AMD Polaris architecture – GCN 4.0

[Mauller]

TBH we don't know the performance of 'Baffin', they were just giving us a power consumption demo for a game scenario running at the same quality settings on a 950 and the EARLY 'Baffin' engineering sample running on early drivers.

they didn't state once if either card was running flat out. just giving a power consumption comparison for equal ingame performance.

 

[Boomstick777]

TBH we don't know the performance of 'Baffin', they were just giving us a power consumption demo for a game scenario running at the same quality settings on a 950 and the EARLY 'Baffin' engineering sample running on early drivers.

they didn't state once if either card was running flat out. just giving a power consumption comparison for equal ingame performance.

It's fairly safe to assume the performance level of the card they demoed is similar to the competitor card it was demoed with (GTX 950).

AMD were literally showing that level of performance in a much lower power envelope.

We could see other midrange cards launch early with GDRR5, not just this particular card. There might be a few different lower - mid end offerings. Just not the HBM 2.0 beasts until later on this year.

One of these low power gaming cards would be spot on for my boys MITX, anything that runs cool / quiet and uses little power is ideal. Looking forward to seeing Nvidia's answer to this segment as well. The replacements for the 750 Ti / 950.

 

[AlamoX]

It's fairly safe to assume the performance level of the card they demoed is similar to the competitor card it was demoed with (GTX 950).

AMD were literally showing that level of performance in a much lower power envelope.

We could see other midrange cards launch early with GDRR5, not just this particular card. There might be a few different lower - mid end offerings. Just not the HBM 2.0 beasts until later on this year.

One of these low power gaming cards would be spot on for my boys MITX, anything that runs cool / quiet and uses little power is ideal. Looking forward to seeing Nvidia's answer to this segment as well. The replacements for the 750 Ti / 950.

not really safe to say , fps was capped, how would you know if the fps unlocked the polaris wouldn't be running at 200 fps

 

[Scougar]

And vice versa.

We saw what they wanted us to see and very little else.

 

[Mauller]

And vice versa.

We saw what they wanted us to see and very little else.

Except the 950 is on the market now and people can bench it themselves in the same scenario. The Polaris is still an unknown.

 

[drunkenmaster]

Again really the major point was it was aimed at guys like Dell, HP, Apple. A card that can achieve 60fps at certain settings is just a performance point those guys would want to fill with one of their systems. If a system can use 86W instead of 140W that is cheaper mobo, cheaper psu, less cooling fans, smaller case they can get away with. Meaning bringing that performance to a cheaper bracket or making a bigger profit.

The power was very impressive regardless. The 950gtx is rated at 90W, leaving about 50W for cpu load which likely won't be anywhere near full whack. That means Polaris is using somewhere in the 35W region of power. Just over 1/3rd of the power for the same performance is incredibly impressive. But this is again why you can tell who the slide is really aimed at... quite aside from showing it to professionals at a professional trade show.

For gamers you would compare those as 90W vs 36W, for Dell you say 140W vs 86W.

 

[pmc25]

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

5 factories. 4 Samsung, 1 GF.

Btw, AMD also confirmed that Polaris is 14nm only. I.E. Samsung & GF only. No TSMC.

Re: the continued discussion about memory type. If there are 4 new GPU sizes. I'd expect the first 2 to use GDDR5, the next one to use HBM1, and what was or is Greenland (big boy) to use HBM2. If there's enough HBM2, then both using HBM2.

 

[nashathedog]

not really safe to say , fps was capped, how would you know if the fps unlocked the polaris wouldn't be running at 200 fps

I wondered that, It could be a much higher end chip running capped or undervolted or down-clocked, who know's? Basically we haven't got a clue and as good as it looks compared to a current gen Nvidia I can't see there being anything in it when compared to there equivalent die shrink.
It tells and shows us nothing we aren't already expecting to see from both sides.

 

[GoogalyMoogaly]

Yeah, I think it's hard to judge if this is impressive or not at the minute.

We're comparing a 14nm HBM2 card that's not out yet (so can't be independently tested) against a 28nm GDDR5 card. How much of that power saving comes just from the move to 14nm and HBM2 rather than anything AMD have done with the Polaris architecture?

We also don't know at what point in the lineup the Polaris card is. Is it high end, is it med range is it the 5450 equivalent?

It also seems to be running Battlefront as the test, which I believe favours AMD cards so if both cards are being capped to 60fps then it should take less power from the AMD card to hit 60fps.

Also, looking at the drivers the Nvidia drivers are now out of date, although we (or at least I) don't know when the slides were produced so they may have been current at the time.
The AMD drivers however seem to be 16.10 betas. As we're currently on 16.1 (hotfix) who knows what's going on there.

I think it's a good sign of things to come, but I'm hoping that this is actually just because of 14nm and HBM2 so that we can expect this sort of thing across the board from both vendors. If 90% of this is architecture and 10% is the move to HBM2 and 14nm then I'm worried 14nm and HBM2 may be overhyped.

 

[TheRealDeal]

Yeah, I think it's hard to judge if this is impressive or not at the minute.

We're comparing a 14nm HBM2 card that's not out yet (so can't be independently tested) against a 28nm GDDR5 card. How much of that power saving comes just from the move to 14nm and HBM2 rather than anything AMD have done with the Polaris architecture?

