ada lovelace Alleged NVIDIA AD106 GeForce RTX 40 GPU (4060? 4060 Ti? 4070 Mobile?) Benchmarks Show Around RTX 3070 Ti Class Performance

[BiG StroOnZ]

Summary

Benchmarks and specifications of an alleged NVIDIA AD106 GPU have appeared on Chiphell forums, @harukaze5719 on Twitter posted all the details. The AD106 GPU in question could be the GeForce RTX 4070 Mobile, RTX 4060 Desktop, or RTX 4060 Ti Desktop. This AD106 GPU has been compared to the GeForce RTX 2080 Super and the RTX 3070 Ti or TU104 and GA104 silicon.

 

 

 

 

Quotes

Quote

Posting on the forum, user "panzerlied" dropped a chart with a comparison of what he claims is a "borrowed" graphics card with an AD106 GPU against cards with GA104 and TU104 processors.

 

The GPU was tested in AIDA64 and 3DMark.

 

The benchmarks include the Speedway, Port Royal, TimeSpy Extreme, TimeSpy, FireStrike Extreme, FireStrike Ultra, and Firestrike scores of the AD106 (RTX 4060 Ti), GA104 (RTX 3070 Ti), and the TU104 (RTX 2080 Super).

 

Assuming these results are legitimate, this gives us a decent picture of what kind of performance we can expect out of x60-class Ada Lovelace GeForce cards. As expected, the AD106 GPU runs an extremely high GPU boost clock, and this gives it great compute and raster performance despite being smaller than the GA104 by a fair bit. However, it is suffering under a rather extreme memory bandwidth deficit compared to the older cards.

 

It beats the RTX 2080 Super in all of the tests, while drawing some 55 W less power at the same time. In some of the benchmarks the wins are within the margin of testing error, for example when it comes to the memory performance in AIDA64. Yet, the AD106 GPU manages to beat the RTX 2080 Super in all of the memory benchmarks in AIDA64.

 

Demonstrated in the 3DMark results, the AD106 chip generally comes out ahead of both TU104 and GA104 in the standard tests at 1080p and 1440p, but in Fire Strike Ultra and Time Spy Extreme—which run in native 4K resolution—it falls behind the older GPUs. It also loses out to the GA104 GPU rather badly in the new Speed Way test that makes heavy use of asset streaming. That's likely because of its PCIe 4.0 x8 interface, which you can see demonstrated in the PCIe test at the bottom.

 

Compared to the RTX 3070 Ti (GA104), things aren't looking nearly as good, although the AD106 GPU still manages to win in a lot of the AIDA64 tests and even in a couple of the 3DMark benchmarks. The margins by which it won in 3DMark is within one to three percent.

 

There's also the consideration that the AD106 part comes out well ahead in the DXR ray-tracing feature test, though, and that's to say nothing of the new card's support for DLSS 3 frame generation.

 

My thoughts

This is an interesting leak/rumor, as the performance isn't as bad as some other rumors were originally suggesting. If a 4060 or 4060 Ti offer around 3070 Ti performance, that really isn't too shabby, all things considered. Even if this is a 4070 Laptop GPU, that's not that bad either, offering desktop 3070 Ti performance. It should be noted that the original poster on Chiphell forums has since removed all these details, so take with a grain of salt and remain skeptical. However, in his updated post he does mention that final performance of AD106 will be higher and this performance isn't representative of future AD106 cards. There is also mention in that thread of someone with a 4060 Ti claiming 3080 performance. So, whatever AD106 part this is, the performance is pretty good if these results are accurate (and if we expect some performance increases in the final silicon). This card appears to be geared towards 1080p or 1440p. Especially with only 8GB of VRAM. Kind of a step backward from the 3060 12GB. Although, the 3060 Ti did have 8GB of VRAM. One of the biggest concerns with this AD106 part is pricing. As if it's priced anything like the Ada cards that have been released thus far, it will probably cost at least $500. While that's a $100 discount from the 3070 Ti MSRP, this AD106 part will need to perform more like a 3080 to be a better value proposition (since the 3080 MSRP was $699). Despite all this, I think the midrange market is suffering and cards like this need to be released soon. I know 40 Series Laptop GPU embargo lifts on the 7th or 8th. Therefore, we might get a better understanding of performance with low to midrange Ada cards around then (since that includes the 4070, 4060, and 4050). 

