• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

AMD did say Polaris would have a 2-2.5x increase in perf/W, so wouldn't that put the R9 480 or 480X at around R9 Nano or R9 Fury levels of perf? I'm not expecting them to be that powerful, but it would really shake things up in the industry if they managed it.

yes the gap between 2x and 2.5x is a big gap when we are making assumptions about performance :)
 
Wouldn't surprised me if these were significantly more efficient in terms of performance per shader, but also that most of the info is based around low clocked mobile models. At least till Vega and beyond the biggest area AMD needs to gain market share is laptops. Wouldn't surprise me if we see relatively big cores at very low speeds for efficiency reasons for laptops but at normal clocks on desktop. In general, though not always true, a larger core with lower clocks is the best and most efficient way to performance. Look at Nano vs Fury/Fury X, you use a fully enabled die at lower clocks for the most efficient form of a card.

Nvidia didn't go that way with the 980m they went more with less shaders though they used 30% lower memory clocks for efficiency reasons. The desktop 980 is 165W chip while the laptop version is a 125W chip.

Consider similar for AMD, 100w laptop chip makes it significantly more efficient and better for battery life than a 125W chip, but that doesn't mean the desktop version won't be a 150W chip.
 
What happened to it supposedly being around 15/20% faster than the FuryX, after it was shown running Hitman at a constant 60fps ?

We don't even know what settings Hitman was running on. Someone pulled max out of his a**.

They said DX12 at 1440p thats all 26:15 offset:

 
Last edited:
Now the one on the far right is an PCIe X32 socket.

How many PCIe lanes was that again?

So what settings are required for a current 390 to run Hitman DX12 1440p at solid 60 fps then? Is it possible? As someone else said, I'm not too fussed about the ridiculous low power usage, the performance is the bit that should be exciting.

I also find it a bit funny how one rumour (similar to what's going on over on the Nvidia thread with alleged 1070->1080ti June release rumours) and everybody starts acting like it's been 100% confirmed true. That said, if it's harmless speculation based on said rumours, then fair enough, apologies for misunderstanding things.

Joining in on harmless speculation, whoever said that this 480 card is likely to have 390/390X level of performance, I'm thinking the same as them. FuryX performance at a 380 price is too far a jump, and sounds too good to be true. An awful lot of current cards (including the top-end like FuryX itself) would be very quickly made obsolete by such a release.

The odd thing is that Hitman benchmark. AMD seemed as if they were aiming Polaris 10 for 1440p60 gaming... which seemed strange at first, but I've gone and had another look at those benchmarks. On max settings, a 390 with seems to average 53.5 and the 390X manages 59.1. Thus perhaps AMD weren't too far off max settings with what they showed, perhaps they turned down a few things like anti-aliasing (I know I would) to get a solid 60. It doesn't help that there aren't too many actual benchmarks of Hitman DX12 out there, not even Digital Foundry managed to do one of their framerate analysis on it. Thus I feel like AMD are being fishy with the choice of game and settings they decided to show off Polaris 10 on.

It would be pretty cool if I was wrong about this and the 480 is Fury X level of performance, but I think I'll honestly be content with having 390/390X performance at the 380 price point. While I personally have no interest in such a card performance-wise (look at what I'm currently on), I know some folks who would like this alleged 480. But we are also forgetting that the rumour doesn't mention 490/490X or Fury2 (whatever the successor would be). So the performance of those cards must be made to fit around the alleged performance of the 480. That would put the 490 at FuryX level and the Fury2 who knows where higher.

At this point, I'm not even looking at the 490 level cards, I'm looking straight at the top. One card to outperform them all. Vega can't come sooner enough, I'm excited for that too considering how Polaris suggests it will perform!
 
Do we have any idea on what the names are for the new GPUs as I thought AMD were changing from the standard 4##/##X naming scheme in order to align them with the Fury/Fury X naming scheme or am I getting confused with the series name changes?
 
How many PCIe lanes was that again?

32 :P

The whole point being that the last 16 Lanes could instead be used for direct links between gpu's on the board without breaking compatibility of the first 16 lanes on the port.

Then it means having a bit of a thicker motherboard to route those extra links between the other ports. so you could end up with 4 direct links between 4 x32 ports. Or 16 links between two ports etc.

http://vrworld.com/2016/04/08/nvidia-mezzanine-nvlink-connector-pictured/

Although each mezzanine connector has 400 pins, the majority of them are probably data, but then this is for an 8 way setup.

each pcie link uses 4 lanes, so 64 for a 16 link, so might not be enough for even a 2 gpu setup.
 
Last edited:
How much faster is a fury x over a 390x in all honesty anyway?

Depending on resolution and game engine the fury x is about 5-10fps ahead of the 390x but the slight advantage gap of the fury narrows the lower the resolution.
At 4k a 20-30% performance increase sounds a lot but in reality it's not when you look at the numbers.
 

When is Nintendo supposed to launch?
Because if Polaris 11 is making a push for laptops with a possible of an important Macbook Pro contract etc, and Nintendo are going to get 14nm Polaris type chips, there may not be much left for desktop users.

Getting those contracts menas that all three consoles will have GCN APUs, and hence most big games should be heavily GCN optimised.
So it should be easy for AMD to win benchmarks, however it won't do them any good if they don't have desktop products to sell.
 
In reply to the above post: that's a good thing if they are using Polaris-like GPUs. Hopefully it means that we'll finally have a console that can maintain 60fps at 1080p.

Of course... it's yet another rumour so pour salt on it, especially considering many believe the NX will launch Holiday 2016.
 
When is Nintendo supposed to launch?
Because if Polaris 11 is making a push for laptops with a possible of an important Macbook Pro contract etc, and Nintendo are going to get 14nm Polaris type chips, there may not be much left for desktop users.

new sony too :o
there was that news story they bought fab time from samsung
*shrug*
 
Joining in on harmless speculation, whoever said that this 480 card is likely to have 390/390X level of performance, I'm thinking the same as them. FuryX performance at a 380 price is too far a jump, and sounds too good to be true. An awful lot of current cards (including the top-end like FuryX itself) would be very quickly made obsolete by such a release.

The only confirmation of that I would believe would be stocks running dry on FuryX, no replacements in sight, etc. If AMD start visibly clearing their sales pipeline then you know they're releasing something that obsoletes their old products. If they don't do that, you can make an educated guess that they're almost certainly not.

My expectation is they bring in Polaris in the mid-level and the high-end is phased out after that in preparation for Vega. Remember, a lot of their interest in Polaris is not about performance, but about power consumption for mobile devices. Step one is to get Polaris out there and eating up market share. Claiming the high end is, imo, going to be step two.
 
Back
Top Bottom