AMD Polaris Thread: Radeon RX 480, RX 470 & RX 460 launching June 29th

[Silverforce11]

1080- 20-25% faster than 980 Ti. 2x the performance in VR
1070- ~980Ti in gaming, 60% more performance in VR compared to 980Ti

Really, 2x in VR? Which games? Can you show 2x in available VR games? -_-

How did Maxwell's claimed awesome VR feature worked out?

Infact, let's just come out with the truth now that it's been a few years.... NV claimed Maxwell was awesome for DX12 when it launched back in 2014. Some of you people were hyping up claiming FL12_1 support, better than GCN for DX12!! LOL

 

[IntelUser2000]

How does same/similar performance equate to same architecture?

They can say whatever they want, and its true that it might do exceptional in few corner case scenarios.

390X - 1.05GHz with 2816 = 5.9TFlops
RX 480 - Assuming 1.25GHz, and 2304, its 5.76TFlops

2% difference in Flops and 3% faster in benchmark is 5-6% faster in total. That's for all intents and purposes, the same.

Have you guys forgot already that these used to be called mediocre in CPU territory? Why does a company making a processor doing embarassingly parallel compute aimed at mostly performance oriented folks bring a product that's 5% faster but 50% the power reduction, and price it low? Can't they do any better?

Infact, let's just come out with the truth now that it's been a few years.... NV claimed Maxwell was awesome for DX12 when it launched back in 2014. Some of you people were hyping up claiming FL12_1 support, better than GCN for DX12!! LOL

<--No such claims made. Nvidia isn't doing any better BTW. RS's claim that Nvidia has been increasing prices so a low-mid card went to high-end/enthusiast pricing is exactly true.

 

[DarkKnightDude]

the leak from WCCF Tech tomorrow sheds more light on it.

Don't believe anything till reviews. Clickbait sites trying to get views.

 

[Silverforce11]

They can say whatever they want, and its true that it might do exceptional in few corner case scenarios.

390X - 1.05GHz with 2816 = 5.9TFlops
RX 480 - Assuming 1.25GHz, and 2304, its 5.76TFlops

2% difference in Flops and 3% faster in benchmark is 5-6% faster in total. That's for all intents and purposes, the same.

Have you guys forgot already that these used to be called mediocre in CPU territory? Why does a company making a processor doing embarassingly parallel compute aimed at mostly performance oriented folks bring a product that's 5% faster but 50% the power reduction, and price it low? Can't they do any better?

Assuming AMD's TDP rating remains the same, gaming power usage of RX 480 is well below 150W, about 7850 levels, or ~100W.

Almost a 2.5x TFlops per watt improvement.

Also a mainstream part for $199 that beats a 390X... and you're bad mouthing it... quite sad actually.

At first, some of you people said why is it so cheap?! Now you're saying "why isn't it faster?!"...

Do you not know the answer?

There are many tiers of GPUs. It's nothing new. Dude, come on, you are being silly.

Honestly, in the other thread, some of you guys are saying if Polaris overclocks like a champion, it means AMD failed and left too much performance on the table... stuff like that was never uttered about Maxwell. -_-

 

[Ansau]

They can say whatever they want, and its true that it might do exceptional in few corner case scenarios.

390X - 1.05GHz with 2816 = 5.9TFlops
RX 480 - Assuming 1.25GHz, and 2304, its 5.76TFlops

2% difference in Flops and 3% faster in benchmark is 5-6% faster in total. That's for all intents and purposes, the same.

Have you guys forgot already that these used to be called mediocre in CPU territory? Why does a company making a processor doing embarassingly parallel compute aimed at mostly performance oriented folks bring a product that's 5% faster but 50% the power reduction, and price it low? Can't they do any better?

Yes, but that's just at dx11 in a quite old synthetic benchmark.

We still don't know the true power consumption (anyone thinking the rx480 will consume 150W should realize he's wrong).
We still don't know how far it will overclock.
We still don't know the real performance in dx11 games.
We still don't know the true performance in dx12. AoS is a bit skewed because it only pushes for async compute, but dx12 is much more.
We still don't know how much better it is in VR.
We still don't know what are the full package of extra features. These also count as architecture improvements. So far we know DP 1.4 and HDR, but how about improved VCE, VSR, color compression...