 

Sources

https://www.hardwaretimes.com/nvidia-rtx-4060-ti-as-fast-as-the-rtx-3070-ti-in-leaked-3dmark-benchmark-with-a-tbp-of-180w-rumor/

https://www.techpowerup.com/304461/alleged-nvidia-ad106-gpu-tested-in-3dmark-and-aida64

https://www.eteknix.com/nvidia-4060-benchmark-leak-performance-par-3070-ti/

https://hothardware.com/news/alleged-ad106-gpu-benchmarks

[HenrySalayne]

Is it just me or is the performance spread with the 40 series much more pronounced than on previous generations.

4090 -> 4080 -> 4070 TI -> 4060 each being like 20 - 30 %?

[leadeater]

12 minutes ago, HenrySalayne said:

Is it just me or is the performance spread with the 40 series much more pronounced than on previous generations.

4090 -> 4080 -> 4070 TI -> 4060 each being like 20 - 30 %?

I think so, I'll wait for more models to come out though. Certainly is for the 4090 vs 4080.

 

I like that it is, I've always hated barely any difference between each model, large price increases and more models than necessary due to minimal performance differences. I won't be complaining much if this is the case.

[porina]

Interesting comparing vs 3070Ti. 4060 has half external memory bandwidth, but double internal copy? We know they increased cache but I wonder what other internal changes there were. Core count down more than made up from running those still there quite a bit faster.

 

I think it'll serve as an entry level 4k GPU too, a spot held by 3070 and 2080Ti. 8GB may be getting a bit tight so some tradeoffs may have to be made in settings to ensure a smooth performance, most likely not having highest resolution textures.

[starsmine]

1 hour ago, HenrySalayne said:

Is it just me or is the performance spread with the 40 series much more pronounced than on previous generations.

4090 -> 4080 -> 4070 TI -> 4060 each being like 20 - 30 %?

With the exception of  4090-> 4080, its just you. 

Not going to compile data from all recent gens.

3090ti- 233  - Δ 9% faster
3090 - 216 - Δ 1% faster
3080ti -214- Δ 12% faster
3080 - 191 - Δ 21% faster
3070ti -157- Δ 6.8% faster
3070 - 147 - Δ 17.6% faster
3060ti - 125 - Δ 25% faster
3060 - 100 - Δ 37% faster
3050- 73 -

These numbers are compressed because full line up. The launch window line up looks more like this

3090 - 216 - Δ 13% faster
3080 - 191 - Δ 30% faster
3070 - 147 - Δ 17.6 faster
3060ti - 125 - Δ 25% faster
3060 - 100 

Take out 3090 doing the NORMAL flagship thing of diminishing returns and you have performance spread of 17.6% to 30%

Ill do Turing sans super as well. (super refresh is more like a half gen, like 500/700 were)

 

TITAN - 198 - Δ 21%

2080 ti - 163 - Δ 17%

2080 - 139- Δ 20% 
2070 - 116- Δ 16%
2060 - 100 -Δ 17.6%
1660ti - 85 -Δ 15%
1660 - 74 - Δ  51%
1650 - 49 - Δ
 

The launch for this line up was unusual and very rapid, the entire line up coming out in two quarters. but each delta being 16-20% (dat 1650 to 1660 50% delta tho)

[HenrySalayne]

27 minutes ago, starsmine said:

With the exception of  4090-> 4080, its just you. 

Not going to compile data from all recent gens.

Guess it's not as much as it feels like.

With the 4k average from TechSpot's 4070 TI review:

4060 (3070 TI) -> +50% -> 4070 TI -> +25% -> 4080 -> +30% -> 4090 

Overall a factor of 2.44x from 4060 to 4090. .

 

With the 4K average from TechSpot's 3070 TI review:

3060 -> +62% -> 3070 TI-> +23% -> 3080 -> +13% -> 3090

Overall a factor of 2.26x.

 

 

[ouroesa]

Based on the -50% memory read/write, I'd assume this is a mobile part, maybe 4070 Mobile?
Otherwise they will have like a 64bit memory bus