We also don't know at what point in the lineup the Polaris card is. Is it high end, is it med range is it the 5450 equivalent?

It also seems to be running Battlefront as the test, which I believe favours AMD cards so if both cards are being capped to 60fps then it should take less power from the AMD card to hit 60fps.

Also, looking at the drivers the Nvidia drivers are now out of date, although we (or at least I) don't know when the slides were produced so they may have been current at the time.
The AMD drivers however seem to be 16.10 betas. As we're currently on 16.1 (hotfix) who knows what's going on there.

I think it's a good sign of things to come, but I'm hoping that this is actually just because of 14nm and HBM2 so that we can expect this sort of thing across the board from both vendors. If 90% of this is architecture and 10% is the move to HBM2 and 14nm then I'm worried 14nm and HBM2 may be overhyped.

Well unless it's been said otherwise it's a gddr5 card as was said pages back. So the power saving is just down to the process change and new architecture.

 

[Dave2150]

Yeah, I think it's hard to judge if this is impressive or not at the minute.

We're comparing a 14nm HBM2 card that's not out yet (so can't be independently tested) against a 28nm GDDR5 card. How much of that power saving comes just from the move to 14nm and HBM2 rather than anything AMD have done with the Polaris architecture?

We also don't know at what point in the lineup the Polaris card is. Is it high end, is it med range is it the 5450 equivalent?

It also seems to be running Battlefront as the test, which I believe favours AMD cards so if both cards are being capped to 60fps then it should take less power from the AMD card to hit 60fps.

Also, looking at the drivers the Nvidia drivers are now out of date, although we (or at least I) don't know when the slides were produced so they may have been current at the time.
The AMD drivers however seem to be 16.10 betas. As we're currently on 16.1 (hotfix) who knows what's going on there.

I think it's a good sign of things to come, but I'm hoping that this is actually just because of 14nm and HBM2 so that we can expect this sort of thing across the board from both vendors. If 90% of this is architecture and 10% is the move to HBM2 and 14nm then I'm worried 14nm and HBM2 may be overhyped.

I suggest you read more of the articles - this was a GDDR5 card being tested. No HBM

 

[GoogalyMoogaly]

I suggest you read more of the articles - this was a GDDR5 card being tested. No HBM

Didn't see a link to an article (still don't), sorry.
Still, my points still stand, minus the HBM2 stuff.

 

[tommybhoy]

Didn't see a link to an article (still don't), sorry.

Look it up then, are your G skills that rubbish?

1. www.google.com

2. Type 'Amd Polaris' into the search bar and press enter

It's simple stuff, but to save you the bother, it reduces current gpu power consumption substantially, but apparently it wasn't important then important if you run Nvidia, swap it round for AMD.

 

[flopper]

Look it up then, are your G skills that rubbish?

1. www.google.com

2. Type 'Amd Polaris' into the search bar and press enter

It's simple stuff, but to save you the bother, it reduces current gpu power consumption substantially, but apparently it wasn't important then important if you run Nvidia, swap it round for AMD.

It happens when people are lazy and cant read up on the current information and only want to confirm their own confirmation bias.

Its a world class leading power save card demonstrated by AMD, kicking the snot out of Nvidia and then someone cant believe that due to them thinks they know better than the people creating the card and demonstrating the card for the world.

Nvidia can put woodscrews and demonstrate a card that did not exist!
The CEO lied to every shareholder and every consumer on stage for the world. without even blinking.
You can google that also.

The future is brighter with AMD Polaris.

 

[stephenb]

Hmm....the words pot, kettle and black spring to mind. As others have stated you can't draw any conclusions about how good Polaris is until we see Nvidia's equivalent on the same process node.

 

[FoxEye]

The future is brighter with AMD Polaris.

Why are you ending your posts with this crap? Do you work for AMD or are you just trying to show off your AMD colours?

It's just as stupid as ending every post with....

nVidia. The Way It's Meant To Be Played











(which I'm not going to do because it would be ******* stupid)

 

[andybird123]

Hmm....the words pot, kettle and black spring to mind. As others have stated you can't draw any conclusions about how good Polaris is until we see Nvidia's equivalent on the same process node.

its been said before but I'll say it again - TSMC quote a 70% power reduction going from 28nm to 16FF+, that is already 3x perf/w, and as Boom points out - we don't know if in this demo they were also running frame limiters and low voltage to deliberately over-play the "efficiency"

 

[JediFragger]

Go easy of Flopper Fox, I think he's a little simple (and foreign to boot!)

 

[humbug]

I think both sides are going to do quite well in terms of performance per watt, maybe people wanting to continue on with power consumption wars on 14/16nm GPU might look a little silly.

IMO they do now.....

 

[Mtom]

TBH even if Nvidia maintains the lead a step ahead of amd in power consumption, the AMD cards will be efficient enough to stay in the cool and quiet range, no more power hungry volcano beast jokes could be thrown around.
The general opinion was that the power consumption is not a problem because of the few £ a year bill, but because of the heat output, so even if NV takes one step further on efficiency (we don't know yet if they will or not), what amd brings will be enough to create great cards.