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

[VulgarDisplay]

I feel NDA for reviews being end of June means they know they've got a dud on their hands. You see because if they were confident in their product they would lift the NDA as soon as possible so that all those millions of people ready to buy 1070 would read the positive reviews of Polaris and decide to wait for it instead.
With NDA so late most people will have already bought 1070 since no reviews for Polaris that they could find.

Oh oh I can do this too!!!

I feel for NDA for reviews being pushed up to May against what most industry people expected meant nvidia knows they've got a dud on their hands. You see because if they were confident in their product they would not have lifted the NDA with this hilariously and intentionally confusing pricing structure meant to price gouge early adopters who would be buying these cards without seeing reviews for polaris.

With NDA so early, early adopters will already have bought the latest and greatest because nvidia paper launched with limited supply and ridiculously inflated prices for garbage reference cards.

Ok, so you see how easy it is to put negative spin on whatever you perceive is happening. I will personally just wait for reviews from both sides before drawing any conclusions.

 

[Slaughterem]

Already giving excuses for the AMD cards?

Enthusiasts who don't mind spending $700+ aren't the only people on this forum...



This is what I expect from the price/performance crowd, and AMD will have a good price/performance part, but not enough of a deal to make people run out and get it. When price drops/holiday sales come, that's when the price/performance crowd will be interested.

What person who is extremely price/performance conscious is running out to get GPUs on launch day? Even if the 1060 is 2 months late to the party, they're ok.

Wow just wow that you would draw this conclusion from my statement that a poll on an enthusiast site is predictable and not really telling of peoples purchasing decisions.
I am amazed at how much you are in tune with the general public's purchasing habits. We should all thank our lucky Polaris stars that we have individuals like you who have such extensive marketing knowledge available to advise us on these forums.

 

[SteveGrabowski]

Food for thought.

When was the last time AMD had a node jump and a new architecture?

GCN on 28nm, correct?

What's Polaris 10 similar to in size and market target? Pitcairn.

Just take a look at the 7870, compare that against the GTX 580 (Big Fermi) and 6970 (Hawaii for it's time).

I'll make it easy for you.

^ Reference 7850 and 7870. Basically 2 SKUs that Polaris 10 will undoubtedly have something similar. Note, there were custom 7870 that were 10% faster.

I'd love to see that, but nothing about the way AMD has been marketing Polaris makes it sound like that's the case. If it was they should be screaming from the rooftops about previous gen flagship performance for half the price, and they should have reviews out for it before the GTX 1070 goes on sale. Because if they could match what they did with the 7870 I think a lot of people hyped for the 1070 would get Polaris instead.

 

[C@mM!]

I'd love to see that, but nothing about the way AMD has been marketing Polaris makes it sound like that's the case. If it was they should be screaming from the rooftops about previous gen flagship performance for half the price, and they should have reviews out for it before the GTX 1070 goes on sale. Because if they could match what they did with the 7870 I think a lot of people hyped for the 1070 would get Polaris instead.

They have been..... Hell, they demoed the goddamned card earlier this year showing it.

 

[SteveGrabowski]

They have been..... Hell, they demoed the goddamned card earlier this year showing it.

They have been talking about VR for the masses and power efficiency, not previous gen flagship level performance.

 

[ShintaiDK]

In terms of VR, they better have the same or better features than Nvidia. Else that part is way gone. And power efficiency needs to be really good.

 

[trane]

They have been talking about VR for the masses and power efficiency, not previous gen flagship level performance.

After the R600 debacle all they talked about is moving away from big dies and going with a small die strategy. Talking about efficiency, perf/$, hitting the sweet spots, such and such. It's all very familiar.

Yet, the 4870 shocked everyone by convincingly beating the 260 for $100 less and getting uncomfortably close to the 280 at over double the price and die size. Even the 4850 beat the previous gen flagship (9800 GTX). It was ~250mm^2. Same ballpark as Pitcairn, Barts and P10.

Not saying P10 will beat 1070, but it doesn't need to. Just need to deliver solid perf/$.

 

[topmounter]

Usually, when something new comes out, there is a big clear out sale for the last generation stuff, and that is when most people upgrade that I have seen.

Lots of people don't care for the newest, and are still using 1080p or lower monitors, and the last generation cards are pretty darn good at that resolution, and the price is really hard to beat, which is where I would expect you to be looking at.

My current monitor is 1080p/144, my intent is to buy a 27" 1440p FreeSync monitor later this year. The 390x is off the table since it would require a beefier power supply. If I can get ~390x performance at 1440p from P10 for less than $299, then it is worth the wait for me.

