NVIDIA GeForce RTX 5000 Series “Blackwell” TDPs Leaked, Entire Lineup Unified with 12+4 Pin Power Connector

[Zach_01]

90s are the new Titans

 

[freeagent]

"look at my massive e-balls" card in my eyes, which is exactly what SLi/CF were all about.

I didnt look at it like that, but I was an earlyish adopter. As soon as they let you mix vendors I was all up in it with my 6600GT lol.

SLi was awesome. What wrecked it was the industry making try hards feel like they were the hardcore gamers that they thought they were once real money started to be tossed around.

Once overclocking started to get blended with gaming all bets were off on all sides lol. Not only that, but hardware was accelerating in leaps and bounds.

It the old world of HP, Dell, and other prebuilds, overclockers were the underbelly of the internet, now it is glamorous

 

[londiste]

Ultra high end stuff has kind of been there all the time but the range of everything and especially the roof on what is possible/sold has basically just flown off.

 

[ShrimpBrime]

Will the cards still come with pcie 8 pin to 12 pin adapters? Know my 4070 Super came with it, or I wouldn't be using the card....

 

[ShiBDiB]

I didnt look at it like that, but I was an earlyish adopter. As soon as they let you mix vendors I was all up in it with my 6600GT lol.

SLi was awesome. What wrecked it was the industry making try hards feel like they were the hardcore gamers that they thought they were once real money started to be tossed around.

Once overclocking started to get blended with gaming all bets were off on all sides lol. Not only that, but hardware was accelerating in leaps and bounds.

It the old world of HP, Dell, and other prebuilds, overclockers were the underbelly of the internet, now it is glamorous

I think it was more hardware lagged behind software, now it's reversed or more equal. SLi/CF just became unnecessary to enjoy games at good frames.

 

[JustBenching]

Doesn't matter. What matters is whether that singular gaming system with a top end GPU in it, uses 300-350W (2016), or over 600W (2024). And that's not touching the OC button, mind you, that's running both systems 'conservatively'. In the end, physics hasn't changed, you're still getting an ever increasing energy bill and heat production. Efficient or not.

Again, it's called power limit. Solves the problem. Would you be complaining about the out of the box brightness of your TV being too high and using a lot of electricity? Probably not, you would just turn it down a notch right?

 

[Vya Domus]

The 5090 for example, will have double the shader count of the 5080 and a 200 W higher TDP - you could fit an entire product stack in between.

Nvidia knows the product at the top of the stack is the only one that will sell for sure at this point in time, which is why they are widening the gap between it and everything else.

The disadvantage of holding the majority of the market share with a non subscription based product is that it's becoming increasingly difficult to compel people to upgrade, especially when that upgrade proves insignificant so this is a way to ensure that at least some products will keep selling.

 

[Shou Miko]

Of course undervolting doesn't lower power draw, that's because you are just moving the frequency curve to the left, meaning your card is trying to hit higher clockspeeds. So by UV, you are just gaining performance with the same power. If you want to lower power, power limit it.

So you are saying that Bryan from TYC and MSI AB is lying when it shows about 130W for this Zotac RTX 4070?

I do that manually in MSI AB, I managed to achieve the out of the box clocks (2730MHz) at 985mv, cut power from 205-210W to 170-180W. You need a volt/freq curve editor for that (Ctrl+F).

I been there done that still no power cut down so, I just left it and bumped up the mem clock but I might see what I can push out of the gpu clock but I doubt that 180MHz will do much for it because it's running 2550MHz in the OC mode from the factory.

 

[JustBenching]

So you are saying that Bryan from TYC and MSI AB is lying when it shows about 130W for this Zotac RTX 4070?

No, I'm saying you don't understand what you are watching. In the TYC video he didn't just undervolt the card, he also locked the clockspeeds. That's why power draw dropped. If you don't lock clocks or power, undervolting will not drop your power draw cause your card will simply try to boost higher.

Undervolting simply means that your card will require less voltage for any given frequency. Let's say your card runs 1volt @ 2.5ghz by default. If you undervolt it it will run 2.5ghz @ 0.9 volts, but your card will still boost to 1v only this time at 2.7 ghz. Since it still hits 1v your power draw will remain the same before and after the undervolt.

 

[Shou Miko]

No, I'm saying you don't understand what you are watching. In the TYC video he didn't just undervolt the card, he also locked the clockspeeds. That's why power draw dropped. If you don't lock clocks or power, undervolting will not drop your power draw cause your card will simply try to boost higher.

