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

[IntelUser2000]

Nvidia claimed new circuitry for faster clocks. Well, the percentage increase is the same as claimed for process itself.

One can think of that as needing the new circuitry to take advantage of 14nm FinFET. Without it, it might not have been possible. To the engineers, its "new architecture", but not in a way we think of it.

Back when Intel went to 0.13u from 0.18 with Northwood, they did need some circuitry modifications, but nowhere near extensive as modern day processes require.

End of Moore's Law(the traditional scaling part) approaches not with noticeable stop and panic, but slowly. Like this.

 

[Abwx]

If you ask me, i would trust videocardz results than the one in wccftech. Wccftech's looked too good to be true.

Yet Videocardz got the same score than WCCF s 1015MHz 390, at 1266MHz and 2304 SPs the 480 should be 12% faster if it was a Hawai die shink.

We also know that Tonga brought about 12-14% better perf/MHz in Firestrike, so the 480 should be 25% faster than said 390 in this test, given that WCCF got 2742 for the 390 the 480 should be at about 3400, wich is in line with WCCF scores, Videocardz score look like it was performed at 1080MHz, that is, with boost disabled.

http://www.hardware.fr/articles/926-24/tonga-vs-tahiti.html

http://wccftech.com/amd-rx-480-faster-than-nano-980/

 

[Abwx]

Nvidia claimed new circuitry for faster clocks. Well, the percentage increase is the same as claimed for process itself.

One can think of that as needing the new circuitry to take advantage of 14nm FinFET. Without it, it might not have been possible. To the engineers, its "new architecture", but not in a way we think of it.

It is not, frequency of 16FF+ is stated as being 65% higher than 28nm HP while the tests show that it s 40%, this latter number is the amount that is gained fom their 20nm to 16FF+ according to TSMC...

 

[trane]

Yet Videocardz got the same score than WCCF s 1015MHz 390, at 1266MHz and 2304 SPs the 480 should be 12% faster if it was a Hawai die shink.

We also know that Tonga brought about 12-14% better perf/MHz in Firestrike, so the 480 should be 25% faster than said 390 in this test, given that WCCF got 2742 for the 390 the 480 should be at about 3400, wich is in line with WCCF scores, Videocardz score look like it was performed at 1080MHz, that is, with boost disabled.

That is a pretty solid argument. And indeed, the scores line up well were it 1080 MHz.

The other thing is the chiphell leak, the youtuber leak and wccftech all agree with each other. The chiphell 1080 MHz leak is right on queue with Videocardz. Also, while there's bogus stuff on wccf, Khalid has a decent track record.

 

[PeckingOrder]

http://wccftech.com/amd-rx-480-faster-than-nano-980/

These guys claiming they also have retail samples and not under NDA. They posted on reddit yesterday that today will be a lot of info reveal.

wccf is known for their guesswork. might as well take everything fudzilla publishes for granted.

 

[3DVagabond]

https://translate.google.ca/transla...6/amd-radeon-rx-480-vr-benchmark/&prev=search

So it looks like the graphs are from AMD.

What I said earlier was right. The 14nm and new architecture benefits are moved entirely to lower prices. And the price alone makes up for what's a middling product, in VR, in gaming, in general.

But as a technological advancement its mediocre. It's a 390X shrink to 14nm. It doesn't looks like it has VR specific enhancements like Nvidia had for Pascal.

What hoopla both companies make for die-shrinked chips!

Post-Vega for AMD and Volta for Nvidia is what true next generation architecture GPUs are. Of course, looking at what these "ground breaking" FinFET/TriGate transistors brought for Intel, it's going to look like Haswell. Ok, but not breathtaking as promised.

How do you come to this conclusion?

 

[AtenRa]

People focus too much on the 3D Mark performance lately, I guess real gaming DX-12 performance is not important

 

[DarkKnightDude]

People focus too much on the 3D Mark performance lately, I guess real gaming DX-12 performance is not important

Didn't you know? Nobody actually games. :sneaky:

Everyone uses benchmarks and synthetic benchmarks.

 

[Abwx]

That is a pretty solid argument. And indeed, the scores line up well were it 1080 MHz.

The other thing is the chiphell leak, the youtuber leak and wccftech all agree with each other. The chiphell 1080 MHz leak is right on queue with Videocardz. Also, while there's bogus stuff on wccf, Khalid has a decent track record.

He also published power drain wich he say is about 100W, and this is quite an interesting info...

Given that 8GB GDDR5 use 35-40W the GPU itself is about 60W TDP, in principle its frequency can be raised by 40% at 120W for the GPU (160-170W for the whole card), custom cards should be surprising to say the least, well , we ll know more in two weeks...

 

[ul256]

Probably already said but you can't compare AMD and Nvidia using firestrike. Just looking at the lower videocardz.com results we can tell that the 480 will be ahead of the 390x and 980 in most real gaming. The 390x is about 9% behind the 980 in these benchmarks but the 390x is ahead of 980 in many actual games, and that's just at dx11.

 

[Silverforce11]

He also published power drain wich he say is about 100W, and this is quite an interesting info...

Not at all interesting. AMD's reference card TDP as always is max board power. It has no accuracy for gaming power usage.

http://www.anandtech.com/show/5625/...-hd-7850-review-rounding-out-southern-islands

7850 = 150W TDP
7870 = 190W TDP

I was expecting the same from Polaris 10, full chip will be ~120W, while cut chip ~100W.

