AMD: Speculation, Rumors, and Discussion (Archive)
- Thread starter iMacmatician
- Start date
- Status
Silent_Buddha
Legend
It kind of makes me think that this time around AMD did things the opposite of the Fury lineup. Rather than going well past the knee of the power usage curve, like the Fury X, they are instead launching the product well before the power curve, like the Fury Nano.
I wonder if that means the overclock headroom and performance gain would be somewhat similar to if you started with the base Fury Nano (stock Rx480) and overclocked it to Fury X (partner O/C Rx480 cards) levels. Assuming anyone had released a Fury Nano variant with 6 + 8 power connectors, suitable cooling apparatus, and suitable VRMs, etc. Of course, no one did that for Fury Nano since that was what Fury X basically was.
Or to put it another way. Instead of AMD launching a Fury X version of Polaris 10, they are going to let their partners do it.
Regards,
SB
Yep. On isovoltage, raising clock speeds increases dynamic power linearly. If you need to raise the voltage, the power used by each clock goes up, so power goes up to the second power. If you cannot keep thermals stable, resistance goes up, and this gives you a third-order term.
Also, upping clocks does not increase leakage power, but raising voltages and temps does.
When i like the rest of your post, why will you it "need" 6+8pin ? you can put 250W on a 6pin if you want( and i have push way more than that by the past ) or use 120W on a 6+8pin connectors.. its not because i have a 1000W PSU that my PC need 1000W.
Honestly, if partners goes the road of high clock + OC marketing, they better to goes for the 6+8pin roads too. will justify the premium price ( and marketing ). This said if this "information" have something right. The point is with a margin on power delivery, they will interest the overclockers market, who at their turn are followed by the enthusiast gamers.
Honestly this dont change much, the real question is the quality of the PWM.
Last edited:
Yeah, theoretically, overclocking a chip without increasing its voltage can increase its power-efficiency since it only increases dynamic power, and does so linearly, so the global power increase is less than linear, whereas performance may scale linearly.
In practice, if you can overclock your chip without raising its voltage, it's likely that its original voltage was excessive to begin with.
In practice, if you can overclock your chip without raising its voltage, it's likely that its original voltage was excessive to begin with.
Effectively, and with actual power management you have on cores, let say you have less to care about the ratio "voltage > TDP".. as the voltage will fluctuate all the time. You can set an higher voltage than you will have done by the past ( its exactly what is doing Nvidia ).
I remember the old era, whatever it was on CPU or gpu's, the first tweaks we was apply was increase the voltage, as they was nearly just good for stability on stock. ( i dont tell you what we was doing on DDR1 BH5 and CH6 )
hoom
Veteran
6+8 comes from that guy not me <shrug>
a 1500mhz 40cu part would could tho. Of course we have no clue if such a thing even exists. But that could account for them adding more power to the board.
He also says it's in DX12 titles so from that I read 'not as fast as 1070 in DX11 titles'.
Seems plausible and we don't know if they clocked up memory as well.
It will be interesting to see how the OC translates into actual performance when it is finally released, for the moment it looks like the best for that was previous gen Maxwell 2.
Here is what NordicHardware said when they reached out to their source regarding the WCCF article:
I emphasised performance as I think this is not about the clock but relative actual gains.
The 1.5GHz clock could be a great feat by AMD or just what one expects, depending upon how it has to be done by the AMD and AIB partners.
That OC is roughly 18% above boost.
To put it into context, for a 1070 a comparable clock would be 1985MHz (actual 1683MHz reference), and for the 1080 it would be 2051MHz (actual 1733MHz reference), putting aside the headache of the messed up basic profiles on the FE models this is achievable (when increasing power and temp targets critically along with adjusting fan profile to be more aggressive at 80c and % to 65 at that threshold).
Now consider the 1080/1070 with custom AIB and the issues with the FE mostly go away, albeit critically one still needs to manually make OC changes for the cards to hit that 18% OC - the EVGA Scan function in Precision may help but still not the simplest solution to just reach 18% if it can as more data is needed with the utility to see if it is conservative.
So for the 480 it could be great or structured in a way that it is charged at a premium, or only applicable to certain models to attain this, and maybe easy/difficult to enable, and importantly is there further headroom above 1500MHz (this would be a great feat for AMD against Pascal)
And it is not clear whether AMD is doing this on their reference card or it is down to the AIB partners, so a lot more info is needed to know if it is great or more in line with traditional expectations.
But one aspect that stand out that some has mentioned; the need for 6+8pin.....
I think it is bit of a red herring because the 1070FE can hit with some effort from the gamer the 18% OC with a single 8-pin, while AIB even for Nvidia seem to go 6+8pin without breaking that ceiling by much if at all.
The issue with the 1070FE (and custom) was not power but voltage limiter protection (restricted to below 1.10V).
So if the 1500MHz is the ceiling, seems that the 480 has similar challenges IMO; shame the rumours do not mention if that is the ceiling or further headroom.