Undervolting simply means that your card will require less voltage for any given frequency. Let's say your card runs 1volt @ 2.5ghz by default. If you undervolt it it will run 2.5ghz @ 0.9 volts, but your card will still boost to 1v only this time at 2.7 ghz. Since it still hits 1v your power draw will remain the same before and after the undervolt.

I know and I locked the clicked in what he showed, no stability issues but TDP is still showing 195watts even in GPU-Z so I guess my card just don't listen to MSI AB.

 

[Timbaloo]

I wonder why people expect miracles in terms of performance and efficiency with every new GPU generation. This ain't the 2000s anymore where each gen came along a significantly better node. The architectures are refined enough since quite a couple of gens so big generational leaps in terms of architectural improvements are not to be expected either. The GPU business has reached the CPU business in terms of performance and efficiency gains per generation, where this apparently is an acceptable state since many many years. Deal with it.

Merry Christmas.

 

[JustBenching]

I know and I locked the clicked in what he showed, no stability issues but TDP is still showing 195watts even in GPU-Z so I guess my card just don't listen to MSI AB.

What do you mean TDP? TDP from GPU Z will always be 195w, that's just what it reads from the bios. Have you checked your power draw?

 

[agent_x007]

Of course undervolting doesn't lower power draw, that's because you are just moving the frequency curve to the left, meaning your card is trying to hit higher clockspeeds. So by UV, you are just gaining performance with the same power. If you want to lower power, power limit it.

In my book, that's just using UV wrong...
Stock + maxed power slider :

Undervolt/Underclock + maxed power slider :

Unless you can invalid Power(W) = Voltage(V) x Current(A) equation, decrease in voltage WILL decrease power.
HOWEVER, in modern GPUs (especially GeForce 30-series and newer), due to dynamic boosting and different number of transistor that switch (due to different loads), actual true maximum power usage is always hidden by "limit by power" cap.

Second point (about different loads having different voltage to show difference), is best shown by comparing previous game results to Furmark2 :
Stock + Max power slider :

Undervolt/underclock + maxed power slider :

^Regardless of voltage use in this case, power limit will be reached (apparently "breaking" equation I mentioned earlier).
Same settings used for both game and Furmark, latter doesn't care about voltage, and former scales power nicely with voltage.

In short : The bigger the die, the more you have to decrease the voltage to even see a dent in actual power consumption show in GPU-z under (in GAMES).

 

[Zach_01]

Some just UV for some performance gains when power and clocks are not constrained.
It’s not wrong usage of UV, just different goal.

Whoever wants power down reduces power limit. Doing UV on top of that is a different story. Making it more efficient with minimum performance losses.

It’s like doing negative v/f curve on Ryzens. If you don’t limit power the CPU will reach higher clocks within the same initial power limit.

At the end… goals are different for everyone

 

[agent_x007]

Some just UV for some performance gains when power and clocks are not constrained.
It’s not wrong usage of UV, just different goal.

Fine, I get that. But his statement still casually breaks laws of physics though...

 

[Zach_01]

What happened with the efficiency where Maxwell and Pascal shined? 500W TDP for the flagship, really?

500W is absurd yes, but in all fairness the 4090 for example is much more efficient than the pascal 1080.

GTX1080: 180W (100% perf)
RTX4090: 450W (423% perf)

x2.5 power increase
x4.23 performance increase

Power to perf increase ratio:1.69 if we compare them linearly through out the power band which won’t do justice on the 4090.

Heck even the “bad” 7900XTX has much better efficiency from 4090 in comparison to 1080.

GTX1080: 180W (100% perf)
7900XTX: 355W (345% perf)

x1.97 power increase
x3.45 performance increase

Power to perf increase ratio:1.75

Want a more similar comparison?

GTX1080: 180W (100% perf)
RTX4060Ti: 160W (165% perf)

x0.89 power increase
x1.65 performance increase

Power to perf increase ratio: 1.85

So, to conclude…
Today GPUs up to current series have better (70~90%) efficiency from GTX1080.

Total power alone means nothing if you leave performance out of the equation.
And I didn’t even mention other capabilities that old GPUs don’t have. Others did mentioned them though.

Yes efficiency does not improve as it used to like 15-20y ago.
But this is normal as the current chip manufacturing tech is reaching slowly to its limits while performance demands keep growing faster.