Given they kept on hammering perf/w, the clocks will be conservative (a lot of performance left on the table if you prefer that phrase).

1266mhz is pretty low for FinFet. The entire point of FinFet is to reduce current leakage and this comes into play to allow much higher clocks.

I don't expect 40-50% OC like the 7850 and 7950, but 20-25% is realistic, though unless they changed it, one would need vcore access. Hopefully it comes at release if AMD learnt their lesson.

 

[cen1]

If wccftech is right I am building a new system with 480. The only question remains whether to wait for Zen or not. Because that makes it closer to Vega. Hard decisions.

 

[AtenRa]

I have just run AoTS with my R9 390 Catalyst 16.6.1 fix at 1080p High and got 48fps

According to AMD slide, RX 470 had 46fps

But i dont know if we can directly compare the results because of different AoTS versions.

 

[flopper]

Given they kept on hammering perf/w, the clocks will be conservative (a lot of performance left on the table if you prefer that phrase).

That is what everything points to.
A King and the card of a decade.

 

[sirmo]

Yup, and they even hinted at this in one of their presentations. By saying AIBs are free to get more creative, where power consumption sweet spot isn't the focus.

 

[Silverforce11]

http://www.nordichardware.se/nyhete...r-battre-r9-nano-och-overklockat-gtx-980.html

^ These guys just confirmed the Wcftech article, saying RX 480 does have a base clock of 1080 and a boost clock of 1266mhz. Interesting.

"Enligt uppgifterna ska Radeon RX 480 arbeta vid en basfrekvens på 1 080 megahertz medan boostfrekvensen går upp til 1 266 megahertz, siffror vi även kunde bekräfta själva på Computex i Taipei förra veckan."

"According to the data shall Radeon RX 480 operate at a base frequency of 1080 megahertz, while boosting the frequency goes up til 1266 megahertz, numbers we could even confirm themselves at Computex in Taipei last week."

That means the Chiphell leaker on youtube who first raised 1080mhz clocks (his sample didn't boost for some reason... non-final bios?) is accurate. He manually cranked it to 1400mhz.

 

[Abwx]

1266mhz is pretty low for FinFet. The entire point of FinFet is to reduce current leakage and this comes into play to allow much higher clocks.

Finfets main advantage is that they can convey the same current at much lower voltage, 80% of the improvement come from this parameter.

Leakage come as a second advantage only, for instance we can estimate that when fully loaded a 290/290X use about 200W for the GPU and that 70W are lost in leakage..

With GF s 28nm SLP the value would be 50W for leakage while for 14nm LPP it would be 15W.

 

[PeckingOrder]

ok boys, I'm on the hype train

sold my 4-year-old HD 7850 2GB for 65$. RX 480 8 GB better be available on the 29th or my i5-3450's iGPU will drive me crazy. Should be a worthy upgrade for me no matter what.

 

[Glo.]

http://www.nordichardware.se/nyhete...r-battre-r9-nano-och-overklockat-gtx-980.html

^ These guys just confirmed the Wcftech article, saying RX 480 does have a base clock of 1080 and a boost clock of 1266mhz. Interesting.

"Enligt uppgifterna ska Radeon RX 480 arbeta vid en basfrekvens på 1 080 megahertz medan boostfrekvensen går upp til 1 266 megahertz, siffror vi även kunde bekräfta själva på Computex i Taipei förra veckan."

"According to the data shall Radeon RX 480 operate at a base frequency of 1080 megahertz, while boosting the frequency goes up til 1266 megahertz, numbers we could even confirm themselves at Computex in Taipei last week."

That means the Chiphell leaker on youtube who first raised 1080mhz clocks (his sample didn't boost for some reason... non-final bios?) is accurate. He manually cranked it to 1400mhz.

I suppose the benchmarks that Videocardz obtained were for RX 470, not 480.

 

[DDH]

Heres hoping that AIB boards with high clocks and more power delivery are available 29th also

 

[cytg111]

ok boys, I'm on the hype train

sold my 4-year-old HD 7850 2GB for 65$. RX 480 8 GB better be available on the 29th or my i5-3450's iGPU will drive me crazy. Should be a worthy upgrade for me no matter what.

Confirm something akin to SMP and I am on the 8GB 480 too for my Vive .. 230$'s gonna be hard to beat.

 

[prtskg]

AMD will be conservative on clocks so that just by increasing clocks they can rebrand the polaris chips next year.

 

[Glo.]

Actually If you think about this. Intel offers 35W quad core CPUs. Combined with P10 which might have gaming load of 100W of power it would end in very efficient and quiet desktop computer.

I quite like this idea.

 

[Qwertilot]

Or the 1050/60, but yes, it is a really quite attractive sort of idea.

The other thing is that the Skylake 35w quads are actually quite fast nowadays! Good clock speeds/boosts. Would be rather surprised if they bottlenecked anything significantly.

 

[Glo.]

Or the 1050/60, but yes, it is a really quite attractive sort of idea.

The other thing is that the Skylake 35w quads are actually quite fast nowadays! Good clock speeds/boosts. Would be rather surprised if they bottlenecked anything significantly.

Not in DX12 .