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

[JDG1980]

Also, if the embargo is lifted on these AMD slides (the fine print says June 13, 11am PST) then why hasn't the deck been posted by any reputable site? I checked Anandtech, nothing. TPU, nothing. Tom's Hardware, nothing. TechReport, nothing.

 

[Elixer]

Different clock rates (NDA traps, or maybe just different AIB cards with different factory OC) could account for most if not all of these disparate benchmarks. We aren't talking about that huge of a variation in performance, just 10%-20%.

He said that they said it was retail, and last I looked, retail products are pretty much final in terms of clockspeed.
Drivers are a different matter though.

 

[Abwx]

270x doesn't use 180 watts. It uses much less. http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1070/22.html

I find it wonderfull that suddenly AMD s previous GPUs are granted such high efficency, more seriously the 270X use 157W in Anno 2070 but only 134W in BF4...

http://www.hardware.fr/articles/941-5/consommation-efficacite-energetique.html

Whatever the number it s likely that the 470 s 110W TDP will be in the same variabillty, that s 95W at most.

 

[jpiniero]

Why is the 390 -> 390X gap from 79% to 97%? That doesn't line up with the scores above.

It's a typo. It should be 89%, which is what you would expect.

 

[metalliax]

Why is the 390 -> 390X gap from 79% to 97%? That doesn't line up with the scores above.

Edit answered already above me

 

[JDG1980]

He said that they said it was retail, and last I looked, retail products are pretty much final in terms of clockspeed.
Drivers are a different matter though.

Clock speeds are at the discretion of the AIB. Unless they're all reference cards, they could and probably will have different factory clocks. This is typical of any new GPU release.

 

[Abwx]

Clock speeds are at the discretion of the AIB. Unless they're all reference cards, they could and probably will have different factory clocks. This is typical of any new GPU release.

That s surely the plan, to release a very efficient reference line and then let the OEMs sqeeze out the MHz in their custom offerings, i dont think that 1400MHz is even close to the max potential, so we may well have some surprises in the coming month.

 

[Sweepr]

The last few months of hype and the inevitable disappointment:

1. AMD will beat Nvidia to release because it's a new node. (false)
2. Samsung 14nm FF is superior to TSMC FF and achieves 2.5x density, so 230mm2 Polaris should easily match Fury X and will be a disappointment if not. (false)
3. Polaris 10 early benchmarks aren't disappointing because it's a cut down chip. (false)

Now we're looking at RX 480 potentially being slower than GTX 980 and R9 390x, and definitely being slower than an OC'd GTX 980. Man this is turning into a huge disappointment.

Couldn't agree more.

SteamVR puts it below Radeon R9 390 / Geforce GTX 980, and barely above the 'good enough' for VR. Note that they didn't even mention VR for the Radeon RX 470, probably because it's below the minimum required.

Seriously, the best they got out of 232mm² worth of 14nm LPP silicon (more if it were using Pascal's less dense process) is barely matching the old Hawaii (if that)? No matter how you twist it, that's fairly unimpressive. If we look at the competition, 314mm² GP104 is considerably faster at almost half GM200's size.

If you look at perf/watt, their 110W (TDP?) part delivers 80% Geforce GTX 970 performance @ Fire Strike. Problem is, Geforce GTX 970 is a 2-year old 145W TDP part on 28nm planar. GP106 could very well deliver better perf/watt and perf/mm² while matching Polaris 10's performance based on today's results.

 

[casiofx]

Couldn't agree more.

SteamVR puts it below Radeon R9 390 / Geforce GTX 980, and barely above the 'good enough' for VR. Note that they didn't even mention VR for the Radeon RX 470, probably because it's below the minimum required.

Seriously, the best they got out of 232mm² worth of 14nm LPP silicon (more if it were using Pascal's less dense process) is barely matching the old Hawaii (if that)? No matter how you twist it, that's fairly unimpressive. If we look at the competition, 314mm² GP104 is considerably faster at almost half GM200's size.

If you look at perf/watt, their 110W (TDP?) part delivers 80% Geforce GTX 970 performance @ Fire Strike. Problem is, Geforce GTX 970 is a 2-year old 145W TDP part on 28nm planar. GP106 could very well deliver better perf/watt and perf/mm² while matching Polaris 10's performance based on today's results.

Dude why did you worship "half gm200 size" and then blasts polaris 10? It is half the hawaii size

 

[Sweepr]

Dude why did you worship "half gm200 size" and then blasts polaris 10? It is half the hawaii size

Except today's SteamVR score puts it below Hawaii XT, meanwhile GP104 is 31.5% faster than GM200.