Don't judge anything until you have the whole picture.

 

[IntelUser2000]

Assuming AMD's TDP rating remains the same, gaming power usage of RX 480 is well below 150W, about 7850 levels, or ~100W.

Almost a 2.5x TFlops per watt improvement.

Flops/watt only gained traction recently.

Also a mainstream part for $199 that beats a 390X... and you're bad mouthing it... quite sad actually.

Perhaps I should ask you if you are invested(emotionally or literally).

2 full generations after 290X should net you far lower prices, especially when competition is there and you have 10-20% marketshare. I am not saying $200 is bad, but for the sake of technological advancement its mediocre.

At first, some of you people said why is it so cheap?! Now you're saying "why isn't it faster?!"...

No I am not being silly. I am just bewildered by why silicon manufacturers can't pull much performance wise in the PC space. What's the 14nm for? All the benefits are being moved not just to perf/watt, but watt.

It's just like the first company that used MuGFETS(multi-gate FETS), Intel. They took a 35W with Sandy Bridge, and kept lowering power, without increasing performance much. In fact, that's what the Skylake U chips are. The 28W chips are so rare and expensive that they practically don't exist.

I've watched PC space and benchmarks very closely for more than a decade. Even right after Pentium 4's heat debacle, it was still all about performance. Suddenly, even in the extreme enthusiast category - AKA GPUs, its all about watts. Lowering it.

 

[Thala]

Why does a company making a processor doing embarassingly parallel compute aimed at mostly performance oriented folks bring a product that's 5% faster but 50% the power reduction, and price it low? Can't they do any better?

That is maybe, just maybe, we are talking about RX480 and not RX490... I hope you noticed that 390X has a "9" in the name.
RX480 brings incredibly good value to mainstream.

If you are looking for high end performance you should look into the upcoming 490 series (Vega).

I am not saying $200 is bad, but for the sake of technological advancement its mediocre.

If you are essentially doubling perf/W you call this mediocre? You have a very strange view i must say.

 

[Silverforce11]

Flops/watt only gained traction recently.

Perf/w is what killed AMD.

Kepler vs Tahiti/Hawaii was not bad for AMD. Until Maxwell came and blew away other GPUs in Perf/w.

They are right to focus on perf/w. Better perf/w allows them to scale up to higher performance to the high-end.

If you are concerned about raw performance, there's Vega 11 & 10 from AMD coming later. Certainly it isn't lacking in progress at all.

You're just feigning disappointment about a mainstream chip that offers great perf/$ and potentially perf/w for reasons that do not make any sense.

This is almost like Kyle @ [H] bashing Polaris for not competing vs the 1080.

Btw, did you feel this disappointment about the GTX 750Ti and 960? "Oh, why can't the GTX 960 compete versus the 980?!"...

 

[trane]

Perf/w is what killed AMD.

Kepler vs Tahiti/Hawaii was not bad for AMD. Until Maxwell came and blew away other GPUs in Perf/w.

Actually, Maxwell ain't that great. It's just that Hawaii is a power hog and AMD decided to push it to it's limit way beyond the optimum clocks/W curve.

Hawaii aside, you'll see that vanilla Fury pretty much matches the GM200 cards, while Fury Nano is probably the most efficient card last generation. Granted, it's a binned specialty product. Fury X is not that far behind either.

Same story in other places. R9 380X is within 20% of GTX 960 and 950. Yes, Maxwell wins, but it doesn't blow away anything. Same story for R7 370 vs 750 Ti.

Finally, if we look at compute and DX12 perf/W, AMD is actually ahead across the board. (Again, except Hawaii)

 

[Silverforce11]

Actually, Maxwell ain't that great. It's just that Hawaii is a power hog and AMD decided to push it to it's limit way beyond the optimum clocks/W curve.

Hawaii aside, you'll see that vanilla Fury pretty much matches the GM200 cards, while Fury Nano is probably the most efficient card last generation. Granted, it's a binned specialty product. Fury X is not that far behind either.

