as a whole, it seems that Pascal isn't that much better than Maxwell. If you factor out the die shrink it's actually worse overall. On a good note it seems that there is a TON of untapped power in the design for Nvidia to refine over the next 5 or more years of refreshes. Good thing their last design had so much extra in the tank to allow this design to raise the clock rate significantly.

That's what I got out of the article. Again thanks for the read.

 

[bboynitrous]

Isn't this pretty much the pentium 4 of GPU's?

 

[NKD]

I knew and I am sure most here know it. But I am in the middle when it comes to nvidia vs amd. I knew it from the start and it was obvious that if you clock down 1080 it will probably be the same as 980ti or may be slower. It has frickin 800 mhz clock difference to boot, thats no fricking joke so its damn fast.

AMD knows they can't boost clocks on their architecture and will always be 500mhz behind. 980ti could clock to almost 1500 now max for 390 was probably 1150. You will see that carry over this generation but AMD is trying to make up a little difference 10-20% whatever they can get upgrading all parts around the shaders to make them more efficient.

But 1080 has got the speed no matter how they get it. They will eventually have to tweak the front end down the road but we shall see. They achieved raw performance and maxwell was always hungry for clocks. With pascal they didn't have to do much than do what they needed the most.

 

[razor1]

1080 at 980ti clock speeds will definitely be slower, it just doesn't have enough ALU's to compensate for the same clocks. I would like to see if they have the same theoretical flops..

 

[Daniel_Chang]

Ultimately, it's more powerful than the prior cards and with less power draw. I don't care how they get there, so long as they eventually get there.

 

[misterbobby]

I guess some people are forgetting that Maxwell had a very nice uplift in IPC and even raised the clocks a bit on top of that. Pascal simply just jacked the clocks up. It seems so bizarre to go to a 75% smaller process yet do little more than add 25% more cores that are no faster per clock yet end up with a higher TDP than the card it replaces.

 

[RPGWiZaRD]

So what if it clocks well and TDP remains reasonable? Which it appearently does, I'm buying for cards for the results, not based on how it does to reach those results.

 

[evilsofa]

So what if it clocks well and TDP remains reasonable? Which it appearently does, I'm buying for cards for the results, not based on how it does to reach those results.

What happened to the good old days when the Air Force was complaining that our PCs were drowning out the sound of their jets taking off?

 

[RPGWiZaRD]

What happened to the good old days when the Air Force was complaining that our PCs were drowning out the sound of their jets taking off?

Personally I never intend to pickup a reference design card when you can get for example Gigabyte, MSI or ASUS own designs which works typically better, I'm on my 3rd Gigabyte WindForce cooler design card (GTX 460, 760 and 970) and I love these, especially since the Gigabyte cards can often be picked up for roughly the same price as reference and typically doing a MUCH better job meanwhile being almost inaudible silent at medium fan speeds which has been enough for overclocking too.

 

[MangoSeed]

So, Pascal is Maxwell optimized for higher clocks and that's a bad thing? If power consumption was off the charts then maybe that would be a problem. Talk about grasping at straws.

 

[Ieldra]

This IPC thing makes no sense.

Consider a highway with 2560 lanes.

The cars start moving in their lanes at a rate of two billion per second.

Another highway has 2816 lanes

The cars move in their lanes at a rate of 1.5 billion per second.

The first highway clearly has higher car throughput.

Now when you start talking about IPC you basically ignore the fact that the first highway allows more cars to move per lane per second, and you just look at the number of lanes.


It makes very little sense to me

GP104 has fewer ALUs, fewer ROPs, fewer TMUs, of course it's slower clock for clock. Obvious statement is obvious, also irrelevant because they don't run at similar clocks

 

[CSI_PC]

I guess some people are forgetting that Maxwell had a very nice uplift in IPC and even raised the clocks a bit on top of that. Pascal simply just jacked the clocks up. It seems so bizarre to go to a 75% smaller process yet do little more than add 25% more cores that are no faster per clock yet end up with a higher TDP than the card it replaces.