Also if NordicHardware is correct, the OC to performance on the 480 is not translating as great as we would like, which we also see happening with the 1070/1080.
Possibly re-inforcing the challenges posed by the node shrink density/physical properties/etc for both manufacturers when operating just beyond the ideal performance envelope.
Cheers
Here is what NordicHardware said when they reached out to their source regarding the WCCF article:
http://www.nordichardware.se/nyhete...480-ska-na-15-ghz-med-vanlig-luftkylning.html
I emphasised performance as I think this is not about the clock but relative actual gains.
The 1.5GHz clock could be a great feat by AMD or just what one expects, depending upon how it has to be done by the AMD and AIB partners.
That OC is roughly 18% above boost.
To put it into context, for a 1070 a comparable clock would be 1985MHz (actual 1683MHz reference), and for the 1080 it would be 2051MHz (actual 1733MHz reference), putting aside the headache of the messed up basic profiles on the FE models this is achievable (when increasing power and temp targets critically along with adjusting fan profile to be more aggressive at 80c and % to 65 at that threshold).
Now consider the 1080/1070 with custom AIB and the issues with the FE mostly go away, albeit critically one still needs to manually make OC changes for the cards to hit that 18% OC - the EVGA Scan function in Precision may help but still not the simplest solution to just reach 18% if it can as more data is needed with the utility to see if it is conservative.
So for the 480 it could be great or structured in a way that it is charged at a premium, or only applicable to certain models to attain this, and maybe easy/difficult to enable, and importantly is there further headroom above 1500MHz (this would be a great feat for AMD against Pascal)
And it is not clear whether AMD is doing this on their reference card or it is down to the AIB partners, so a lot more info is needed to know if it is great or more in line with traditional expectations.
But one aspect that stand out that some has mentioned; the need for 6+8pin.....
I think it is bit of a red herring because the 1070FE can hit with some effort from the gamer the 18% OC with a single 8-pin, while AIB even for Nvidia seem to go 6+8pin without breaking that ceiling by much if at all.
The issue with the 1070FE (and custom) was not power but voltage limiter protection (restricted to below 1.10V).
So if the 1500MHz is the ceiling, seems that the 480 has similar challenges IMO; shame the rumours do not mention if that is the ceiling or further headroom.
Also if NordicHardware is correct, the OC to performance on the 480 is not translating as great as we would like, which we also see happening with the 1070/1080.
Possibly re-inforcing the challenges posed by the node shrink density/physical properties/etc for both manufacturers when operating just beyond the ideal performance envelope.
Cheers
Well we dont have much numbers for game yet, only Firestrike, and we dont know in what condition they have been made. 3Dmark numbers in the databse have this special, that anyone will try got the better score, even at stock speed .
it have allways been a competition benchmark, even if not by direct OC, by remove TDP limit, by remove AF quality, AA, push the fan at max. and what more. I have enough been part of team OC, for know that even a stock speed results are rarely 100% stock .
But even if it slighy slower and match maybe the 1070 in some case ( honestly, i will not push too much what say this guy, i have absolutely no idea if it is true or not ), But the price is still correct.. ( I heard allready the peoples who will goes after the " 1070 can be OC too" . )
I think, the speaking about a different way of think the "entry market level " ( under 300$ ) come more from the fact, you got a 4-8gb gpu's, and who can finally goes on part with middle range gpu's. with decent performance at stock and some OC margin, it could end not as bad.
Maybe the most interessant price ( cost / perf / watts ) could even been the 470 if it clock tooo really high.
Last edited:
When was the last time that a card with a single 6pin power needed another 8pin in order to sustain OC with air cooling? because unless we are seeing the best of the best aircooling system i find it really hard to believe that the huge increase in power consumption doesn't translate in a huge amount of extra heat.
AOTS, performance numbers for Dx12, the 1070 would be pretty far ahead, and that is if we take AOTS numbers from AMD's dual rx480 figures and extrapolate from there.
Hitman is the only one I could see a possible tie. And ROTR, wouldn't be even close.
Last edited:
If they aim to market thoses "special bios OC card" for high overclock etc... it make sense, from a marketing point of view.. Again this dont mean you need it for sustain it. because this is the first thing you hear from overclockers: too bad, only a 6pin.. too bad only 2x6pin.. too bad, this card will be good if it had a second 6pin.. Not all overclockers like to hard mod their gpus.
Now, what could be interessant, is if the clock speed scale welll with voltage increase ( not like the FuryX ) will be too see what this gpu can do with high end watercooling and a big jump in voltage .
At first i was not much interested because i wait the big pascal or Vega. .. But if i can sustain 1500-1600mhz under H20 for 250-280$, i take 2 for my OpenCL renders, along that my other gpu's. should just give me something as a 14Ttflops gain for cheap. When Vega is out, i take 1 or 2 for complete the setup as main gpu#s. ( its not for gaming primary,, but for raytracing )
Last edited:
- Status
