Radeon RX 480 "Polaris" Launched at $199

[Kaapstad]

The new 16nm architecture has still masively improved power efficiency though, even over and above Polaris/480. A 150w 1070 can outperform a 250w 980tI.

You have not met my 980 Ti's.

 

[Kaapstad]

Maxwell and Pascal have exactly the same performance per clock ratio (according to PcPer) - what you perceive as inferior is because what we have now has less Shaders, Rops and memory bandwidth than Titan or Ti.

Overclock one of these to GTX 1080 clocks and the difference is massive.

 

[Kaapstad]

Its got nothign to do with sacrafcing quality.

When you analyze the theoretical clock speed of a processor you have to look at the instructions that take the longest times, which invariably means an electrical signal that has to travel through the most number of transistors. Each transistor has a switching speed governed by the process. A certain lengthy instruction that must travel through X transistors each switching at Y nano seconds will take X*Y nanosecond to compute, giving you he highest possible clock speed of 1/X*Y. If you find these critical long instructions and improvement them so they say less time, say 0.8*X then you can increase the clock speed to 1/0.8*X*Y. Sometimes you can't optimize the instruction, so you split that instruction to be computed over multiple clock cycles.Lets say you double the number of clock cycles o commute that instruction, your clock speed is now 1/0.5*X*Y, however whenever that instructions needs computing it will take twice the number of clock cycles.In doing so you have reduced the IPC. The IPS for that instruction has remained the same, but the average IPS of all instructions has increased.


So for about the 1 billionth time, IPC is absolutely meaningless in determining performance. NVidia have official said they have performed extensive critical path optimization on Pascal in order to maximize clock speeds, at the potential expense of IPC (although most reports are indicates there is a pretty minimal difference in IPC). It is completely meaningless how Nvidia achieved the increase in performance, all that matters is real world gaming performance.

The same applies to AMD. The 480 may not be as fast as the 390x in theory, but all that matters is real world performance.

As I was saying.

 

[Kaapstad]

That's exactly what I said ._. The difference would be massive just because of the extra Rops and Shaders.

1080 and 1070 are carried only by their crazy clocks, as far as everything else they are indeed midrange.

You realise the 980 Ti and 1080 both have about the same number of active transistors

 

[Kaapstad]

No, you were saying soemthign about sacrificing quality, which is meaningless garbage.

Think what you want, your previous post got it about right.

 

[Kaapstad]

This is what I see, besides clock speeds they don't have anything over Titan X and Ti.

With the 1080 it is all about clockspeeds, there is nothing wrong with high clockspeeds but that is where the performance comes from.

The transistor count on that chart for the 980 Ti are not all active as the chip has some disabled compared to the TitanX.

 

[Kaapstad]

Lets see what happens when you run a 980 Ti at normal 1080 clockspeeds.



http://www.3dmark.com/hall-of-fame-2/fire+strike+3dmark+score+performance+preset/version+1.1/1+gpu

What is also interesting is there are no 1080s in the top 20.

I think the 980 Ti wins easy on efficiency.

 

[Kaapstad]

I think we're saying the same thing

We are.

It does not matter how you get the performance as long as it is there.

 

[Kaapstad]

Does make the situation seem even more artificial though (as in let's keep stuff off the card for when we'll need to get a Titan out).

Pascal Titan will be a monster but this time big Vega should give it some real competition.

 

[Kaapstad]

The 1060 is rumoured for July. BTW you didn't answer my question?

Easy to get a rough idea what the performance of the 1060 will be.

Half the spec of the 1080 and half the performance.

 

[Kaapstad]

Why is there a bandwidth difference between the 4gb and 8gb versions.

 

[Kaapstad]

Cool, so if you're buying now you'd say "No I won't have newer faster tech for less money as it's new tech so should have been a bigger jump, I'll buy the old one" .... ?



That... is a bit annoying

It could be a bit GTX 970

Simply going from 4gb to 8gb should make no difference to the bandwidth.

 

[Kaapstad]

No its just using slower Ram speeds for some reason. It was explained somwhere in this thread. I think it was to give the 8gb version more performance.

Yea it is just lower clocks.

They must have some pretty poor RAM to get rid of if it is clocked lower than the 8gb versions.

 

[Kaapstad]

Blown out of proportion I expect, as AMD would know the implications of selling an out of spec Pci-e card.

Until others come back with similar tests, it could be a mistake with their testing methodology (not making excuses, just needs confirming)

I can think of a few cards that could blow up a poor mother board but the RX 480 is not one of them.

 

[Kaapstad]

I don't know why some users thought that a 380 replacement should be a worthy upgrade to a 290/290x/970 user.

The FX 480 does exactly what it is supposed to do.

It is good value for money and offers good performance.

 

[Kaapstad]

I do wonder if certain people in this thread even play games with their PCs.

Like who ?

 

[Kaapstad]

Was this a sarcastic swipe at the 480...by Flopper?!!

You mean Flip Flopper.