die size

[GK104]

hey folks, Is that right that a larger gpu die size causes a higher temperatures and higher power consumption?

the GTX TITAN is manufactured at 28nm proccess and NVIDIA put 7100M transistors on 550mm^2 die size, so why the R9 290X dues to 95C at full load if AMD put 6200M transistors on 438mm^2 at proccess of 28nm?

R9 290X, smaller die size and less transistors than TITAN, both proccessed at 28nm, but the 290X is hotter, why is that?

 

[n0x1ous]

hey folks, Is that right that a larger gpu die size causes a higher temperatures and higher power consumption?

the GTX TITAN is manufactured at 28nm proccess and NVIDIA put 7100M transistors on 550mm^2 die size, so why the R9 290X dues to 95C at full load if AMD put 6200M transistors on 438mm^2 at proccess of 28nm?

R9 290X, smaller die size and less transistors than TITAN, both proccessed at 28nm, but the 290X is hotter, why is that?

The bigger the die the more space there is for the heat to be spread out. the Transistors on Hawaii are packed closer together than GK110. the 95C is simply the function of fan speed. If AMD had made a better cooler or you turn the fan up then it will be less than 95C

 

[PingviN]

hey folks, Is that right that a larger gpu die size causes a higher temperatures and higher power consumption?

the GTX TITAN is manufactured at 28nm proccess and NVIDIA put 7100M transistors on 550mm^2 die size, so why the R9 290X dues to 95C at full load if AMD put 6200M transistors on 438mm^2 at proccess of 28nm?

R9 290X, smaller die size and less transistors than TITAN, both proccessed at 28nm, but the 290X is hotter, why is that?

You're forgetting about cooler efficiency. Titan and R9 290X have different coolers, with the 290s being worse.

More transistors mean higher power consumption. Packing transistors closely (like on the R9 290) means there is less space to allow for heat dissipation. Shrinking the process also has effect on power consumption as it means you can lower the voltage between the transistors (less power, less heat).

In short:

Cooler matters
Die size matters
Transistors overall matters
Space between transistors matters

 

[GK104]

The bigger the die the more space there is for the heat to be spread out. the Transistors on Hawaii are packed closer together than GK110. the 95C is simply the function of fan speed. If AMD had made a better cooler or you turn the fan up then it will be less than 95C

so basically, bigger die size = better?

 

[DiogoDX]

so basically, bigger die size = better?

I tink is less transistor density = better.

GK110 = 7.1B/550mm2 = 12.9M/mm2
Hawai = 6.2B/438mm2 = 14.1M/mm2

 

[SiliconWars]

so basically, bigger die size = better?

The bigger the die size, the easier it is to get higher performance due to larger dies normally having more transistors and more space. The flipside of this is that you get less dies per wafer and are more open to other yield problems like defects, so the cost goes up.

 

[RaulF]

so basically, bigger die size = better?

Maybe in the sense that larger surface area can dissipate more heat, or also can pack more transistors depending on tech being used.

But no, not necessarily better.

 

[el etro]

so basically, bigger die size = better?

More Stream Processors/Cuda Cores = better(a bit inaccurate, but gives a good idea of what performance the GPU will have). Performance(and power consumption) don't increase linearly with each transistor added to GPU.

AMD and Nvidia spend different amounts of transistor for each stream processor added to GPU.

R9 290X, smaller die size and less transistors than TITAN, both proccessed at 28nm, but the 290X is hotter, why is that?

A inaccurate but explainable symptom of bad thermals of 290X is its poor cooling system.

 

[parvadomus]

The more GPU units a die has (shaders, rops, tmus, etc.) the more efficient it will be to reach certain performance level (assuming this units are balanced, not something like Tahiti).
GK110 may be a little more efficient than Hawaii thanks to having more performance per mhz (Im not 100% sure about this, as no review site has locked both GPUs at same frequency for testing).
Power consumption depends on these:
- Core voltage (a big die, requieres less mhz to reach certain speed, and then less voltage is needed)
- Temperatures (the higher the temp. a chip runs the higher the power consumption)
- Core speed (same as 1)

So a bigger die is always better.

 

[Stuka87]

The more GPU units a die has (shaders, rops, tmus, etc.) the more efficient it will be to reach certain performance level (assuming this units are balanced, not something like Tahiti).
GK110 may be a little more efficient than Hawaii thanks to having more performance per mhz (Im not 100% sure about this, as no review site has locked both GPUs at same frequency for testing).
Power consumption depends on these:
- Core voltage (a big die, requieres less mhz to reach certain speed, and then less voltage is needed)
- Temperatures (the higher the temp. a chip runs the higher the power consumption)
- Core speed (same as 1)

So a bigger die is always better.

More shaders/rops/etc does not make the chip more efficient. There is WAY more that goes into efficiency.

A larger die will allow more heat transfer as there is a larger heat sink area. However, larger dies cost more, and are more prone to manufacturing flaws because of their size. Which is why they cost much more to produce.

And your comment on a larger die not needing as high of a core voltage is not entirely true. Again, many factors go into this.

Ultimately there is no "better". Its all a balancing act. It depends on the ultimate goal that the GPU has to fulfill. In some cases a larger die may be better, in others it is worse. You saying they are always better is simply not true.

 

[Harvey]

The issue is not die size, per se. Hundreds to thousands chips are fabircated on large wafers which are then cut into individual dice. Therefore, a smaller die size means the manufacturer realizes more dice, and potentially, more profit per wafer.

Each step toward smaller feature size of various components on the die (transistor parts, resistors, traces, etc) means more components can be squeezed into a given area. This means these features are closer together which results in lowering the breakdown voltage of electrical insulating material between the various components and conductors. This is why newer IC's with smaller features must operate at lower voltages.

Power = Volts x Amps (P = EI). Therefore, the heat generated in an IC can be determined by multiplying the operating voltage across the IC times the current used by the circuit at that voltage. Fortunately, with each drop in feature size, the drop in operating voltage at a given current results in less power for a given number of components. The same is true as new component topologies are developed that require less operating current.

It's a matter of power dissipation over area. In other words, how much surface area is available to disspate the power used by the circuit. The answer is engineering newer, more efficient ways to remove heat from the die. In larger terms, it's similar to the improvements in the efficiency of CPU heatsinks as they moved from large chunks of aluminum with fins and fans to heat pipes.

Smaller feature size also means lower capacitance between features which allows faster circuit operation. Unfortunately, the smaller die size makes it more difficult to remove the heat that IS produced by operation of the circuit.

Another problem with smaller die is that it involves newer technologies that may or may not produce more working dice (yield) as a percentage of all the dice on a wafer. The yield at any given feature size tends to improve over time as manufacturers learn more about how to produce it.

The bottom line is that smaller die size is better... with several caveats. It's all money driven. How many working dice can be manufactured at what cost and with what reliability? IC manufacturers look at many sometimes conflicting cost-benefit curves to determine the optimum technical combination for a given new product line.

The bottom line is that smaller die size is better... with several caveats. It's all money driven. IC manufacturers look at many sometimes conflicting cost-benefit curves to determine the optimum technical combination for a given new product line.

And we haven't even touched the issue of the greater component density realized by going from two dimensional to three dimensional feature designs.