From ATI to AMD back to ATI? A Journey in Futility @ [H]

Nah, i mean the frequency/voltage dependency becoming borderline-linear after 910Mhz.
Voltage requirements change as frequencies rise and fall.
Saying voltage requirements become linear after 910MHz doesnt mean much and could be incorrect around that frequency unless you have evidence.

910MHz is a low speed that is easily attainable, voltage will likely not need a raise around then because other factors determine the minimum voltage. Without a needed raise, it would not be linear.
At some point it will need to start raising and beyond that the rise will become exponential (non linear) and will be not possible without causing damage.

To say that after 910MHz it becomes linear matters little at that frequency and soon becomes incorrect, if it is correct in the first place.
Way before 1500MHz, voltage will not rise linearly with frequency.
Its a pointless statement.
 
Last edited:
Well what he is saying is, I think at least, there is a leveling off of voltage to frequency, to wattage, which yeah it does happen when you get into the ideal range of an architecture *this is why its possible in some instances to get more frequency with a less than linear increase in wattage*, but of course once you hit the upper limit of that range, voltage starts going up again and then you of course have wattage increases which no longer are linear but exponential or more than exponential
 
Voltage requirements change as frequencies rise and fall.
Saying voltage requirements become linear after 910MHz doesnt mean much.

910MHz is a low speed that is easily attainable, voltage will likely not need a raise around then because other factors determine the minimum voltage.
At some point it will need to start raising and beyond that the rise will become exponential and will be not possible without causing damage.

To say that after 910MHz it becomes linear matters little at that frequency and soon becomes incorrect.
Way before 1500MHz, voltage will not rise linearly with frequency.
Its a pointless statement.
it goes from 843 mV to 1150 mV while frequency goes from ~910 to ~1260. Fairly linear, if you ask me.

Also f(V) is not exponential AFAIK, more like a power law,

Well what he is saying is, I think at least, there is a leveling off of voltage to frequency, to wattage, which yeah it does happen when you get into the ideal range of an architecture, but of course once you hit the upper limit of that range, voltage starts going up again and then you of course have wattage increases which no longer are linear but exponential or more than exponential

I think ya'll sometimes forget what exponential law really is. But whatever. What i mean, is that P10 is probably working above it's sweet range, so voltage increase is already required for frequency increases, and said increase is linear with frequency. So, cubic increase in power consumption, and that gets ugly real quick on small die heat-wise.

P. S. There were rumors of RX480 actually having Maxwell-like wacky boost clock, any bets certain folks that called Maxwell's boost nothing but a benchmark inflation measure, will keep silent on that in this card?
 
it goes from 843 mV to 1150 mV while frequency goes from ~910 to ~1260. Fairly linear, if you ask me.
You didnt state a limit, you said above 910MHz.
1260MHz isnt anything to brag about.
Also f(V) is not exponential AFAIK, more like a power law
Power is proportional to voltage squared.
Voltage has an exponent of 2.
I think ya'll sometimes forget what exponential law really is. But whatever.
Interesting...
 
Last edited:
it goes from 843 mV to 1150 mV while frequency goes from ~910 to ~1260. Fairly linear, if you ask me.

Also f(V) is not exponential AFAIK, more like a power law,



I think ya'll sometimes forget what exponential law really is. But whatever. What i mean, is that P10 is probably working above it's sweet range, so voltage increase is already required for frequency increases, and said increase is linear with frequency. So, cubic increase in power consumption, and that gets ugly real quick on small die heat-wise.

P. S. There were rumors of RX480 actually having Maxwell-like wacky boost clock, any bets certain folks that called Maxwell's boost nothing but a benchmark inflation measure, will keep silent on that in this card?


I see what you are saying that makes sense good post!
 
You didnt state a limit, you said above 910MHz.
1260MHz isnt anything to brag about.

Power is proportional to voltage squared.
Voltage has an exponent of 2.

Interesting...
My post implied one have seen that leaked screen of said OC utility with voltage/freq table, but w.e.

Also, fixed exponent is power law per definition.
 
The point is to express my concern with overclocking of Rx480. My opinion is that it will be severely heat limited on any air cooler.
Why the big concern over air cooler? Wouldn't a high end air cooler like the tripple fan cooler on the fury pro's work fairly well? Of course i have come to realize the blower coolers on the current 480's are no where good enough for real cooling. I'll take a 480 with one of these coolers lol
14633528793SEuViJS3c_1_12.jpg
 
up to a certain point but as voltage goes up, temps go up very fast, so even a better air cooler might not do the trick.
 
Why the big concern over air cooler? Wouldn't a high end air cooler like the tripple fan cooler on the fury pro's work fairly well? Of course i have come to realize the blower coolers on the current 480's are no where good enough for real cooling.
It's matter of heat transfer. Air coolers may simply fail to cope with something like 200W+ worth of heat coming from the card. You can see the effect i talk about with overclocked 6950X.
 
