AMD Radeon finally back into laptops?

without boost clocks, I'm sure it can down that far, if not farther if down clocked, with boost clocks the 1050ti stays around 70 watts.

If you lock clocks at base clocks on a 1060 it going down to 65 watts.... a 1070 is down to 75 watts.

They can cut down 50% power usage just by locking clocks (more specifically voltage so boost clocks can't be used)

They don't need to down clock a 1050 or 1050 ti to get to where a r460 is at when down clocked by 25%.

I'm also pretty sure this is why the mobile Pascals seem a bit slower than the desktops, other laptops are still limited by a lower TDP then what the desktop Pascals can do so they don't boost as much.

Then factor in binning for lower voltage chips, there ya have it. Even gaming laptops have a max on their TDP. For a 1060 gaming laptop I would expect a 70watt TDP, no more. 1070 laptop probably 105 watts.

 

The discrete card at base clocks 1508MHz (something like that anyway) in game uses 62W, at 2050-2100MHz it uses 120W, higher figure 130W-140W more to do with custom models and top clocks (both memory and core).
So can extrapolate from that, and that is for the full discrete 6GB card.
Cheers

 

They are guessing, they don't know, other websites have stated 65 and 75 watts.

And hell yes it can get to 35 watts, just search for voltage locks for the 1060 and 1070, people have done them on other forums, they are getting the 50% power savings I have been saying.

This is a moot point any ways, because we know Pascal kills GCN Polaris in perf/watt, you shouldn't even question that a 1050 or 1050ti could get down to 35 watts and still have better performance the question is how much more performance will it have and that end result is most likely retain the base clocks, no boost

Now if we want to have some data behind this, Temps of the 1050 and 1050ti don't increase unitl the core starts boosting

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1050-ti,4787-7.html



Just looking at the temps, they increase 35 degrees C, that alone tells us they can save 35 watts or so if they stop the card from boosting (seriously overclocked card here), so just to point out temps are playing a pretty important roll with power usage when this card is overclocked.

So its regular boost without it the temps will save 10 watts or so, then take into the effect of the voltage change, which does increase power usage considerably more then the temps, you are looking at alot of savings just by disabling the voltage for boost.

 

You don't seem to understand how leakage works and what happens with increased voltages with the boosts. I suggest you look up things like that when you have free time before spouting about EE as if it couldn't be figured out with math.

Are you guessing on that? Cause I don't see the data showing that.........

I think you should search for them. You kinda put your foot in the fire here with pretty absurd comments, one of them is extremetech, so go look

I think it can do it, just that the bios isn't letting them do it, We can see the cooling for most of these 1050ti's are not enough to reach 2k. We can see overclocking without the 6 pin attached they still get to 1750 mhz, which means at 75 watts max board supply they have a butt load of room to push power.

Oh I think you missed the razor blade ultrabook then........ Yes they did.

 

You stated 750ti, the 750ti that were used in notebooks, had TDP's around 35 to 45 watts

Btw You were the one that stated you thought nV can't get to that power rating.

If you want me to quote it here ya go

Don't sit here and try to redirect your line of thought on to others that directly was associated with the what you call derailment now.

The rest of it, I did no such thing I just stated you need to look some stuff up before you post, because you haven't looked for the information before. If I were to do something like that, ya know and everyone knows I have no qualms and calling out what a person truly is by using the exact words, and I don't see you posting like that so I didn't.

http://www.extremetech.com/gaming/1...nt-amd-and-nv-graphics-cards-by-a-huge-margin

GTX 800 series (performance chips) were using GM107 chips, not GM108 which are the ones talked about at extremetech the one I just linked to.

The lower end GTX 800 think they were GTX 840 and 830 were the ones that were GM108 chips.

https://en.wikipedia.org/wiki/Maxwell_(microarchitecture)

 

The radeon.pro page has a (partially obscured, off-angle, limited area in focus) visual representation of what appears to be the 460 die.

Assuming it's representative:
Without doing any visual manipulation, does it seem like the CUs were engineered to be narrower?
The SIMD portions are not symmetrical in these CUs, unlike Polaris 10 and others, but like some of the APU implementations like Carrizo. The inner SIMD section looks to be more heavily reworked.
Various other blocks seem to have been rotated and various SRAM blocks stacked along the length of the CU rather than width.

 

You know your point is bloody stupid and deliberately baiting/twisting my information.
I think you like just arguing with me, just like you insisted the Nintendo Switch was not Nvidia and was most likely DMP.