Remember though it is not just cores that matter but additional aspects such as transistor count, new HW technology such as the changes to the polymorph engine; support for single pass rendering/simultaneous multi display with improved angle correction/etc.
These all take up room, and in some way boost performance either generally or indirectly.
Cheers

 

[euskalzabe]

I read the article and it was quite interesting. The problem that some here don't seem to understand, is that the author is suggesting that we should have had even more performance on the 1080 precisely because of the architecture refinements + die shrink. It seems that all they did was shrink, then boost clocks to get a good performance bump, call it a day. If AMD were more competitive these days, Nvidia would have not been so lazy and rearchitect much more than just that, giving us much more performance on the 1080 for the same price.

I'm still waiting on Polaris 10 being showcased before I decide what to buy.

 

[Ieldra]

I read the article and it was quite interesting. The problem that some here don't seem to understand, is that the author is suggesting that we should have had even more performance on the 1080 precisely because of the architecture refinements + die shrink. It seems that all they did was shrink, then boost clocks to get a good performance bump, call it a day. If AMD were more competitive these days, Nvidia would have not been so lazy and rearchitect much more than just that, giving us much more performance on the 1080 for the same price.

This is ridiculous. No offense, but if this Clarence Spurr fellow was such a digital design expert he probably wouldn't be writing articles for a living.

Architectural changes are needed to allow the design to clock higher. EVERYONE keeps making this mistake, and talking as if all you need to do is "translate" the 28nmFD-SOI maxwell design onto the new 16nm FF process and that's it, higher clocks. THAT'S NOT HOW IT WORKS.

You can complain all you want about "ipc" and "architectural improvements" not being sufficient; it's a complaint that is totally devoid of sense. There is no logic behind this thought-process at all.

NV clearly stated that a big chunk of the R&D budget for pascal went towards routing and critical path optimization, this is what allows to clock so high.

You want to argue that the "IPC due to architectural improvements" should be higher ? Okay then.

Let's pit an 8.6tflop Maxwell GPU vs an 8.6Tflop Fiji GPU and see which achieves a higher effective compute throughput. Obviously the results will vary based on the specific algorithm being used, on the API being used etc etc.

Edit: Another thing I wanted to add, if AMD was capable (in terms of finance) of spending as much as NVIDIA has on R&D and, consequently, clocking their designs as high as Pascal, they would. High clocking designs require less wafer area per unit of performance. It's simply better financially.

Edit #2: Pascal is clocked more aggressively than Maxwell. The stock clocks are closer to the limit than they were on Maxwell.

With Maxwell people complain that the out-of-the-box performance is underwhelming; look at the ruckus caused by a 390X performing similar to 980Ti in AotS.

Now with Pascal people complain the there's not enough OC headroom.

This irritates to me no end, you can't have both.

TLDR; people will bitch and whine, no matter what. 1080 is an extremely impressive piece of hardware, that isn't out yet.

Prices will settle. All will be good.

Funny thing is the NV fan boys would roast AMD for doing the same thing but since its the green team its A-OK!

Thanks for your contribution to the thread.

 

[Elmy]

Nvidia has the upgrade cycle down pat. Boost performance 20-25% at a time and people will buy it. Why would they make a card that is 50% faster and lose the 20-25% step. Its all about making money and Nvidia is very good at it because of their fanbase.

People that Have 980 Ti's and are getting say 80 FPS in a game will buy the 1080 to get 100 FPS in a game. If you did a side by side comparison most people would never be able to tell the difference in that FPS playing on identical systems and the only factor is the GPU.

But they want it.... its called upgradeitis..... its a disease.... LMAO..... there is no cure.... you are all doomed.....

 

[Ieldra]

Nvidia has the upgrade cycle down pat. Boost performance 20-25% at a time and people will buy it. Why would they make a card that is 50% faster and lose the 20-25% step. Its all about making money and Nvidia is very good at it because of their fanbase.