Same story in other places. R9 380X is within 20% of GTX 960 and 950. Yes, Maxwell wins, but it doesn't blow away anything. Same story for R7 370 vs 750 Ti.

Finally, if we look at compute and DX12 perf/W, AMD is actually ahead across the board. (Again, except Hawaii)

You can argue that, but at release, the 970 and 980 embarrassed the 290 and 290X. The 970 match 290X with much less power usage.

These days the 290X is well in front, often matching the 980 so it has improved but it wasn't the case in 2014.

The facts are AMD got down to 18% marketshare because of cards like the 750Ti, 960 and 970. Maxwell's perf/w advantage really pwned AMD hard.

Six months ago, many were skeptical that AMD's next-gen could catch up to Pascal's perf/w given how far they were behind vs Maxwell...

 

[Qwertilot]

No I am not being silly. I am just bewildered by why silicon manufacturers can't pull much performance wise in the PC space. What's the 14nm for? All the benefits are being moved not just to perf/watt, but watt.

Ok. AMD seem to be doing watts because they couldn't do a whole chip line at once - its getting very expensive to do these designs and they're hardly flush with cash just now - and they needed low(er) power/cost things for the console refresh wins, and other semi custom stuff.

With Polaris 10 at this sort of performance, size and power draw it is an absolute certainty that they can do bigger chips scaling notably past it. Those big chips will be able to do that much better if its 100w than if its 150w.

If it consumed 250w or whatever the 290(X) did, then that would stop them scaling it up at all.

 

[LTC8K6]

Really, 2x in VR? Which games? Can you show 2x in available VR games? -_-

How did Maxwell's claimed awesome VR feature worked out?

Infact, let's just come out with the truth now that it's been a few years.... NV claimed Maxwell was awesome for DX12 when it launched back in 2014. Some of you people were hyping up claiming FL12_1 support, better than GCN for DX12!! LOL

Well...

http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-980-ti/specifications

12 API with Feature Level 12_1

 

[3DVagabond]

Flops/watt only gained traction recently.

Perhaps I should ask you if you are invested(emotionally or literally).

2 full generations after 290X should net you far lower prices, especially when competition is there and you have 10-20% marketshare. I am not saying $200 is bad, but for the sake of technological advancement its mediocre.

No I am not being silly. I am just bewildered by why silicon manufacturers can't pull much performance wise in the PC space. What's the 14nm for? All the benefits are being moved not just to perf/watt, but watt.

It's just like the first company that used MuGFETS(multi-gate FETS), Intel. They took a 35W with Sandy Bridge, and kept lowering power, without increasing performance much. In fact, that's what the Skylake U chips are. The 28W chips are so rare and expensive that they practically don't exist.

I've watched PC space and benchmarks very closely for more than a decade. Even right after Pentium 4's heat debacle, it was still all about performance. Suddenly, even in the extreme enthusiast category - AKA GPUs, its all about watts. Lowering it.

I'm not sure what you are expecting. Improved perf/W beyond the process improvement. More rendering features. What more do you want?

 

[zinfamous]

Flops/watt only gained traction recently.

Perhaps I should ask you if you are invested(emotionally or literally).

2 full generations after 290X should net you far lower prices, especially when competition is there and you have 10-20% marketshare. I am not saying $200 is bad, but for the sake of technological advancement its mediocre.

No I am not being silly. I am just bewildered by why silicon manufacturers can't pull much performance wise in the PC space. What's the 14nm for? All the benefits are being moved not just to perf/watt, but watt.

It's just like the first company that used MuGFETS(multi-gate FETS), Intel. They took a 35W with Sandy Bridge, and kept lowering power, without increasing performance much. In fact, that's what the Skylake U chips are. The 28W chips are so rare and expensive that they practically don't exist.

I've watched PC space and benchmarks very closely for more than a decade. Even right after Pentium 4's heat debacle, it was still all about performance. Suddenly, even in the extreme enthusiast category - AKA GPUs, its all about watts. Lowering it.

