When I first heard the rumor that GT300 was going to be a 512 core, 512bit interfaced 2 billion plus transistor monster my first thought was "best of luck achieving a reasonable TDP". Equalling the relatively monstrous 236w TDP of the gTX 280 would require each core to consume less than half the power of their 65nm brethren which seems a little optimistic especially given the rumored leakage problems with TSMC's 40nm process.
Yesterday the following 40nm GT200m DX10.1 products were announced:
<<<<<<<396 gigaflops; 96 processor cores; 550MHz graphics clock; 1375MHz processor clock; 1800MHz; 1GB GDDR5 RAM; 128-bit memory width; 38-watt TDP
360 gigaflops; 96 processor cores; 500MHz graphics clock; 1250MHz processor clock; 1600MHz; 1GB GDDR5 RAM; 128-bit memory width; 28-watt TDP
174 gigaflops; 48 processor cores; 550MHz graphics clock; 1210MHz processor clock; 800MHz; 1GB GDDR3 RAM; 128-bit memory width; 23-watt TDP
158 gigaflops; 48 processor cores; 500MHz graphics clock; 1100MHz processor clock; 800MHz; 1GB GDDR3 RAM; 128-bit memory width; 23-watt TDP
72 gigaflops; 16 processor cores; 625MHz graphics clock; 1500MHz processor clock; 800MHz; 512MB GDDR3 RAM; 64-bit memory width; 14-watt TDP
Power Extrapolations
512cores/96cores * 38w = 202w (From the top of the list)
this means that if GT300 debuts at 550Mhz (less than expected) we can expect a ~202w TDP
But notice something else, A difference in 50Mhz (10%) core clock speed between the top two SKUs resulted in a whopping 35.7% increase in TDP (28w to 38w). From this we can conclude that there must be a voltage bump involved but also this suggests that ramping clockspeed on the 40nm process currently results in a heavy power increase which doesn't bode well for high clocks/overclocking. The 48 core SKUs exhibit the opposite behavior with identical TDPs despite 50Mhz clockspeed disparities. Of course the voltage bump may be greater than required simply to keep the OCers happy on the highest end SKU or who knows what other possible reasons.
TDP Restricted Clockspeed Extrapolations
I chose 236w as the estimated max TDP. As there are not enough data points to construct an exponential graph of clockspeed vs TDP I will use a linear approximation which should be relatively accurate for close in extrapolations.
From the TDP numbers above, the linear relationship is ~3.5:1 power to clockspeed.
This means that 34w is the disparity between the theoretical GT300 at 550Mhz and the 236w limit.
34 watts is a ~17% TDP increase from 202w.
17/3.5 = ~4.9% expected increase in clockspeed for the 17% increase in TDP holding to the 3.5:1 ratio.
4.9% greater than 550Mhz is 576.9Mhz coincidentally almost identical to GTX 260's clockspeed.
At these core counts and clockspeeds the card would do almost exactly 2 teraflops.
So... our theoretical GT300 with a 236w TDP has a core clock of ~576Mhz these calculations exclude the power draw of ram and many other factors but these estimations should be within 10% of the truth assuming core (Oh OK "Stream Processor
") architecture is similar.
These numbers are really a worst-case scenario as TSMc's 40nm process will likely improve and Nvidias architecture will likely improve and I took the worst possible TDP/clockspeed relationship by not factoring in the unchanged TDPs of the 48 core SKUs so the actual GT300 product will likely be better.
Obviously this is all speculation and the GT300 cores may be so architecturally different that my numbers are meaningless but nonetheless I thought the power numbers on the new SKUs were interesting.
_Nate
Edit: I now hear nvidia is targeting a 225w TDP for GT300
Yesterday the following 40nm GT200m DX10.1 products were announced:
<<<<<<<396 gigaflops; 96 processor cores; 550MHz graphics clock; 1375MHz processor clock; 1800MHz; 1GB GDDR5 RAM; 128-bit memory width; 38-watt TDP
360 gigaflops; 96 processor cores; 500MHz graphics clock; 1250MHz processor clock; 1600MHz; 1GB GDDR5 RAM; 128-bit memory width; 28-watt TDP
174 gigaflops; 48 processor cores; 550MHz graphics clock; 1210MHz processor clock; 800MHz; 1GB GDDR3 RAM; 128-bit memory width; 23-watt TDP
158 gigaflops; 48 processor cores; 500MHz graphics clock; 1100MHz processor clock; 800MHz; 1GB GDDR3 RAM; 128-bit memory width; 23-watt TDP
72 gigaflops; 16 processor cores; 625MHz graphics clock; 1500MHz processor clock; 800MHz; 512MB GDDR3 RAM; 64-bit memory width; 14-watt TDP
Power Extrapolations
512cores/96cores * 38w = 202w (From the top of the list)
this means that if GT300 debuts at 550Mhz (less than expected) we can expect a ~202w TDP
But notice something else, A difference in 50Mhz (10%) core clock speed between the top two SKUs resulted in a whopping 35.7% increase in TDP (28w to 38w). From this we can conclude that there must be a voltage bump involved but also this suggests that ramping clockspeed on the 40nm process currently results in a heavy power increase which doesn't bode well for high clocks/overclocking. The 48 core SKUs exhibit the opposite behavior with identical TDPs despite 50Mhz clockspeed disparities. Of course the voltage bump may be greater than required simply to keep the OCers happy on the highest end SKU or who knows what other possible reasons.
TDP Restricted Clockspeed Extrapolations
I chose 236w as the estimated max TDP. As there are not enough data points to construct an exponential graph of clockspeed vs TDP I will use a linear approximation which should be relatively accurate for close in extrapolations.
From the TDP numbers above, the linear relationship is ~3.5:1 power to clockspeed.
This means that 34w is the disparity between the theoretical GT300 at 550Mhz and the 236w limit.
34 watts is a ~17% TDP increase from 202w.
17/3.5 = ~4.9% expected increase in clockspeed for the 17% increase in TDP holding to the 3.5:1 ratio.
4.9% greater than 550Mhz is 576.9Mhz coincidentally almost identical to GTX 260's clockspeed.
At these core counts and clockspeeds the card would do almost exactly 2 teraflops.
So... our theoretical GT300 with a 236w TDP has a core clock of ~576Mhz these calculations exclude the power draw of ram and many other factors but these estimations should be within 10% of the truth assuming core (Oh OK "Stream Processor
") architecture is similar.These numbers are really a worst-case scenario as TSMc's 40nm process will likely improve and Nvidias architecture will likely improve and I took the worst possible TDP/clockspeed relationship by not factoring in the unchanged TDPs of the 48 core SKUs so the actual GT300 product will likely be better.
Obviously this is all speculation and the GT300 cores may be so architecturally different that my numbers are meaningless but nonetheless I thought the power numbers on the new SKUs were interesting.
_Nate
Edit: I now hear nvidia is targeting a 225w TDP for GT300
Facebook
Twitter