People that Have 980 Ti's and are getting say 80 FPS in a game will buy the 1080 to get 100 FPS in a game. If you did a side by side comparison most people would never be able to tell the difference in that FPS playing on identical systems and the only factor is the GPU.

But they want it.... its called upgradeitis..... its a disease.... LMAO..... there is no cure.... you are all doomed.....

No. The performance boost is calculated relative to the 980. THE 980.

It's not replacing the 980Ti. It's replacing the 980. Maybe if you weren't so busy receiving free hardware from AMD you'd stick your head out of the sand and realize you're just talking baseless crap.

Hilariously hypocritical baseless crap as well; how big a boost is Fury X over 290x hm ? Fury X replaces 290x btw. Do they achieve anything close to the 60-70% increase that 1080 has over 980 ? Nope. But you can't talk about that because AMD will cut you off from receiving your cards

 

[CSI_PC]

Nvidia has the upgrade cycle down pat. Boost performance 20-25% at a time and people will buy it. Why would they make a card that is 50% faster and lose the 20-25% step. Its all about making money and Nvidia is very good at it because of their fanbase.
.

But the 1080 is a replacement for the 980, the 980ti has more SM streams and processor clusters (GPC).
So in that respect the 1080 is around 1.65x to 1.78x faster than a 980.

To expand; the 980 has 4 GPC and 16 SM streams, while the 980ti has 6 GPC and 22 SM streams.
Let alone the ROPs/TU difference.

The 1080 uses same 4 GPC and has 20 SM streams (5 SM per GPC that is the new structure for Pascal), and now 176TU.
But these architecture-HW numbers for the 1080 are still less than the 980ti.
Cheers

 

[Ieldra]

But the 1080 is a replacement for the 980, the 980ti has more SM streams and processor clusters (GPC).
So in that respect the 1080 is around 1.65x to 1.78x faster than a 980.

To expand; the 980 has 4 GPC and 16 SM streams, while the 980ti has 6 GPC and 22 SM streams.
Let alone the ROPs/TU difference.

Cheers

The only "clock for clock" advantage GP104 could have is in rasterizer performance and polygon throughput, but judging by the results from the "Beyond3D Suite" there don't seem to be any major improvements at all.

You can of course bitch and whine about there not being improvements in the geometry front-end (increase throughput due to clocking only) but then you'd have to compare to the competition and ... lol

1080 has 5x polygon throughput of Fury X in Strip format
3x in List format.

3x

Remember though, when it comes to geometry heavy scenes/games/benchmarks and GCN cards perform badly, don't blame GCN! Blame NVIDIA! Blame gameworks! Blame the developer! Blame your mother's dog! But GOD FORBID you point a finger at GCN's front-end. That's a big no no.

I can't wait for Polaris to launch so I can watch Fury X users start complaining that AMD is "gimping" their performance. Polaris 10 will TRASH the Fury X in geometry heavy scenes. TRASH.

 

[euskalzabe]

You can complain all you want about "ipc" and "architectural improvements" not being sufficient; it's a complaint that is totally devoid of sense. There is no logic behind this thought-process at all...

TLDR; people will bitch and whine, no matter what. 1080 is an extremely impressive piece of hardware, that isn't out yet.

Whoa, gee. Way to insta-escalate there, buddy; bring it down a notch. All I said was that I found the article interesting and it made a good point: obviously you still need to re-architect if you're changing the manufacturing process, but you can't seriously believe that we're getting everything we can from Nvidia, do you? AMD hasn't been able to compete seriously for a while now (achieving similar performance with crazy TDP doesn't really count), and while it's not as bad as in the CPU side of things, Nvidia is getting a little bit complacent, just like Intel has - for the second time in its history.

