Pascal vs Maxwell **same clocks**

[Griffildur]

Until a whole load of new games come out that its not been optimized for as support has ended for it now so hopefully it doesnt end up in 780ti lane...

a 980 TI will be a great card for years. All this "ended support" means nothing considering how powerful that card is and how great it overclocks.

 

[akarypid]

Indeed. It has taken AMD nearly two years just to bring something out that matches the 970/980 in perf per watt.

It is slightly concerning in terms of what they can do to counter GP102 but hopefully vega has even further improved efficiency over POlaris and isn't just a scaled up Polaris. - HBM will obviously be used to alleviate the efficiency disadvantage as well (like with the fury range).

Even though HBM will surely help with consumption, ultimately both companies will be using it soon. So any respite AMD may get through the use of HBM will be very short-lived.

 

[akarypid]

I really don't see the point of all this. Obviously Pascal and Maxwell have the same successful underlying GPU architecture. Pascal has a couple of VR/Async extra features but the rest is the same. So what?

I mean, you buy a 1080 because it's a 16nm chip that reaches performance via higher clocks with fewer cores, which the 980ti can't do. What's bad about it?

 

[Seanspeed]

Pascal is not a fresh new architectural jump for Nvidia, but it is hardly just Maxwell + die shrink like many are saying.

You dont just produce a GPU that can clock up to 2Ghz+ by sticking a smaller node process on there. There was inevitably a ton of engineering that went into achieving what they did.

 

[Rroff]

Pascal is not a fresh new architectural jump for Nvidia, but it is hardly just Maxwell + die shrink like many are saying.

You dont just produce a GPU that can clock up to 2Ghz+ by sticking a smaller node process on there. There was inevitably a ton of engineering that went into achieving what they did.

People should really read the Pascal whitepapers, etc. before parroting that its just a shrunk Maxwell.

 

[LegendaryDread]

Perhaps this is a stupid question, I'm tired (that's my excuse and I'm sticking to it!):

If a comparison must be made, why not use GTX980 and GTX1070? They seem to be much more similar than 980Ti and 1080.

At least with the 980 and 1070, both are 256-bit and both use the same type of memory. Both also have almost identical ROPs/TMUs/Shader counts - unlike the 1080 and 980Ti.

 

[TheRealDeal]

Perhaps this is a stupid question, I'm tired (that's my excuse and I'm sticking to it!):

If a comparison must be made, why not use GTX980 and GTX1070? They seem to be much more similar than 980Ti and 1080.

Both are 256-bit, both use the same type of memory. Both have almost identical ROPs/TMUs/Shader counts. The biggest difference between the two cards is clock speed.

I think you are confusing the gtx980 with the gtx980ti which has a 384-bit bus and more of everything pretty much in comparison to the gtx1070/1080 except memory. The Pascal cards gain performance through very high clock speeds.

 

[Lokken86]

I don't get it. If you make them the same TFlops of course they are going to perform similar. Kind of a dumb test really.

I think it's funny the idea of AMD reverse engineering Nvidia's stuff haha. I never thought of it like that before but they can learn a lot doing it. Is that legal?

 

[TheRealDeal]

I don't get it. If you make them the same TFlops of course they are going to perform similar. Kind of a dumb test really.

I think it's funny the idea of AMD reverse engineering Nvidia's stuff haha. I never thought of it like that before but they can learn a lot doing it. Is that legal?

Different architectures would most likely not perform the same at the same Tflops. You only need to look at Amds max Tflops v Nvidia cards to see this. Amd cards usually have a higher Tflop but are often slower as they can't use there full potential. As Pascal and Maxwell are very similar they perform very similar at the same Tflops.

 

[LegendaryDread]

I think you are confusing the gtx980 with the gtx980ti which has a 384-bit bus and more of everything pretty much in comparison to the gtx1070/1080 except memory. The Pascal cards gain performance through very high clock speeds.

The "both" I was referring to in the last line was the GTX980 and 1070. Edited for clarity.

 

[Poneros]

Interesting. Too bad the frequency difference isn't small enough (1.5 ghz vs 2.1 ghz / 980 ti vs 1080) to off-set it, though in that case the 1080 would look very questionable.

 

[Lokken86]

Different architectures would most likely not perform the same at the same Tflops. You only need to look at Amds max Tflops v Nvidia cards to see this. Amd cards usually have a higher Tflop but are often slower as they can't use there full potential. As Pascal and Maxwell are very similar they perform very similar at the same Tflops.