 

[agent_x007]

So, to conclude…
Today GPUs up to current series have better (70~90%) efficiency from GTX1080.

Total power alone means nothing if you leave performance out of the equation.
And I didn’t even mention other capabilities that old GPUs don’t have. Others did mentioned them though.

Yes efficiency does not improve as it used to like 15-20y ago.
But this is normal as the current chip manufacturing tech is reaching slowly to its limits while performance demands keep growing faster.

For context (xx4-class full die comparison) :
[GF114] GTX 560 Ti (2011) : 170W TDP
[GP104] GTX 1080 (2016) : 180W TDP
[AD104] GTX 4070 Ti (January 2023) : 285W TDP

TPU perf. number :
[GF114] GTX 560 Ti (2011) : 100%
[GP104] GTX 1080 FE (2016) : 436% (vs. GTX 560 Ti)
[AD104] GTX 4070 Ti (January 2023) : 1154% (vs. GTX 560 Ti)

Efficiency (relative perf. vs. TDP) :
[GF114] GTX 560 Ti : 0.588 (Perf/W)
[GP104] GTX 1080 FE : 4.113 (Perf/W)
[AD104] RTX 4070 Ti : 4.049 (Perf/W)

EDIT (no GF114, GP104 = 100%) :

Performance :
[GP104] GTX 1080 FE (2016) : 100%
[AD104] GTX 4070 Ti (January 2023) : 265% (vs. GTX 1080)

Efficiency (relative perf. vs. TDP) :
[GP104] GTX 1080 FE : 0.5556 (Perf/W)
[AD104] RTX 4070 Ti : 0.93 (Perf/W)

Funny how math works

 

[AusWolf]

Funny how math works

That's the verdict.

Reviews like to math out everything, but life isn't made of numbers, just a bunch of experiences. Buy the GPU that fits your needs, and don't give a damn about the math. At least that's what I do.

 

[Zach_01]

That's the verdict.

Reviews like to math out everything, but life isn't made of numbers, just a bunch of experiences. Buy the GPU that fits your needs, and don't give a damn about the math. At least that's what I do.

I dont care either, but some times you need them to show that thinking intuitively is not always going to bring you to the right conclusion.
Seeing a number alone...

Like I said before GPUs today draw x5-10 the power from 25y ago but perform x300-400+.
A fact is a fact.

 

[Sir Beregond]

Speaking of the 16-pin, I don't understand why they didn't retain how they did the 12-pin on the 30-series FE cards. No hard bend in this cable (a 12VHPWR) which is plugged into my 3080 Ti FE. Honestly seemed like a good idea how they stood the connector up and angled it. Never had any problems.

 

[tpa-pr]

Speaking of the 16-pin, I don't understand why they didn't retain how they did the 12-pin on the 30-series FE cards. No hard bend in this cable (a 12VHPWR) which is plugged into my 3080 Ti FE. Honestly seemed like a good idea how they stood the connector up and angled it. Never had any problems.

Oh that's a brilliant idea. Was that an FE exclusive or did other AIBs do it too? That seems like it'd solve a lot of problems with bending/plug pressure.

 

[x4it3n]

I dont care either, but some times you need them to show that thinking intuitively is not always going to bring you to the right conclusion.
Seeing a number alone...

Like I said before GPUs today draw x5-10 the power from 25y ago but perform x300-400+.
A fact is a fact.

Technology has come a long way in 20+ years for sure. But now it's all about A.I. so metrics are all screwed

 

[AvrageGamr]

I know and I locked the clicked in what he showed, no stability issues but TDP is still showing 195watts even in GPU-Z so I guess my card just don't listen to MSI AB.

Does your 4070 support undervolting? Not all models do. Mine doesn't. Even if I unlock the voltage control in Afterburner, it does nothing.

 

[freeagent]

AB does nothing with my card. Maybe the fans.. cant remember. It isn't the first time AB failed me. I have an Asus card that it is useless with as well.

Kind of a bummer..

 

[Sir Beregond]

Oh that's a brilliant idea. Was that an FE exclusive or did other AIBs do it too? That seems like it'd solve a lot of problems with bending/plug pressure.

This was the 30-series Nvidia 12-pin which is basically the 16-pin 12VHPWR minus the sense pins. As far as I am aware, for 30-series, only the FE cards had them. The 3090 Ti I believe was the first card with the 12VHPWR 16-pin connectors.