For the record: I wasn't complaining. All I said was that I have a feeling Nvidia could have done more with Pascal but it was under no pressure and so didn't try that hard. It's still a great card, just not as good as I had hoped. There's a good amount of innovation in the 2016 generation cards, I expected even better than what we got. That's all.

 

[Ieldra]

Whoa, gee. Way to insta-escalate there, buddy; bring it down a notch. All I said was that I found the article interesting and it made a good point: obviously you still need to re-architect if you're changing the manufacturing process, but you can't seriously believe that we're getting everything we can from Nvidia, do you? AMD hasn't been able to compete seriously for a while now (achieving similar performance with crazy TDP doesn't really count), and while it's not as bad as in the CPU side of things, Nvidia is getting a little bit complacent, just like Intel has - for the second time in its history.

For the record: I wasn't complaining. All I said was that I have a feeling Nvidia could have done more with Pascal but it was under no pressure and so didn't try that hard. It's still a great card, just not as good as I had hoped. There's a good amount of innovation in the 2016 generation cards, I expected even better than what we got. That's all.

But you can't say NV aren't doing all they can, it's just not a claim you can make, because you don't know what they "can" do. The 1080 is leaps and bounds faster than the card it's replacing, and the performance delta between the two is actually bigger than usual, it's certainly not a small improvement. Either way, it isn't even a question of capability, it's a question of financial viability.
As for Intel, Intel hasn't really gotten complacent, it's just that there's only so much you can do to increase CPU performance; it's not like GPUs that you can just scale up and produce them with more "cores". The stagnation is exacerbated by the need to maintain compatibility with x86-64 ISA. The pricing of their CPUs is a result of complacency/lack of competition, but their performance is just stellar

Plus the whole "IPC is lower" thing is only valid when you compare to the 980Ti or Titan X, both cards the 1080 is not replacing...

Using this bogus IPC metric the uncontested GPU champion would be Fiji XT with it's 4096 ALUs

The most important thing here, to me, is that you guys stop thinking that Pascal clocks high just because of the process change. That's not how it works. It requires deep architectural changes and optimization to reduce maximum delay between latched events to the point where you can safely clock to 2ghz. Huge amount of work

Considering 2000mhz then max latency between latched events must be well under half a nanosecond (500 picoseconds)

 

[razor1]

Whoa, gee. Way to insta-escalate there, buddy; bring it down a notch. All I said was that I found the article interesting and it made a good point: obviously you still need to re-architect if you're changing the manufacturing process, but you can't seriously believe that we're getting everything we can from Nvidia, do you? AMD hasn't been able to compete seriously for a while now (achieving similar performance with crazy TDP doesn't really count), and while it's not as bad as in the CPU side of things, Nvidia is getting a little bit complacent, just like Intel has - for the second time in its history.

For the record: I wasn't complaining. All I said was that I have a feeling Nvidia could have done more with Pascal but it was under no pressure and so didn't try that hard. It's still a great card, just not as good as I had hoped. There's a good amount of innovation in the 2016 generation cards, I expected even better than what we got. That's all.

Changes come gradually, we haven't seen huge generation leaps in performance for many years now, and even when we did see then it was associated with a major change in the way API's functioned. Like going from fixed function to programmable and programmable to unified (dx8 to dx9 and dx 9 to dx10 respectively). DX12 is more evolutionary than redefining like the previous API changes. Also we are looking at performance bracket card, enthusiast cards aren't even out yet, I would expect another 25-50% boost from what the 1080 has for its base performance.

And yeah you don't use IPC to figure out which architecture is better. Because GPU's don't rely on IPC to get their performance (well at some level they do, but its not the end all), they rely on the ability to do many things at once simultaneously and always have done this.

Once if you equalize the Tflops then you will get the actual IPC, which that article didn't do, nor did the author even understand that. What he stated IPC, was the entire IPC for the entire GPU not just one core of the chip (which is what we talk about with CPU's, one core of the cpu). This is why its not a good way to measure GPU performance, because there is so many other things involved with the IPC of the entire GPU which constitute its throughput.