Well I don't think it's quite the same as Nvidia vs AMD technology but yea I know what you mean. Still Pascal did not improve much on the Maxwell architecture so much as die shrinking it and cranking up the clock speeds, so in theory Tflop to Tflop they should be close in this case.

 

[SiDeards73]

I watched a video that basically concluded the reason Nvidia Maxwell and Pascal are more efficient than GCN cards is because GCN has a lot of hardware features that even though they don't get used still require power, Pascal and Maxwell lack these and thus don't require the extra power required to run.

This may explain why no hardware support of Async etc, however should Nvidia add this support to hardware their power efficiency will take a hit like AMD has and you may see closer power consumption between manufacturers.

They concluded by saying if Nvidia add hardware support for a lot of features like AMD have on Volta then expect the power consumption to be higher and the cards to be less efficient.

 

[Griffildur]

got anything to share other than ifs and maybes?

 

[Gregster]

I watched a video that basically concluded the reason Nvidia Maxwell and Pascal are more efficient than GCN cards is because GCN has a lot of hardware features that even though they don't get used still require power, Pascal and Maxwell lack these and thus don't require the extra power required to run.

This may explain why no hardware support of Async etc, however should Nvidia add this support to hardware their power efficiency will take a hit like AMD has and you may see closer power consumption between manufacturers.

They concluded by saying if Nvidia add hardware support for a lot of features like AMD have on Volta then expect the power consumption to be higher and the cards to be less efficient.

Interesting. Got a link to that vid?

 

[mahius]

I remember someone asking him to make more Nvidia content, to which he replied along the lines of: What else is there to say about Pascal? We know it has terrible DX12 performance and non-existent A-Sync. Of course his minions were quick to upvote this.

Yup, I'm pretty sure that was me. I asked about potential AMD bias and not enough Nvidia content (he never did a speculation video for the 1060). The reply was like I'd touched a nerve... so I calmly just agreed afterwards in the hopes that I'd not further provoke the guy.

I will say though, this clock for clock comparison seems kinda reasonable. Back when the 1080 released, many of us on these forums were suggesting that it's Maxwel with higher clocks anyway. Of course, they did some work to make sure such clocks were both possible and stable, but at similar clockspeeds, the 1000 series does feel a tad underwhelming. If anything, this just makes the older Maxwell cards more desirable over the new Pascal ones to me. Just because they aren't all as outdated as we think.

I thought some 980tis could reach 1.5 Ghz? So why not test at that since it's much closer to 1080 base clock and less tricky fiddling.

 

[mustrum]

Indeed. It has taken AMD nearly two years just to bring something out that matches the 970/980 in perf per watt.

It is slightly concerning in terms of what they can do to counter GP102 but hopefully vega has even further improved efficiency over POlaris and isn't just a scaled up Polaris. - HBM will obviously be used to alleviate the efficiency disadvantage as well (like with the fury range).

I doubt AMD can come even clocse with performance per watt. Honestly: I don't give a rats ass either. I hope they come up with something that performs somewhat close to GP102. If they don't manage that I see them in trouble.

 

[r7slayer]

If you drive a Ferrari and a Nissan Micra at 60Mph they are both as fast as each other. Save your money, don't buy a Ferrari, a micra is just as fast.

Once of the worst analogies in the GPU section.

 

[Wrinkly]

I don't know if that has been posted before but I've just stumbled across this video with a 980 Ti vs 1080 at the same clocks give or take a few mhz due to GPU boost, as you can see the 980 Ti is faster due to its cuda core advantage
what do you guys make of this? so with everyone mentioning drivers etc for the maxwell cards maxwell and pascal are so similar that I don't think drivers will matter as much as people make out with kepler and maxwell.
https://www.youtube.com/watch?v=nDaekpMBYUA

This was my gut feeling when Pascal launched.

 

[KillBoY_UK]

I watched a video that basically concluded the reason Nvidia Maxwell and Pascal are more efficient than GCN cards is because GCN has a lot of hardware features that even though they don't get used still require power, Pascal and Maxwell lack these and thus don't require the extra power required to run.

This may explain why no hardware support of Async etc, however should Nvidia add this support to hardware their power efficiency will take a hit like AMD has and you may see closer power consumption between manufacturers.

They concluded by saying if Nvidia add hardware support for a lot of features like AMD have on Volta then expect the power consumption to be higher and the cards to be less efficient.

I find it hard to believe unused parts of a GPU are leading to upto 1/3 more power being used than your competitor. Im thinking along the lines of the 1060 and RX480 or 980 and R290/390 etc