By extrapolate it means look at the figure at base clocks/0.8V and go from there in a more sensible way than the wrong figure quoted by PCGamer, not in your pointless example of -5W/850MHz and a voltage figure that is stupid, there is a limit to how stable it will be at minimum voltage values and yeah it is obvious the linear behaviour is only within the silicon-node optimal performance envelope, and you read enough of my posts to know I go on about that a lot in various threads.
If you have the choice of the figure you use that has no basis within measurements, or an actual measurement benchmark albeit for the 1060 6GB discrete that was analysed for volt-frequency-watts-performance from 0.8V to 1.1V I know what I prefer.
So your using as fact the made up estimate from PCGamer to make a point and still want to argue about it....
Case in point the article you use says

But instead of arguing and posting a baiting/sarcastic response, you would know that at 1508MHz it ONLY USES around 62W when measured accurately.
I have posted many times the Tom's Hardware analysis and ironically you also argued briefly against that in the past.
Here is part of it again, and please cut back on the deliberate baiting/taking my posts out of context.



And then they checked their analysis with gameplay trend analysis as well, anyway far more concise and accurate to the figure you want to argue about used by PCGamer: http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1060-pascal,4679-7.html

I guess you want to argue with Nvidia that they are wrong stating their P4 GP104 (comparable to 1080 architecture) Tesla is spec'd at 75W with 5.5TFLOPs FP32, and yeah that has a clock range of 810MHz-1114MHz (75W is the top clock not base).
Anyway out of this discussion now, but I would not use the PCGamer figure, and would use the Tom's analysis to extrapolate the watts, and yes I assume all here have common sense to do that with reason.
Cheers

 

Yeah pretty familiar with wafer thinning, Intel has been using it for quite some time now. This is nothing new.

Well you are guessing, when there is information where you don't need to guess, and that is what CSI_PC and I have been saying.

For a person that is saying a quality test engineer should follow more than one thing, its a hell of a lot better then guessing in the first place isn't it?

 

That seems to involve a manufacturing step more concerned with the Z axis of a chip, perpendicular to the plane the CUs are on.
Perhaps there are considerations for the implementation that can influence this to a limited extent, if for example there are mechanical reasons for specific choices.
With die thinning the big win would be shaving (ed: mechanically grinding) the non-patterned side of the wafer, which makes up most of the height.

 

Sigh,
you do realise it was Tom's Hardware that proved the 480 TDP and power distribution was incorrect by using similar methodology and context, and followed up by PCPer?
Funny that AMD corrected their card based upon Tom's Hardware and PCPer, but hey what do we all know, and Metro gave results good enough for the point back then and is hard on power draw for both manufacturers, hence why it is used.
Yeah time constraints and they spent a solid 2 days just mapping the envelope of 2 GPUs, however again you ignored they double checked by running the game and analysing default/minimum/max clocks to see how it correlated to their earlier model.
Anyway enough, but had to respond as again you skewed the context and my knowledge, well done.

 

Interestingly Metro even though is an old game, when it comes to power consumption its a demanding game so what we are looking at is one of the more power intensive games that give out those kinda of figures.

 

While I don't think anyone needs to point out that you pulled a very dull and banal straw man argument, and with an irritating mocking tone to boot, but the hypocrisy of then accusing others posts of being full of bile was really the cherry on the cake.

Good God man, this is the kind of thing I expect on reddit...

Obviously I cannot speak for Razor or CSI but I am quite certain nothing they said suggested power and frequency scale linearly.

The graph posted above shows fairly linear behavior in the clock range tested.

I'm going to take care to be clear because I don't want you pulling another strawman, pretty soon this thread will be a barn.

1. Indisputable fact; the entirety of a GTX1060 consume 62 watts at 1500mhz, at 2000mhz it's ~120w.

2. Indisputable fact: the Tesla P40 is a a full GP104 config with slower gddr5 clocked at ~1ghz boost. The power draw claimed by nvidia is 50-75w, of the whole board.

So what exactly is your problem?

The 1060 has a 50% reduction in power from a 25% reduction in clock from 2000-1500mhz.

Seriously ? Come on.

 

what happened to the post before you posted that? The one I quoted you on? That just isn't important anymore for this conversation?

I don't get you tots, you start something, but don't like it when people show you contrary to what you stated

I showed you the gp107 did get down there. Yet you keep on going with your imaginary thinking......

BTW I did no such things to you other then stating you should look up some of the things you are posting about, hinting to you are not correct. Maybe next time I will call you out as you are? Would that make you feel better? At least that way when you accuse me of personally attacking you it will actually be real?

 

I'm laughing as I am reading all this flaming. You people take yourselves way too seriously. It's just a message board. Nobody will remember what you wrote 48 hours from now.

 

Now we have come down to Pokemon memes lol! At least its entertaining

BTW the TDP figure for the laptop GPU from the way AMD has been marketing their TDP's this gen in their manuals, is for the GPU only, So yeah the question arises how much is it for the memory too.