The 1070 which will have the same Tflops or close to the Titan X and its performance is close to it. Which shows IPC wise they are the same...... But the 1070 has much less power draw at higher frequencies and a hell of a lot less bandwidth, so over all, it has things the Titan X just doesn't have.

 

[euskalzabe]

It may be the case that I'm used to get better improvements or features from generation to generation. 10 years ago things were more dramatic. 20 years ago, I can't even begin to tell you... every video card upgrade was like a new universe. Advances seem to have become so... humble, in comparison to years ago. Maybe this is the new reality. Maybe I need to buy a GPU every 4 years instead of 2. I'm still rocking my 770 and it's only flaw is the 2gb of VRAM. Then again, I'm only gaming at 1080p, I don't need the excesses we have today with 3 screens, 4K, etc... to game happily, but that's just me.

 

[Ieldra]

It may be the case that I'm used to get better improvements or features from generation to generation. 10 years ago things were more dramatic. 20 years ago, I can't even begin to tell you... every video card upgrade was like a new universe. Advances seem to have become so... humble, in comparison to years ago. Maybe this is the new reality. Maybe I need to buy a GPU every 4 years instead of 2. I'm still rocking my 770 and it's only flaw is the 2gb of VRAM. Then again, I'm only gaming at 1080p, I don't need the excesses we have today with 3 screens, 4K, etc... to game happily, but that's just me.

You are getting improvements and features from generation to generation

If you compare the 1080 to the card it actually replaces, stock vs stock, it's 60+% faster

 

[euskalzabe]

You are getting improvements and features from generation to generation

If you compare the 1080 to the card it actually replaces, stock vs stock, it's 60+% faster

Yeah, yeah. We're getting things, just not as many as I would hope. Talk to me about the original Geforce. Or the 8800GT. Now, those were clear, impressive advancements. Since then, it's all been super evolutionary. The 1080 will be good, but not revolutionary. Also, jacking up the prices is a big mistake. I have a feeling AMD will force these prices down - hence why I don't get why so many people will be wasting their money in 2 days (unless having it before anybody else is worth money to those people... which I also don't understand).

 

[Ieldra]

Yeah, yeah. We're getting things, just not as many as I would hope. Talk to me about the original Geforce. Or the 8800GT. Now, those were clear, impressive advancements. Since then, it's all been super evolutionary. The 1080 will be good, but not revolutionary. Also, jacking up the prices is a big mistake. I have a feeling AMD will force these prices down - hence why I don't get why so many people will be wasting their money in 2 days (unless having it before anybody else is worth money to those people... which I also don't understand).

Man. You're comparing Tesla to GeForce 7 series

That was the jump to unified shaders.

AMD won't have a say in pricing until they actually release a competitor. That's months away.

 

[euskalzabe]

I know, that's why I don't criticize these GPU/CPU releases anymore, there's no point. I'm just disillusioned. Maybe that comes with age. I've seen, what, 20 years of hardware upgrade cycles? Lately they all blend together and while I'm sure technology is getting slower due to diminishing returns, man did we have a good 2-decade run - and I got to enjoy it all from being a kid to becoming an adult. 10 years ago I was happy to buy the fastest option available because it was monstrously better than the rest. These days... I get a x70 range card and I'm fine for 2 years. Not very exciting, specially when games also don't evolve due to sequelitis. Oh well, I'll stop ranting now. I'm bringing myself down

 

[razor1]

Yeah, yeah. We're getting things, just not as many as I would hope. Talk to me about the original Geforce. Or the 8800GT. Now, those were clear, impressive advancements. Since then, it's all been super evolutionary. The 1080 will be good, but not revolutionary. Also, jacking up the prices is a big mistake. I have a feeling AMD will force these prices down - hence why I don't get why so many people will be wasting their money in 2 days (unless having it before anybody else is worth money to those people... which I also don't understand).

