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

[StinkyPinky]

Is there a 480 and 480x or just a 480?

 

[Elixer]

Is there a 480 and 480x or just a 480?

You have to wait until the 29th to find out. Nobody knows yet.

 

[EndlessWargasm]

I suspect that if there is a 40CU Polaris 10, that for the foreseeable future all those chips will be going to Apple and/or mobile parts.

It makes more sense this time around than the wait for full Tonga as AMD has no competition in the mainstream space till GP106 drops.

Once yields come up and the new MacBook Pro has had it's initial run then the non OEM PC will get the 480X.

OR AMD might just be waiting for the 1060 to come out.

We should know for sure once someone takes the top off an RX 480 and does some high quality scans.

 

[Bacon1]

I suspect that if there is a 40CU Polaris 10, that for the foreseeable future all those chips will be going to Apple and/or mobile parts.

It makes more sense this time around than the wait for full Tonga as AMD has no competition in the mainstream space till GP106 drops.

Once yields come up and the new MacBook Pro has had it's initial run then the non OEM PC will get the 480X.

OR AMD might just be waiting for the 1060 to come out.

We should know for sure once someone takes the top off an RX 480 and does some high quality scans.

Why would the high end chips go to Apple? Don't they use the mobile gpus for products? I'd expect them to be all over the Polaris 11 chips not high end 10.

 

[Stuka87]

Why would the high end chips go to Apple? Don't they use the mobile gpus for products? I'd expect them to be all over the Polaris 11 chips not high end 10.

Same reason all the fully enabled Tonga chips went to Apple. The Tonga that was sold at retail was a cut down version.

 

[R0H1T]

Why would the high end chips go to Apple? Don't they use the mobile gpus for products? I'd expect them to be all over the Polaris 11 chips not high end 10.

Apple would definitely want the best bins, be it P10 or P11, & I doubt AMD can deny them that.

 

[Bacon1]

Apple would definitely want the best bins, be it P10 or P11, & I doubt AMD can deny them that.

Why do they want gaming cards?

 

[SpaceBeer]

They dont. They need them for GPU acceleration in foto/audio/video software

 

[R0H1T]

Why do they want gaming cards?

Just gaming cards? I'm fairly certain AMD is the best (PC) GPU vendor when it comes to OpenCL, Apple loves OpenCL since they hatched it in the last decade, also they'll accept lesser margins than Nvidia thus Apple's preferred choice.

 

[Bacon1]

They dont. They need them for GPU acceleration in foto/audio/video software

There isn't much difference though

https://youtu.be/mF9hF9GNL6Y?t=26

 

[EndlessWargasm]

There isn't much difference though

https://youtu.be/mF9hF9GNL6Y?t=26

Not really a valid comparison, and a very narrow one at that. OSX is written with OpenCL in mind, not CUDA.

 

[Sweepr]

The myth of the 2560SP Polaris 10.

I almost want to make a moonbogg bet here that there's no 2560SP Polaris 10 card.

So far, zero concrete leak about this part, and we've had 36 CUs Polaris 10 entries @ SiSoftware for months. Some can say wait and see, but it's looking less and less likely that a 40+ CU version exists.

 

[mohit9206]

I'm sure there will without a doubt be a 480X/485 card with more SPs and CUs.
And ofcourse cut down RX 470($149) and 460($99) as well.
Waiting for the RX460 as that should be a substantial jump from my GT730.

 

[kraatus77]

So far, zero concrete leak about this part, and we've had 36 CUs Polaris 10 entries @ SiSoftware for months. Some can say wait and see, but it's looking less and less likely that a 40+ CU version exists.

iirc we never saw anything but rumors about full tonga. and 2304 shaders seems weird for a full chip.

 

[maddie]

When the RX480X enters the arena, it will be a sensation beyond the norm. I look forward to seeing reactions.

As several have already stated, a cursory examination of P11 [assumed 1280 SP at 115 mm^2] and P10 [2304 SP @ 232mm^2] will lead to this dilemma. Where in the world have the transistors gone?

I think we will find out soon to the dismay of some.

 

[MrTeal]

When the RX480X enters the arena, it will be a sensation beyond the norm. I look forward to seeing reactions.

As several have already stated, a cursory examination of P11 [assumed 1280 SP at 115 mm^2] and P10 [2304 SP @ 232mm^2] will lead to this dilemma. Where in the world have the transistors gone?

I think we will find out soon to the dismay of some.

Out of curiosity, how many transistors do you think P10 will have assuming a 232mm² die size?

 

[Sweepr]

As several have already stated, a cursory examination of P11 [assumed 1280 SP at 115 mm^2] and P10 [2304 SP @ 232mm^2] will lead to this dilemma. Where in the world have the transistors gone?

Are these numbers set in stone? So far I haven't seen anything official from AMD.

I think we will find out soon to the dismay of some.

They better have something faster than RX 480 till Vega, not sure how many Fury cards will be sold from now on.

 

[maddie]

Out of curiosity, how many transistors do you think P10 will have assuming a 232mm² die size?

Within a very small margin of error, I will say double P11. Absolute numbers are irrelevant to my argument.

The video engine will be common to both P10 and P11 meaning that you will have a minimum of double the area for the rest. P11 has a 128bit bus, P10 has a 256bit bus. It is beyond reasonable to have double the space available, equating to double the transistors and not have double the SP units. Look at the block diagram and eliminate the common shared units.

If we don't have 2560 SP.
Where in the world have the transistors gone?

 

[MrTeal]

Within a very small margin of error, I will say double P11. Absolute numbers are irrelevant to my argument.

