AMD Polaris architecture – GCN 4.0

[Steampunk]

What is a pipe cleaner in this context please? (I tried looking it up but couldn't find anything concrete)

It's a smaller, simpler chip pushed through as a test of a new design, or more often when the major jump to a smaller process node happens. Then when your bigger, more complex chips go through the new process, you can have more confidence on things like yields, or to correct any problems that were observed with the simpler product.

It's considered safer than the old idea of launching first with your most complex, top-of-the-range product, and then producing the smaller, cut down versions for the cheaper price segments. Both AMD and Nvidia have been caught out in the past when moving to a new process node with the hardest chip to make coming through the manufacturing process first.

 

[eddyr]

Early Polaris must be targeting either pitcairn or oland level surely. They must have come some way clearing stocks of them with OEM help by now.

 

[FoxEye]

Why did they show the efficiency gains of Polaris at 1080p/ 60 FPS/ medium settings? Why compare to a GTX 950?

Presumably they could have shown the same efficiency gains at 1080p/ 60 FPS/ ultra settings. And compared to a GTX 970.

This makes me suspect that the first products will be 950/960 ballpark performance, just much more efficient in terms of perf/watt.

It's already been rumoured that the Zen APU has the same GFX perf as current gen consoles, so it's safe to say that Zen does not mean that APUs destroy the low-end GPU segment.

So there will still be low-end GPUs, at GTX 950/960 performance, and these could very well be what was demonstrated at CES. Very efficient, very low end.

 

[Mtom]

AMD to Launch Polaris Graphics Architecture with a Mainstream Card

 

[LoadsaMoney]

I suspected that and imagine the same for Nvidia. It would appear as expensive or short supply for them to go that route.

Nvidia will be using the newer GDDR5X (or whatever its called) on their non high end cards.

 

[Dave2150]

I suspected that and imagine the same for Nvidia. It would appear as expensive or short supply for them to go that route.

Unfortunately NVIDIA don't have priority access to HBM2, AMD does. So I imagine we'll see more parts from AMD with HBM2 compared to NVIDIA's lineup

 

[Gregster]

Nvidia will be using the newer GDDR5X (or whatever its called) on their non high end cards.

Yer, good point that.

Unfortunately NVIDIA don't have priority access to HBM2, AMD does. So I imagine we'll see more parts from AMD with HBM2 compared to NVIDIA's lineup

You do know there is more than one supplier of HBM?

 

[humbug]

AMD to Launch Polaris Graphics Architecture with a Mainstream Card

Thus, it is no surprise that AMD’s/RTG’s first 14nm FinFET part is targeting mainstream, high-volume market. Internally known as Baffin, this graphics chip is expected to launch in just two months (March 2016 timeframe – think GDC San Francisco)

Woohoo.

 

[SiDeards73]

So the low end stuff apparently in March, followed by the mid range later in the year, which means we might even see the high end stuff before the end of the year? im holding out for the Greenland stuff, my 290 will keep me going until then, holding off a new monitor also, interested to see what monitors arrive with this new tech as well.

 

[FoxEye]

AMD to Launch Polaris Graphics Architecture with a Mainstream Card

Just as I suspected, then. In AMD nomenclature, "maintstream" is what we'd call "low-end". (Enthusiast > Performance > Mainstream).

Baffin is quite a small GPU targeting the $149.99 price point range, and will square off against Nvidia’s GeForce GTX 950, as compared to in an AMD-sourced video you can watch above. Ellesmere will replace the 300 series, and go against GeForce GTX 960/970, while Greenland will go and hit 980 Ti / Titan X and their Pascal-based successors.

And I was correct about the new card being GTX 950 performance. Sadly

All the signs were there tho.

e: actually the wording of that article is worrying. We might not be getting anywhere near the performance increases you've alll been dreaming about. Massive decreases in power, sure. But then we've heard for ages that 14nn FF was not a process well-suited to high power, high performance applications.

This gen might net us a 20-30% increase and that's your lot. Only it will use 40% less power as well.

 

[Mtom]

Yer, good point that.



You do know there is more than one supplier of HBM?

Thats why Samsung could be good for AMD, GPU and HBM in house. If they can make the interposer, then all the stuff can be made by them,.

 

[Mtom]

Woohoo.

Sounds exciting, but this is not an official AMD statement so hold the horses

 

[Poneros]

Even if things get released early, HBM will still mean supply limitations so I don't expect being able to grab a high-end card any sooner than Winter '16.

 

[D.P.]

On full card perf/watt the Nano won out in most reviews I saw. It uses HBM so NVIDIA will of course get that power saving moving memory tech too, though NVIDIA has been using more complex power management which is where their sudden jump in perf/watt came from which AMD are saying they've now added - of course we know very little about how the two compare.

Overall... I've no idea who'll come out ahead or if it'll be too close to call/situational.



Yeah probably a pipecleaner, seems the sensible but boring option.

The Nano uses hand-picked cores and are under clocked and undervalued big chips. Do the same with a 980TI and you would get better performance per watt than the Nani even with 50% more memory and GDDR5.

 

[D.P.]

Unfortunately NVIDIA don't have priority access to HBM2, AMD does. So I imagine we'll see more parts from AMD with HBM2 compared to NVIDIA's lineup

The most recent rumours suggest Nvidia have priority access to HBM. Pick and choose which rumour you want to beleive in.

 

[Gregster]

Thats why Samsung could be good for AMD, GPU and HBM in house. If they can make the interposer, then all the stuff can be made by them,.

Not sure to what level Samsung are involved? but yer, if they are tucked up with AMD, that would be very good. AMD's experience with Samsungs resources would make a great partnership.

 

[Orangey]

Just as I suspected, then. In AMD nomenclature, "maintstream" is what we'd call "low-end". (Enthusiast > Performance > Mainstream).



And I was correct about the new card being GTX 950 performance. Sadly

All the signs were there tho.

e: actually the wording of that article is worrying. We might not be getting anywhere near the performance increases you've alll been dreaming about. Massive decreases in power, sure. But then we've heard for ages that 14nn FF was not a process well-suited to high power, high performance applications.

This gen might net us a 20-30% increase and that's your lot. Only it will use 40% less power as well.

And tell me, why exactly do you believe some random blog? It's clearly all speculation & plain old made-up waffle based on previous rumours.

You need to develop a sort of sixth sense when dealing with these sites. I have it, and it allows me to sense when an author has done actual real work with real sources and when it's just blogspam.

Theo Valich, formerly of BS News. Nuff sed.

 

[Perfect_Chaos]

This gen might net us a 20-30% increase and that's your lot. Only it will use 40% less power as well.

Sounds pretty lame.. But 30%+ over 980Ti and may consider. Whichever company brings out the best card this gen

 

[Mtom]

The most believable: none of them has priority.

So far AMD seems to get the headstart, already demoing the Polaris, have access to 2 factories for production, and i can imagine they have a combined deal with samsung for gpu-hbm production. We'll see how the actual products compare when they both release them

 

[LoadsaMoney]

The most believable: none of them has priority.

So far AMD seems to get the headstart, already demoing the Polaris, have access to 2 factories for production, and i can imagine they have a combined deal with samsung for gpu-hbm production. We'll see how the actual products compare when they both release them

Well i'll not be falling for the Hype, not after the Titan Killer that turned out not to beat the 980