Kopite7kimi has revealed the power limit of NVIDIA’s Ada Lovelace GPU on the next-generation graphics card “GeForce RTX 40”. The power limit includes both desktop and mobile parts.
Details of NVIDIA GeForce RTX 40 Desktop & Mobile GPU Power Limits. Flagship desktop “AD102” equipped with Ada Lovelace has a maximum of 800W, flagship mobile “AD103” has a maximum of 175W
The NVIDIA GeForce RTX 40 series lineup with Ada Lovelace graphics architecture is available for both desktop and mobile.
At the flagship, I heard various reports that it will be equipped with a TGP of up to 900 W, but as you know, the power limit is the maximum value that the GPU can reach, and it does not reflect the actual TGP / TBP value. do not have.
These limits are significantly overclocked and may only be reachable by a custom cooled version.
A truth. The power limits:
AD103 (DT), 450W, AD103 (Mobile), 175W;
AD104 (DT), 400W, AD104 (Mobile), 175W;
AD106 (DT), 260W, AD106 (Mobile), 140W.
But I don’t think we need to use the full power cap.
— kopite7kimi (@kopite7kimi) June 18, 2022
So, when it comes to power limits, there are four types of NVIDIA Ada Lovelace GPU SKUs: AD102, AD103, AD104, and AD106.
All GPUs except the AD102 are available in mobile variations because the flagship chip has high power requirements that laptop platforms can’t support.
So did the GPU on the GA102, making the GA103 the fastest mobile product.
From the desktop lineup, the NVIDIA AD102 GPU features an 800W power limit, the AD103 GPU features a 450W power limit, the AD104 GPU features a 400W power limit, and the AD106 features a 260W power limit.
For mobile, the AD103 mobile GPU has a power limit of 175W, the AD104 mobile GPU has a power limit of 175W, and the AD106 mobile has a power limit of 140W.
Finally, there are numbers like this.
NVIDIA Ada Lovelace Desktop GPU：
- AD102-Maximum power limit 800W
- AD103-Maximum power limit 450W
- AD104-Maximum power limit 400W
- AD106-Maximum power limit 260W
NVIDIA Ada Lovelace Mobile GPUs：
- AD103-Maximum power limit 175W
- AD104-Maximum power limit 175W
- AD106-Maximum power limit 140W
This significantly raises the power limits for desktop and mobile products.
The power limit of 800W for AD102 is already dead, but the actual graphics card that uses it ends up with TGP600W.
The GPUs on the AD103 and AD104 are so similar that the TGP numbers on mobile are about the same, with a difference of 50W on the desktop segment.
The AD106 is a mainstream 260W segment, but the graphics card actually used ends up with a TGP close to 200W.
Mobile products are maximized with a 175W power limit because the platform is the best it can do in a given area.
Beyond that, existing laptops and coolers for them need to be significantly redesigned, and most manufacturers don’t seem to want to make such an additional investment right now.
Due to the difference in power limits between desktop and mobile products, there will again be a big performance difference between the two NVIDIA GeForce RTX GPUs.
However, that is solely due to mobile platform limitations, and while desktops are still taking the next level, mobile performance is expected to improve significantly.
NVIDIA Ada Lovelace “GeForce RTX 4000” GPU block diagram:
ソース：wccftech – NVIDIA Ada Lovelace ‘GeForce RTX 40’ GPUs Power Limits Detailed: AD102 at 800W, AD103 at 450W, AD104 at 400W, AD106 at 260W
The RTX4000 series has different power limits for desktop and mobile.
AD103 / 104/106 has mobile and desktop.
Of course, it’s still in the rumor stage, but I think it’s okay to understand that it’s almost confirmed.
This time AD103 seems to come out on both mobile and desktop.
The AD102 seems to have much more isolated performance (and power consumption) than the AD103 and below.
Of course, it is still in the rumor stage, but if AD102 and AD103 are separated, there will be a performance difference of about 10: 6 between RTX4090 and RTX4080.
The desktop version of AD102 seems to have a power limit of 800W, but since this is a single chip, it is true that the entire board is 900W.
It feels a little scary.
The Titan A / RTX4090 / Ti, which is likely to use the AD102, is likely to be a product that is difficult for the general public to reach in terms of price and power consumption.
Similarly, AD103 is 450W, AD104 is 400W, and AD106 is 260W, so it seems that only AD107 (if any) can be cooled by a single fan.
Also, the AD106 is the only model that can supply power with 8Px2, which is not the PCIe 5.0 standard.
It may be possible to use some AD104, but my view is that it may be difficult because the chip itself is 400W.
To be honest, I thought, “Well, who will be happy?”
Isn’t it a power supply specification that even abandoned gamers hesitate?
I don’t know what you think that RTX is only 35% widespread, even though nVidia and AIB, which went crazy with miners due to mining special demand, shipped their products like crazy. If this crazy spec isn’t accepted by the market, I’m wondering if there’s even more chance of straying.
At least I’m pretty much against the Titan A / RTX4090 / Ti where the AD102 is used.
The RTX4080 also has an upper limit of 450W, so it is a specification that can exceed 450W with the board alone.
Even RTX4080, the custom OC model may become a basic 4-slot cooler.
It feels like gamers around the world are testing how far they can keep up with this spec and price.
Most gamers have dropped out early due to soaring special mining demand, so I wonder if most of them will not follow. (In the sense that my heart is broken)
I think that the high-end model of RTX4000 may be killed by explosion.
nVidia RTX 3000 series GPU
¥249,800 (2022/06/18 18:53:53 As of Amazon research-detailed)
¥209,800 (2022/06/19 03:49:54 As of Amazon research-detailed)
RTX3080 10GB LHR
¥125,000 (2022/06/19 03:56:34 As of Amazon research-detailed)
¥97,800 (2022/06/19 02:20:19 As of Amazon research-detailed)
Geforce RTX3070 8GB GDDR6 LHR
¥125,244 (2022/06/19 03:49:55 As of Amazon research-detailed)
¥74,422 (2022/06/19 03:52:56 As of Amazon research-detailed)
RTX3060Ti 8GB GDDR6
¥99,860 (2022/06/19 04:14:48 As of Amazon research-detailed)
RTX3060 12GB GDDR6
RTX2060 is reprinted with 12GB memory
¥53,783 (2022/06/19 04:11:54 As of Amazon research-detailed)
nVidia Geforce RTX 1600 series
¥37,000 (2022/06/19 04:11:55 Amazon research-detailed)
¥28,400 (2022/06/19 03:52:56 As of Amazon research-detailed)