Anyways is it the small size of the die that makes it so hard for the coolers to deal with the heat properly?
 
Anyways is it the small size of the die that makes it so hard for the coolers to deal with the heat properly?

the physical limitations of the GPU and its architecture doesn't change too much, just look at Fiji, you can slap an AIO on it but it didn't help overclocking.

Partial but not the only thing. Transistor density, can affect hot spots, but there are many other factors in the design. As ebduncan stated before, Polaris at stock already uses more voltage than a bigger and higher clocked Pascal......
 
Last edited:
On second thought i overestimated the heat problem, the bigger problem is simple voltage scaling making any serious overclock require unsafe voltage.
 
It's matter of heat transfer. Air coolers may simply fail to cope with something like 200W+ worth of heat coming from the card. You can see the effect i talk about with overclocked 6950X.

A card with a max TDP of 150W will not be dissipating more than 200W of heat.

And those coolers were very effective with some of the higher wattage cards, like the 380x/390x.
 
with bigger chips they are more efficient, but the problem with a smaller die size, you have a physical limitation of how much heat can be removed because of the smaller surface area so it really doesn't matter how much fans or even AIO you put on it at some point, that physical limitation can't be superseded. This is all in theory of course till we actually see the card in reviews.
 
A card with a max TDP of 150W will not be dissipating more than 200W of heat.

And those coolers were very effective with some of the higher wattage cards, like the 380x/390x.
I talked overclocking heat/power consumption. Also, my argument is that 380x/390x's heat is relatively easier to dissipate.

I mean, it won't pop until the voltage limit is removed, and i overestimated the power law a little, but it's there.
 
There are means to deal to that small die but not cost effective, plus of limited benefit. Other issue is thermal conductivity limits of the silicon and electrical limits of underlying transistors and other components.
 
the physical limitations of the GPU and its architecture doesn't change too much, just look at Fiji, you can slap an AIO on it but it didn't help overclocking..

No, because they needed to clock higher too attack a 980ti instead of the projected 980. Do you really think they slapped on that cooler for fun? Because it looked cool? Because they felt the added expense would be neat-oh? The cooler was enabling its OC. It just became stock because that was how far they could push. Built in AIO = close to max clock. Get it?

Do you think an 480 needs an AIO so it can destroy mid-range? I think the AIB partners will be able to do quite well on air with custom 480's me thinks!
 
with bigger chips they are more efficient, but the problem with a smaller die size, you have a physical limitation of how much heat can be removed because of the smaller surface area so it really doesn't matter how much fans or even AIO you put on it at some point, that physical limitation can't be superseded. This is all in theory of course till we actually see the card in reviews.

This is only 50% true.

The smaller dies mean there is less surface contact between the die and the sink, but this does not impose any hard 'limit' on how effective it can be cooled down. After All, that is the job of a heatsink, to give a small source of heat more surface area to dissipate that heat. Keep the sink cool and the die will keep cool. If the die is getting too hot, then the sink will follow suit.
 
Hmmm that's interesting to note...
<sarcasm>Well, you know me, I just make a lot of shit up in the name of page hits and yellow journalism, and then tell you that it came from "sources," so I would be extremely careful believing anything I put on the interwebs.</sarcasm>
 
Another disappointing launch in a row. Running hot is now synonymous with AMD. Come on, get your act together. Now don't come back with the bullshit that oh, you can run two of these suckers for a par performance with 1080 GTX. 2 GPUs is never equal to a single GPU experience, NEVER!
 
Looks like I have done the right move. Holding off for AIB versions. A bit more clock room and better cooling.
 
I guess the running hot and not hitting the clocks they wanted rumours are true... or at least have some truth.
 
So what we have is its hot, slower than expected , wait a minute didn't someone write an article about that lol.
B-but it's the drivers! And he used HBAO+ so it's biased! And he broke the NDA the evil bastard!
Seriously read the comments, it's comedy gold.
 
B-but it's the drivers! And he used HBAO+ so it's biased! And he broke the NDA the evil bastard!
Seriously read the comments, it's comedy gold.


Dude I wish I could post there its waste of time though, its not going to increase the performance that great, I will wait for more reviews, but I can tell ya right now, don't expect miracles. How many times have we heard drivers will do the trick and months later if not years we still don't see it.

Great to have hope but we have seen this review, the video reviews which were taken down. The overclocking bit yeah drivers might help there, but I would like to see the power usage, The guy already stated temps of the GPU, I don't see it changing much there so I don't think you will see "greatly" lower power consumption because of drivers or more overclocking head room because of drivers. (not geared at you but at the comments of his review)
 
Now to see how overclocking fares on AIB i guess, but even if 1500mhz is achievable, that's also going to come with a price increase not to mention diminishing returns in terms of perf: power . How inefficient will it get ?!
 
Back
Top