 

[moveml]

Dude why did you worship "half gm200 size" and then blasts polaris 10? It is half the hawaii size

Because he loves Nvidia?


Threadcrapping and trolling are not allowed
Markfw900

 

[Silverforce11]

Except today's SteamVR score puts it below Hawaii XT, meanwhile GP104 is 31.5% faster than GM200.

Let me ask you guys a serious question.

Why do you think there's only 1 Polaris 11 based SKU, the RX 460?

Think about it and use your logic.

Then do you really think Polaris 11 is only 1024 SP or 16 CU?

 

[Sweepr]

Let me ask you guys a serious question.

Why do you think there's only 1 Polaris 11 based SKU, the RX 460?

Think about it and use your logic.

Then do you really think Polaris 11 is only 1024 SP or 16 CU?

Seriously you should just give up on your hopes of unannounced SKUs by now. AMD's slide is perfectly clear to me:

Even if your theory was true, I doubt they will release these imaginary products with higher CU count for regular desktops.

 

[Silverforce11]

Even if your theory was true, I doubt they will release these imaginary products with higher CU count for regular desktops.

AMD's slides are PR material mate. You people should know that (they said Tonga was 256 bit bus, we know that's false, it is cut down). It does them no good to admit it's a cut chip they are selling as next-gen.

As for them being released for desktops, I've been saying it won't happen (in the short term) because volume goes to Apple.

Apple sells ~5M Macs per quarter. Even if dGPU options are 20%, that's still a ton of full good chips that AMD have to supply. If you sign a deal to supply Apple and you do not deliver, Apple can sue you for damages.

http://seekingalpha.com/article/3981684-amd-polaris-gpus-will-get-inside-new-macs

And yes, the reason there's only 1 Polaris 11 SKU, is it's the harvested part, 1024 SP. I posted this yesterday that if the SP count is 1024, you can be sure it's a cut chip because earlier SiSoft leak list the ES has 20 CU, 1280 SP. If you think 1024 SP is the full chip, you have to ask yourself: Is AMD selling cut Polaris 11 to Apple? What's the more logical situation here..

 

[moveml]

So with the gains showcased in those leaks (R370X -> RX470 and R380X -> RX480), could you extrapolate the gains of 290x to Vega 10 (which should compete with 1070)?

In addition Vega was stated to have an even higher perf/w, probably due to HBM?

This could be compared to the gains from 970GTX to 1070GTX for relevant metrics (performance or perf/watt) and ballpark who'll be ahead for that battle... Ofc Nvidia could release a 1070Ti as well....

 

[Silverforce11]

So with the gains showcased in those leaks (R370X -> RX470 and R380X -> RX480), could you extrapolate the gains of 290x to Vega 10 (which should compete with 1070)?

In addition Vega was stated to have an even higher perf/w, probably due to HBM?

This could be compared to the gains from 970GTX to 1070GTX for relevant metrics (performance or perf/watt) and ballpark who'll be ahead for that battle... Ofc Nvidia could release a 1070Ti as well....

I think at this time without more info, it will all be extrapolation and it comes down to whether you are an optimist, realist or a pessimist about AMD.

Higher perf/w is due to HBM2 and Vega GCN.

 

[casiofx]

Except today's SteamVR score puts it below Hawaii XT, meanwhile GP104 is 31.5% faster than GM200.

Sorry, after overclocked it is more like http://forums.anandtech.com/showpost.php?p=38275319&postcount=1

So basically you're paying the same price for 23% more performance.

Now take a look at Polaris 10, it is 65% the price of the cheapest 390X ($350) but offers around the same performance as 390X in firestrike. You do need to remember that not all people in the world lives in America (Americans population in the world is only 4.3%), the rest of the world had to pay 390X at MSRP, so Polaris 10 is 53% of the price of 390X ($429).
at almost half the price you get nearly similar performance.

And that is a negative? Wow the amount of negative shilling is very apparent here.

I also don't understand why the situation now is so bad. I mean we will have both high end and mainstream covered by both companies, end users are winning, why complain?

 

[Abwx]

And yes, the reason there's only 1 Polaris 11 SKU, is it's the harvested part, 1024 SP. I posted this yesterday that if the SP count is 1024, you can be sure it's a cut chip because earlier SiSoft leak list the ES has 20 CU, 1280 SP. If you think 1024 SP is the full chip, you have to ask yourself: Is AMD selling cut Polaris 11 to Apple? What's the more logical situation here..

Dunno for P10 but P11 with only 1024 SP doesnt make sense even economicaly as it would cost too much to use a 2304/2560 SP part as basis for say 1280-1536 SP offerings.



There s simply a too big gap between those two chips, that said it s possible given that they are enough to cover most of the DT and and even more of the mobile markets..