The video engine will be common to both P10 and P11 meaning that you will have a minimum of double the area for the rest. P11 has a 128bit bus, P10 has a 256bit bus. It is beyond reasonable to have double the space available, equating to double the transistors and not have double the SP units. Look at the block diagram and eliminate the common shared units.

If we don't have 2560 SP.
Where in the world have the transistors gone?

Ah, I see where you are going with that. I agree with you there, though keep in mind we still don't know the die size for P11 and AFAIK are still basing it on people comparing the die to someone's thumb. If their size estimate was off by 15%, that could be the difference there.

Aside, I'm guessing slightly less than Hawaii, probably right around 5.8-6B.

 

[maddie]

Are these numbers set in stone? So far I haven't seen anything official from AMD.




They better have something faster than RX 480 till Vega, not sure how many Fury cards will be sold from now on.

No the numbers are the most believed ones. TBD. Notice I said assumed.

Except for the lowest of low cards for individuals only needing a video outputs, I think between P11, P10, 1070 and 1080, every single 28nm card is obsolete.

If I was AMD I'd want to stop Fury production ASAP, so this is good.

Just like you would never advise someone to buy a 980Ti presently, would you still recommend a GTX960 4GB card over a P10, or a 750Ti over a P11? P11 and P10 demolishes every Nvidia card below the top end. I predict a huge swing in marketshare over the next few quarters.

 

[Qwertilot]

NV not being insane - and very well resourced - they're not going to get anything like a few quarters uncontested before 1050/60 turns up and there's a real fight

With AMD basically giving up on the top x% of the market entirely, they'll have to do very well with P11/10 to keep their overall market share steady. Not that it matters so much, they'll be happy if they can turn a profit somehow.

 

[tviceman]

No the numbers are the most believed ones. TBD. Notice I said assumed.

Except for the lowest of low cards for individuals only needing a video outputs, I think between P11, P10, 1070 and 1080, every single 28nm card is obsolete.

If I was AMD I'd want to stop Fury production ASAP, so this is good.

Just like you would never advise someone to buy a 980Ti presently, would you still recommend a GTX960 4GB card over a P10, or a 750Ti over a P11? P11 and P10 demolishes every Nvidia card below the top end. I predict a huge swing in marketshare over the next few quarters.

People are forgetting the economics behind these chips. AMD's single best selling point for the RX 480 vs. the competition is it's price. It's definitely not faster, and doesn't appear to be more efficient. We all now know it's got performance somewhere around 390x to Fury levels and approximately a 230mm2 die size. Compare P10 and GP104 against Pitcairn and GK104 in terms of performance. GTX680 at release was about 30% faster than HD7870, but now we're looking at GTX1080 being 64% faster (480 at 390x speeds) to 47% faster (480 at Fury Vanilla speeds). Nvidia substantially increased the performance delta between itself and AMD with similar die sizes on finfets vs. first gen 28nm. So now the question becomes this: whats to stop Nvidia from dropping prices of the 1080 and 1070 and releasing a 192-bit GP104 for $250 that still handily beats the RX 480 by 10-15%? Do you think Nvidia is going to sit on it's laurels and let a smaller, cash strapped competitor undermine them? Of course not, and the economics of perf/mm2 clearly show that Nvidia has the upper hand.

At current pricing, I would definitely buy an RX 480 if I didn't absolutely need GTX 1070 or faster performance. But I don't expect Nvidia to sit idle and simply let AMD run wild in the lower priced performance segment and steal gobs of marketshare. GP106 will be out soon and Nvidia will probably release a third GP104 sku to fill in the ~$250 price bracket. I expect very stiff competition in the coming months for my money.

 

[maddie]

NV not being insane - and very well resourced - they're not going to get anything like a few quarters uncontested before 1050/60 turns up and there's a real fight

With AMD basically giving up on the top x% of the market entirely, they'll have to do very well with P11/10 to keep their overall market share steady. Not that it matters so much, they'll be happy if they can turn a profit somehow.

I'm sorry to report but Nvidia still is bound by physical constraints. They can't just wave a magic wand and produce cards.

TSMC will have hard orders for many customers. Unless there is a fair amount of spare capacity, the option to just produce whatever you want whenever you want is imaginary.

If they do release earlier than planned then the numbers will not be there so you can't sell anyhow.

GP106 almost certainly will have 2 GPC and I can't see it reaching RX480 performance levels. The real fight might be with the RX470, which will be even cheaper.

Don't forget when you have a company earning high margins and having the majority of marketshare already, the last thing you want to do is drop prices and margins. Share price falls follow such actions. At this point in time Nvidia needs high margins to keep the share price high giving them a very bitter choice of loosing marketshare or having its shares fall in price.

One other thing, if polaris is as good at mining as previous generations and pascal still inferior, we might witness the creation of a totally new full market segment of crypto-currency. Anyone reading widely will see the migration of crypto to the mainstream, with even Wall Street banks wanting to have their own. A whole new market totally separate from gamers. Just imagine the incredulous situation of AMD beating Nvidia in marketshare with the majority of gamers still on Nvidia.

 

[Sweepr]

Just like you would never advise someone to buy a 980Ti presently, would you still recommend a GTX960 4GB card over a P10, or a 750Ti over a P11? P11 and P10 demolishes every Nvidia card below the top end. I predict a huge swing in marketshare over the next few quarters.

Polaris 11 is not available yet (no show at Computex?) and Polaris 10 won't be for a month or so. Considering all Pascal chips were already taped-out, AMD will have competition sooner rather than later, so I don't believe in a 'huge' swing in marketshare.

 

[ShintaiDK]

I predict a huge swing in marketshare over the next few quarters.

Sounds right, but it wont be what you expect.