 

[sze5003]

I also may be upgrading my monitor but I'll stick to 1080p for now. Just sucks they can't release any info about it now so I can at least decide.

 

[tviceman]

Food for thought.

When was the last time AMD had a node jump and a new architecture?

GCN on 28nm, correct?

What's Polaris 10 similar to in size and market target? Pitcairn.

Just take a look at the 7870, compare that against the GTX 580 (Big Fermi) and 6970 (Hawaii for it's time).

I'll make it easy for you.

^ Reference 7850 and 7870. Basically 2 SKUs that Polaris 10 will undoubtedly have something similar. Note, there were custom 7870 that were 10% faster.

7870 10% faster than the 6970? Cayman was closer to Tahiti in die size than Hawaii. The comparison is very tough to make since 28nm was so prolonged and AMD went bigger than Cayman twice. 7870 also had similar bandwidth to Cayman, while P10 is going to be at a huge defecit vs. Hawaii (let alone Fiji).

P10 is going to be right around 25-30% slower than GP104. If it's closer to GP104 than that, it's an engineering win for AMD. If it's more than 30% slower it'll be another situation where AMD is losing in perf/mm2 and perf/transistor.

I think P10 is going to be in between 390x and Fury at 1080p, but probably drop off a little st higher resolutions and perform at 390x levels.

 

[sweetusernames]

P10 is going to be right around 25-30% slower than GP104. If it's closer to GP104 than that, it's an engineering win for AMD. If it's more than 30% slower it'll be another situation where AMD is losing in perf/mm2 and perf/transistor.

Man I secretly just hope AMD plans to make a 480x GHz edition or something like that just to close that 25-30% even further.

 

[tviceman]

Deletedb

 

[trane]

7870 10% faster than the 6970? Cayman was closer to Tahiti in die size than Hawaii. The comparison is very tough to make since 28nm was so prolonged and AMD went bigger than Cayman twice. 7870 also had similar bandwidth to Cayman, while P10 is going to be at a huge defecit vs. Hawaii (let alone Fiji).

P10 is going to be right around 25-30% slower than GP104. If it's closer to GP104 than that, it's an engineering win for AMD. If it's more than 30% slower it'll be another situation where AMD is losing in perf/mm2 and perf/transistor.

Two things - the jump from 28nm to 14nm FinFet is significantly larger than 40nm to 28nm. There was no major clock speed boost.

Let's skip die size and look directly at number of transistors. Pitcairn was 2.8B transistors, Cayman was 2.65B. Polaris 10 is going to be at least 6B at 232mm^2 if not 6.5B. I'd even wager 7B is possible at 14LPP. That compares 6.2B to Hawaii. Plus, this time, the clock speed is going to increase much more.

As for bandwidth, Polaris 10 is going to feature 3rd gen memory compression while Hawaii had none. As for Fiji, it is quite obvious its bottlenecks were elsewhere - its 4096 SPs were rarely kept fed.

All that said, I agree that P10 will be 25%-30% slower than 1080. But that's still 1070 / Fury X / 980 Ti performance right there. If they can get it down to $300, we have a winner.

 

[tviceman]

All that said, I agree that P10 will be 25%-30% slower than 1080. But that's still 1070 / Fury X / 980 Ti performance right there. If they can get it down to $300, we have a winner.

You are out of your mind if you think 7b transistors is possible. You honestly believe P10 will havr 40% better density than GP104. And keep in mind 30% slower is NOT really that flattering, as it would mean GP104 would be 43% faster than P10.

 

[jpiniero]

You are out of your mind if you think 7b transistors is possible. You honestly believe P10 will havr 40% better density than GP104.

It's possible. The scaling that nVidia got is pretty lousy if you think about it, although it might have been deliberate if it saved development time (and oh by the way allowed for higher clock speeds)

 

[trane]

You are out of your mind if you think 7b transistors is possible. You honestly believe P10 will havr 40% better density than GP104. And keep in mind 30% slower is NOT really that flattering, as it would mean GP104 would be 43% faster than P10.

AMD has traditionally had much better transistor density across all generations. Particularly when new to a process, Nvidia are very conservative.

Though, Nvidia did catch up greatly by the end of the Maxwell generation on the mature 28nm process. Still though, Fiji had 12% greater transistor density than GM200. On top of that, Samsung/GF 14LPP has ~10% greater density than TSMC 16FF+. So 25% greater than GP104 is status quo, really.

That gets us to 6.65B transistors for Polaris 10 at 232mm^2. All I said was 7B is possible - but they probably won't do it as that would offset other gains, in clock speed and power consumption. But 6.5B is a pretty safe bet.