AIB partner cards are going to be reviewed in 2 days, and launch of AIB and founder edition cards are the same time now, so lets see if that statement holds up about paying more, I think a lot of people will pick their favorite AIB's and not worry about the founder's edition.

 

[Ieldra]

I know, that's why I don't criticize these GPU/CPU releases anymore, there's no point. I'm just disillusioned. Maybe that comes with age. I've seen, what, 20 years of hardware upgrade cycles? Lately they all blend together and while I'm sure technology is getting slower due to diminishing returns, man did we have a good 2-decade run - and I got to enjoy it all from being a kid to becoming an adult. 10 years ago I was happy to buy the fastest option available because it was monstrously better than the rest. These days... I get a x70 range card and I'm fine for 2 years. Not very exciting, specially when games also don't evolve due to sequelitis. Oh well, I'll stop ranting now. I'm bringing myself down

Another thing you may not be aware of, but the cost per transistor is now higher than it was on 28nm.

If Nvidia just copy paste their design on the new node, say gm200, it'll actually cost more for them to produce. It sucks but that's how it is

 

[euskalzabe]

AIB partner cards are going to be reviewed in 2 days, and launch of AIB and founder edition cards are the same time now, so lets see if that statement holds up about paying more, I think a lot of people will pick their favorite AIB's and not worry about the founder's edition.

Oh I wasn't even considering the FE, just talking about the $379 for the 1070 vs the $330 it should have been kept at. Not that it's a problem for me, but that'll lock out a lot of people who would have upgraded to this. A 50 buck increase is not inconsiderable.

Another thing you may not be aware of, but the cost per transistor is now higher than it was on 28nm.

If Nvidia just copy paste their design on the new node, say gm200, it'll actually cost more for them to produce. It sucks but that's how it is

That's a good point, 16nm is new for these GPUs. That means that by next year we should expect prices to go down again... but once they're up, I'm not so sure they'll go down - although this has happened before, admittedly.

 

[Ieldra]

Oh I wasn't even considering the FE, just talking about the $379 for the 1070 vs the $330 it should have been kept at. Not that it's a problem for me, but that'll lock out a lot of people who would have upgraded to this. A 50 buck increase is not inconsiderable.



That's a good point, 16nm is new for these GPUs. That means that by next year we should expect prices to go down again... but once they're up, I'm not so sure they'll go down - although this has happened before, admittedly.

They'll go down because of yields if anything.

BTW, cost per transistor is 7% higher according to tsmc, and that assumes maximum density, for lower density parts (like gp104) it will be even higher.

Compare gm204 to gp104, more transistors, higher cost per transistor = price hike


Price increase is

 

[razor1]

Oh I wasn't even considering the FE, just talking about the $379 for the 1070 vs the $330 it should have been kept at. Not that it's a problem for me, but that'll lock out a lot of people who would have upgraded to this. A 50 buck increase is not inconsiderable.



That's a good point, 16nm is new for these GPUs. That means that by next year we should expect prices to go down again... but once they're up, I'm not so sure they'll go down - although this has happened before, admittedly.

well 50 buck increase is a 15% increase and when you look at transistor amounts of the 970 to the 1070, so its actually not that much of an increase.

 

[Ieldra]

well 50 buck increase is a 15% increase and when you look at transistor amounts of the 970 to the 1070, so its actually not that much of an increase.

Lol. Beat you to it

 

[CSI_PC]

You are getting improvements and features from generation to generation

If you compare the 1080 to the card it actually replaces, stock vs stock, it's 60+% faster

And in some games that Maxwell 2 did not seem to be optimal the 1080FE is even 75% to 85% faster.
In fact HardwareCanuck show it being 100% faster in Quantum Break than 980, Maxwell/Kepler has really big issues with the post processing volumetric/global lighting that game engine has (which is more aligned towards AMD and GCN).
Cheers