Polaris refresh?

old gamer · 15 Oct 2016 at 00:15

arc@css said:
4870 / 4890?

Yep and I bought em both

r7slayer · 15 Oct 2016 at 01:27

LoadsaMoney said:
It doesn't sound like the 480/70/60 will be refreshed from reading that.

well if not then xfx have managed something quite good. Near 1500Mhz and under 150 watts and on air too. Was just tailing a fury card.

genericjim101 · 15 Oct 2016 at 01:57

r7slayer said:
well if not then xfx have managed something quite good. Near 1500Mhz and under 150 watts and on air too. Was just tailing a fury card.

Are you going off that same golden sample Jayztwocents was given?

nashathedog · 15 Oct 2016 at 07:19

DarrenM343 said:
If a Polaris refresh happens so soon that would surely be indicative of an AMD rush job on the first Polaris cards?

Definitely, They needed to get something out the door so they did.

genericjim101 said:
Are you going off that same golden sample Jayztwocents was given?

It'll be interesting to see how others fare with the same card.

DarrenM343 · 15 Oct 2016 at 07:53

nashathedog said:
Definitely, They needed to get something out the door so they did.

It'll be interesting to see how others fare with the same card.

Agree. My comment was a bit tongue in cheek as I've read so many others talking about the Nvidia 10 series being a "rush job". Yeah, right......

Looking forward to early next year when hopefully new cards from both sides, a least a refresh anyway

Rroff · 15 Oct 2016 at 08:31

Not sure its a rush job as such if what I think is the case is true AMD were forced to make a choice of making the best of what the process could do or waiting the ~6 months of development it would take to work out a certain issue with 14nm.

Panos · 15 Oct 2016 at 08:37

DarrenM343 said:
If a Polaris refresh happens so soon that would surely be indicative of an AMD rush job on the first Polaris cards?

Point the finger to Nvidia mate and the rush job of the 1080, with the power & voltage limiters imposed to it also.
No other chip EVER started throttling from 32C

Not to AMD.

Rroff · 15 Oct 2016 at 08:50

Panos said:
Point the finger to Nvidia mate and the rush job of the 1080, with the power & voltage limiters imposed to it also.
No other chip EVER started throttling from 32C

Not to AMD.

Part of the problem there is that the chip is pushing way past the kind of clockspeeds 16nm FF+ was expected to support - ostensibly anything over ~1.86GHz is expecting a lot at ~300mm2.

AllBodies · 15 Oct 2016 at 08:58

Panos said:
Point the finger to Nvidia mate and the rush job of the 1080, with the power & voltage limiters imposed to it also.
No other chip EVER started throttling from 32C

Not to AMD.

Eh?

Sure there seems to be some voltage/power limit causing a brick wall for clocks around 2114 MHz, no matter the board/manufacturer. But what are you smoking saying it throttles from 32C?

The standard throttle temp is ~82C, and upping the power limit increases that to ~92C.

Rroff · 15 Oct 2016 at 09:02

Not sure where 32C came from but some people have shown better results if you can keep it below 42C.

For some reason my 1070 always thermal throttles once it goes over ~80C even if I increase the power/temp target but with a slight voltage bump can run mostly ~2100MHz with the odd drop to 202xMHz due to design power.

queamin · 15 Oct 2016 at 10:15

Rroff said:
Part of the problem there is that the chip is pushing way past the kind of clockspeeds 16nm FF+ was expected to support - ostensibly anything over ~1.86GHz is expecting a lot at ~300mm2.

But in away it worked out for Nvidia it made AMD push the 470/80 away over the sweet spot and that is why they use so much power.

added- I think Nvidia took AMD by surprise clocking so high.

Rroff · 15 Oct 2016 at 10:23

queamin said:
But in away it worked out for Nvidia it made AMD push the 470/80 away over the sweet spot and that is why they use so much power.

added- I think Nvidia took AMD by surprise clocking so high.

As much as anything I think it is that they hit the same problem TSMC did at an earlier stage of development where as you approach 300mm2 for some reason power efficiency, etc. was going out the window - I'm convinced given all the aspects of the 480 that they hit the same kind of thing with 14nm but being behind TSMC (despite claims to the otherwise by certain people) were forced to run with it or delay a long time - once that has been progressed past should see a lot better results and much higher clock speeds coming out of AMD parts.

queamin · 15 Oct 2016 at 10:36

I have a feeling that they based early estimates off Samsung not GF and paid the price for it so to speak thinking it would be about the same stage as Samsung.

Rroff · 15 Oct 2016 at 10:43

Samsung introduced a new variant (can't remember what combination of L, P and/or C, H or E letters it is off the top of my head) to address concerns from those making bigger high performance parts too late for AMD to have it running for the RX480 as well which doesn't help.

EDIT: Oh its still called LPP at Samsung - GF apparently has it unofficially named "LPP+".

Nasher · 15 Oct 2016 at 10:46

Rroff said:
Not sure where 32C came from but some people have shown better results if you can keep it below 42C.

For some reason my 1070 always thermal throttles once it goes over ~80C even if I increase the power/temp target but with a slight voltage bump can run mostly ~2100MHz with the odd drop to 202xMHz due to design power.

Yea you can't change it on Nvidia cards very easily even though most cards are rated for 90c. On AMD you can with a driver option, which is nice.

r7slayer · 15 Oct 2016 at 10:49

genericjim101 said:
Are you going off that same golden sample Jayztwocents was given?

It's more than a "golden sample" Getting lucky with the silicon lottery doesn't normally yield 40-50 watts less power usage.

Rroff · 15 Oct 2016 at 11:01

r7slayer said:
It's more than a "golden sample" Getting lucky with the silicon lottery doesn't normally yield 40-50 watts less power usage.

Samsung made some adjustments for high performance use after GF started working with their 14nm process - looks like GF now either has that or produced their own equivalent unofficially badged as "LPP+".

AllBodies · 15 Oct 2016 at 11:53

r7slayer said:
It's more than a "golden sample" Getting lucky with the silicon lottery doesn't normally yield 40-50 watts less power usage.

Rroff said:
Samsung made some adjustments for high performance use after GF started working with their 14nm process - looks like GF now either has that or produced their own equivalent unofficially badged as "LPP+".

Yeah but isn't that readout from JayZ's video the GPU die power only? i.e. no memory/memory controller.

So a GTX 1080 reading the same thing would say ~120W.

Same reason why AMD has occasionally quoted the Polaris 10 power consumption as 110W, as opposed to the 'typical board power' being 150W.

humbug · 15 Oct 2016 at 14:32

AllBodies said:
I believe this is a misunderstanding of Afterburner's readout.

As far as I know that number is the Watts drawn by the GPU die only, so no memory/memory controller power draw.

AMD have also quoted the RX 480 as drawing 110W a couple of times in a slightly cheaty way, because that's what the GPU itself draws.

The point is this 480 is pulling 40 Watts less than others, even with a mild factory overclock.

AllBodies · 15 Oct 2016 at 23:04

humbug said:
The point is this 480 is pulling 40 Watts less than others, even with a mild factory overclock.

Is it though? I cannot find a Jayz video where he's tracking the same reading on a reference (or any other) RX 480. So there's no identical comparison.

What I'm saying is if I'm interpreting that reading correctly then 100-110W is what you'd expect from a normal RX 480. So that XFX card isn't consuming less.