You're still trying to compare AMD's not yet-released mainstream chip with their previously released generation(s) cut-down enthusiast chips.

This is ~half the cost for the same performance and less power usage compared to the top tier class. How does this not meet all of your demands?

Perhaps your irrational rant will be better placed when you have a cut-down Vega to compare to the 390X, you think?

I wouldn't be surprised if they top out their full Vega chip at $500 MSRP following this pricing. Maybe a "limited edition hyper enthusiast" version at 600-650 to stoke the passions of the " only top price matters!" crowd in the interest of trading blows with spurious nVidia marketing.

Imagine low-end Vega chips starting at 300-320 and topping out at $500.

 

[sontin]

How did Maxwell's claimed awesome VR feature worked out?

Much better than AMD's "lead" in VR. :thumbsup:

 

[Dribble]

Perf/watt doesn't matter so much for desktop - you can get away with being less efficient, although if the gap is big enough (e.g. 970 vs 290) it does start to have an effect.

Where it matters more is mobile (i.e. laptops). You simply can't compete if you can't stay close in performance/watt. In that market even Kepler was more efficient then anything AMD had, let alone Maxwell which was far ahead. AMD need to redress this or they are locked out of that market.

Hence the importance of polaris, and clearly as apple are buying it they have had some success particularly with polaris 11.

 

[sontin]

Apple bought Cape Verde und put Tonga into their iMac. Apple doesnt care about perf/watt. They only care about money.

 

[Sweepr]

Weren't there two XBOs announced at the E3, the updated XBOne (s) & then Project Scorpio? If so the final specs of the true nexgen© PS can still be manipulated enough to beat the next gen Xbox, probably the same reason why there was no announcement here. Then there's still the possibility of a minor refresh ala PS4.5

Rumours point to PS4 Neo launching later this year (before October?), so hardware specs are pretty much locked down at this point. Yes, they can raise clocks like MS did to Xbox One's CPU cores close to launch, but that's it. Releasing ~1 year later (Fall 2017 according to The Verge) MS can probably do the same to Project Scorpio to retain their advantage. Exciting times ahead.

Just noticed there's 12 chips on top of the APU:

12/24 GB GDDR5 7 Gbps @ 384-bit?

 

[Glo.]

Apple bought Cape Verde und put Tonga into their iMac. Apple doesnt care about perf/watt. They only care about money.

Tonga in Retina iMac had 125W TDP, and 3.7 Tflops of compute power.

It gives 27.9 GFLOPs per watt. Only GPUs that have higher performance per watt are Fiji series, Polaris 10 and GP 104.

Stop using techpowerup gaming benchmarks as a reference. Efficiency always is counted from gigaflops per each watt consumed.

 

[khon]

Perf/watt doesn't matter so much for desktop - you can get away with being less efficient, although even if the gap is big enough (e.g. 970 vs 290) it does start to have an effect.

Perf/watt is directly linked to perf/$.

If the card is not as efficient, you have to spend more on a larger power supply, and spend more on your electricity bill.

 

[HiroThreading]

Apple bought Cape Verde und put Tonga into their iMac. Apple doesnt care about perf/watt. They only care about money.

Everybody cares about money.

Also, what's wrong with the perf/watt of Cape Verde and Tonga?

 

[flash-gordon]

Has anybody measured the die from RX470 picture? It's P10 or P11?

 

[sm625]

12/24 GB GDDR5 7 Gbps @ 384-bit?

It has to be 12GB. These games have to be backward compatible, which is going to limit RAM usage.

 

[coercitiv]

Apple doesnt care about perf/watt.

Sometimes, in the quest for the favorite brand, pearls like this appear. I can only wonder at such strength of belief.

 

[SPBHM]

It has to be 12GB. These games have to be backward compatible, which is going to limit RAM usage.

are you sure?
they mentioned a million times all Xbox One stuff being compatible with project Scorpio, but I can't remember they saying xbox scorpio games would be compatible with the xbox one,

at this point I think their strategy differs from Sony on this, and that's one reason why they are jumping from 1.3 to 6TF while Sony is "only" a 2.3x increase?

12GB GDDR5 makes perfect sense.