This could be compared to the gains from 970GTX to 1070GTX for relevant metrics (performance or perf/watt) and ballpark who'll be ahead for that battle... Ofc Nvidia could release a 1070Ti as well....

Perf/Watt gain would be much more mitigated if people did thought that the 1070 s 8GB GDDR5X use 20W less than the 8GB of GDDR5 of say a 480...

 

[Armsdealer]

If you look at perf/watt, their 110W (TDP?) part delivers 80% Geforce GTX 970 performance @ Fire Strike. Problem is, Geforce GTX 970 is a 2-year old 145W TDP part on 28nm planar. GP106 could very well deliver better perf/watt and perf/mm² while matching Polaris 10's performance based on today's results.

Where are you getting FS results for rx470? I'm seeing on vcz they got ~13,500 3dmark11 points which matches 970 and a top overclock above 16,000 which is above 980 ref

http://videocardz.com/61005/new-amd-radeon-rx-480-3dmark-benchmarks


http://www.guru3d.com/articles-pages/nvidia-geforce-gtx-970-and-980-reference-review,20.html

 

[Silverforce11]

Sorry, after overclocked it is more like http://forums.anandtech.com/showpost.php?p=38275319&postcount=1

So basically you're paying the same price for 23% more performance.

Now take a look at Polaris 10, it is 65% the price of the cheapest 390X ($350) but offers around the same performance as 390X in firestrike. You do need to remember that not all people in the world lives in America (Americans population in the world is only 4.3%), the rest of the world had to pay 390X at MSRP, so Polaris 10 is 53% of the price of 390X ($429).
at almost half the price you get nearly similar performance.

And that is a negative? Wow the amount of negative shilling is very apparent here.

It's not the same price.

$699 1080 vs $649 980Ti. Most custom 1080 are around $699 as well, with some of them even higher. -_-

People keep on using this false MSRP when the cards are selling way above that.

Right now in my region, 1080s are ~$1179 AUD while 980Ti custom models are $899.

Actually you can tell who are shilling by their claims about the price of the 1070 and 1080. Some folks love to lie that it's selling for $379 and $599. A quick check in the USA (where the prices are the best), Newegg, you find they are selling for much higher than that. So why do they keep on repeating the lie?

 

[Sweepr]

Where are you getting FS results for rx470? I'm seeing on vcz they got ~13,500 3dmark11 points which matches 970 and a to overclock above 16,000 which is above 980 ref

http://videocardz.com/61005/new-amd-radeon-rx-480-3dmark-benchmarks


http://www.guru3d.com/articles-pages/nvidia-geforce-gtx-970-and-980-reference-review,20.html

I'm using the only official result provided by AMD so far, 9.090 pts @ Fire Strike.

 

[iiiankiii]

I'll tell you what. If that VR score is real, Polaris 10 is a freakin' monumental fail on a technical level. You have to remember that 14nm @ 232mm, that's basically ~510mm @ 28nm (2.2x 28nm). At ~510mm, that's LARGER than than the 390/x (438mm). How can they make something that is larger and be effectively slower? That is pathetic. It really is. The only saving grace for Polaris 10 is its price. Man, AMD freakin' sucks

edit: on the plus side. Polaris 10 will be in the PS4K. That ensures continue support and optimization for Polaris 10 for the next 2-3 years. VR games will run well on Polaris because of PSVR. I guess that's something positive. The performance of Polaris 10 will only increase. Consoles is saving AMD. LOL.

 

[casiofx]

It's not the same price.

$699 1080 vs $649 980Ti. Most custom 1080 are around $699 as well, with some of them even higher. -_-

People keep on using this false MSRP when the cards are selling way above that.

Right now in my region, 1080s are ~$1179 AUD while 980Ti custom models are $899.

Must be supply issue, basic economics. Nvidia needed to release the cards even when the quantities are limited. This is not even on the same page as jacked up 290/290X's prices because that was hogged by miners.

Oh yeah, 1080 FE at my place costs $850 (including 6% VAT)

 

[Armsdealer]

I'm using the only official result provided by AMD so far, 9.090 pts @ Fire Strike.

http://www.guru3d.com/articles-pages/nvidia-geforce-gtx-970-and-980-reference-review,21.html

970 ref is at 9568 here, or 5% faster


I agree with your broader argument though, assuming that steam VR score is more representative of p10 than what vcz et al are showing.

 

[Glo.]

We know that RX 470 is 67DF:C4 model, right?

http://ranker.sisoftware.net/show_d...a09dac8ae2dfeaccb489b89efb9ea393b5c6fbc3&l=en

It has 36 CU's as it looks like.