I wasn't thinking straight on the other point. You're right, and if you put it that way, I disagree. I highly doubt 1080 will be 43% faster than 480X. Heck, if all they did was direct die shrink Hawaii to 14LPP, that still wouldn't be the case.

 

[VulgarDisplay]

You are out of your mind if you think 7b transistors is possible. You honestly believe P10 will havr 40% better density than GP104. And keep in mind 30% slower is NOT really that flattering, as it would mean GP104 would be 43% faster than P10.

I don't think 7b is likely, but glofo/Samsung is more dense than tsmc and (10-15% if I remember correctly) and amd's designs are typically quite a bit more dense than nvidia.

It all depends on if they are going for lower clocks with as many transistors as possible or a higher speed less dense design.

It's important to remember that I know next to nothing about this but this is the little information I have picked up on.

Sent from my XT1575 using Tapatalk

 

[flopper]

All that said, I agree that P10 will be 25%-30% slower than 1080. But that's still 1070 / Fury X / 980 Ti performance right there. If they can get it down to $300, we have a winner.

Yup.
Would be the king of all cards the last decade as such

 

[boozzer]

And because it was on the internet it must be true.

just by reading this forum extensively for the last 2 years, yes, it is true, 100%.

 

[trane]

Yup.
Would be the king of all cards the last decade as such

That would be the 4870.

To be better than a 4870, R9 480X would have to beat the 1070 by 10%-15% and get within 10% of 1080. For $299.

By the way, the 480 non-X would have to beat 980 Ti. For $199.

And the 1080 die size be 575 mm^2 and not 315!

Seriously, RV770 is the finest graphics chip this millennium. Good times!

(PS: R300 is a close second. Cypress, G92, there are some other pretty great ones too.)

 

[sam_816]

some one tagged me to read this article at tom's hardware

AMD To Chase Market Share Over High End With First Polaris Offerings

Don't know if this has already been shared on not.
it has some charts and graphs that ppl here love /jk

 

[sze5003]

some one tagged me to read this article at tom's hardware

Don't know if this has already been shared on not.
it has some charts and graphs that ppl here love /jk

404 page not found. At least that's what it says on my phone.

 

[sam_816]

404 page not found. At least that's what it says on my phone.

http://www.tomshardware.com/news/amd-polaris-10-mainstream-gpu,31879.html
this is the direct link. i hope it works

 

[ZZZAAA]

That article doesn't say much since AMD should be releasing Polaris 10 and 11 at about the same time. If Polaris 11 is the affordable budget option (1050/1060) as it should be with AMD previously having compared it to the GTX 950 with energy consumption being the theme of the test, Polaris 10 could target the mainstream performance segment with Vega 10 targeting the high-end enthusiast market. It would be best (at least for me) if Polaris 10 XT would slot right between the 1080 and 1070 and undercut both of the GTX's at price/performance. That being said, I know some AMD fans would want pure performance at any cost to take out the 1080 which seemed unlikely at this point. However, with the recently-revealed NDA being an entire month away, we basically know NOTHING of the Polaris' performance potential. Just my two cents.



Imagine if the sources that stated under $300 (229?/279?) really meant the Polaris 11 Pro and XT with each of the cards being able to outperform or matchup against the 970/980 which would make them feasible for VR? I just don't see any reason for AMD to fall behind NVidia by so much in one generation.

 

[railven]

Food for thought.

When was the last time AMD had a node jump and a new architecture?

GCN on 28nm, correct?

What's Polaris 10 similar to in size and market target? Pitcairn.

Just take a look at the 7870, compare that against the GTX 580 (Big Fermi) and 6970 (Hawaii for it's time).

I'll make it easy for you.

<snip>

^ Reference 7850 and 7870. Basically 2 SKUs that Polaris 10 will undoubtedly have something similar. Note, there were custom 7870 that were 10% faster.

But then factor in price.

Unless AMD is going to drop the price back down to pre-GCN days, $250-350 is very likely still possible.

And if this is +10% over 390X...well it still doesn't paint a good upgrade path for current AMD users, only those still sitting on GCN 1.0 hardware. I personally don't see Polaris 10 hitting Fury levels of performance. Unless we use those games where for whatever reason Fury level hardware was borked. (390X should never have been close to Fury to begin with).






AMD pushing the NDA out so far does strike me as odd. But then again, I don't know what this NDA is actually for. The rumors for this release have been all over the place along with the performance expectations. Remember, 90% of GTX 1080 @ DX12?

I'm just going to get ready to blame GloFo at this point.