NVIDIA Reveals the RTX 4070 Ti & Ada Mobile GPUs
The wait has come to a close and CES 2023 is underway. NVIDIA just introduced the RTX 4070 Ti, while boasting some serious performance claims. Buckle up fellows, because this ride will get real fast, real quick. The RTX 4070 Ti is a direct threat to AMD’s upcoming Navi32-based GPUs. Despite being based on a Monolithic design, the RTX 4070 Ti has a decent MSRP, of course, when talking relatively.
Oh but that’s not all. NVIDIA went one step ahead and introduced the Ada Lovelace Mobile lineup. These GPUs are officially the fastest mobile GPUs on the planet. Performance does come at the cost of power, but these GPUs look extremely promising in terms of efficiency. Enough with the chatter, more with the news.
RTX 4070 Ti, Budget Gamers Rejoice
The RTX 4070 Ti is about to be the latest addition to many mid-range gamers’ arsenal. Now, readers should note that this GPU is a rebranded RTX 4080 12GB, which mind you, was canceled. In any case, the specifications were known many weeks before launch. In fact, based on the metrics NVIDIA provided for the RTX 4080 12GB, we knew the performance level of the RTX 4070 Ti beforehand.
The raw compute performance stands at 40 TFLOPs, similar to the RTX 3090 Ti. Furthermore, the RTX 4070 Ti offers 93 RT TFLOPs and a whopping 642 Tensor TFLOPs as per NVIDIA. For those wondering, the 3090 Ti boasts 78 RT TFLOPs and 320 Tensor TFLOPs of power.
Featuring NVIDIA’s AD104-400-A1 GPU, the RTX 4070 Ti packs 7680 Cuda Cores, 25% more than the 3070 Ti. That equates to around 60 SMs, which is less than half of what the RTX 4090 offers. This GPU has a base frequency of 2.31GHz which on boosting goes as high as 2.61GHz. Team green went with much, and I repeat, much more L2 Cache this time around with the 4070 Ti equipped with 48MB of fast L2 Cache.
The bus width stands at just 192 bits, which on using 21Gbps memory is around 504GB/s of effective memory bandwidth. The RTX 4070 Ti has 12GB of Micron’s GDDR6X memory. A respectable TGP of 285W will indeed make this GPU favorable for most gamers. Also, we should mention that this is technically the power limit for this GPU as NVIDIA in a very sneaky manner, changed the definition of TGP. In short, the GPU will almost always stay under the 285W threshold, which is quite efficient if you ask us.
Pricing
Now for the deciding factor, the MSRP, drumroll, please…….NVIDIA has priced the RTX 4070 Ti at just $799. This leaves a considerable difference between the RTX 4080 16GB and the RTX 4070 Ti. Should Jensen go all out and launch the RTX 40 SUPER series to fill in the price gap? That is yet to be seen, but then again, this is NVIDIA we’re talking about.
Availability
The RTX 4070 Ti will only arrive in custom AIB designs, so no FE variant from NVIDIA. The image marketed by team green is indeed misleading, but then again, the same was the case with the RTX 3050. The AD104 GPU, on which the RTX 4070 Ti is based, consists of 35.8 Billion transistors which is much higher than GA102 (RTX 3090 Ti).
Reviews will go live tomorrow and on the 5th of January, the RTX 4070 Ti will be available to consumers.
Performance
Coming to the performance metrics, NVIDIA claims that the RTX 4070 Ti is faster than the RTX 3090 Ti. In Cyberpunk 2077, using the RT Overdrive Mode the RTX 4070 Ti is up to 3x faster than the Ampere king.
Across a wide variety of games, the RTX 4070 Ti does actually manage to beat the RTX 3090 Ti. But, we will know for sure when this GPU is reviewed. Actually, the RTX 4070 Ti as per NVIDIA was slower than the RTX 3090 Ti when it was branded as the RTX 4080 12GB. So do not take these charts as a representation of the performance.
RTX 40 Mobility Lineup
We did tease the performance of NVIDIA’s upcoming Mobile GPUs above and let us reassure you that these GPUs are as good as they sound. In fact, many of the upcoming Ada laptop GPUs are almost on par with their Ampere desktop counterparts. Just let that sink in for a moment. We currently have 5 different SKUs unveiled by team green. Let us go over them one by one.
Bringing forth the Ada architecture, the RTX 40 Mobile GPUs are now available to the market. NVIDIA has announced 2 different categories of these laptops divided across 5 GPUs which we have touched on below.
RTX 4090M & RTX 4080M
Behold the RTX 4090M, the fastest mobile GPU in history. As dramatic as it may sound sounds, that is no exaggeration. The RTX 4090M houses 76 SMs which equates to 9728 Cuda cores. This GPU ships with 16GB of GDDR6 memory. The RTX 4090M is based on NVIDIA’s AD103 GPU and has a wattage that varies from 80-150W.
Next up, the RTX 4080M has 12GB of GDDR6 memory. Don’t let the lower memory count fool you as this GPU comes close to the desktop RTX 3080. Specifications-wise, this GPU is equipped with 58 SMs which translates to 7424 Cuda cores (128 Cuda cores per SM). The RTX 4080M ships with the AD104 GPU, which is also used on the RTX 4070 Ti.
The RTX 40 flagship laptops feature NVIDIA’s RTX 4090M and the RTX 4080M. Offering almost 2x faster video export than their predecessor, these laptops will be available starting at $1999 from February 8th. NVIDIA claims that these laptops can drive three 4K monitors and still deliver 60 FPS. That is simply insane, considering all of that is done under 200W of power (GPU).
RTX 4070M, RTX 4060M & RTX 4050
The RTX 4070M comes packed with 8GB Of GDDR6 memory. Moving over to the RTX 4060M, this GPU yet again features 8GB of GDDR6 memory. Both the RTX 4060 and the RTX 4070 have a base 115W profile which can go up to 140W. Furthermore, the RTX 4070M and the RTX 4060M pack 4608 and 3072 Cuda cores respectively.
Last, but not least, we have the RTX 4050M with 6GB of GDDR6 memory. This GPU has a TGP range of 115W–140W. However, for laptops aiming at power efficiency, expect the TGP to drop as low as 85W. The RTX 4070M is based on NVIDIA’s AD106 GPU, whereas the RTX 4060M and the RTX 4050M utilize the AD107 GPU.
These GPUs are faster than the last-gen flagship, but use almost 1/3rd of the power to offer the same performance. The next-gen laptops promoted by NVIDIA can power 80 FPS, 1440p ultra-gaming, and render scenes in Blender, which used to take two and a half hours, in just 10 minutes.
Performance
Undoubtedly, the Ada architecture is much faster than Ampere. However, team green once again boasts of their achievements and shows that Ada is up to 4x faster than the last generation.
With power, comes great responsibility. High performance at high power consumption can work for desktops. But laptops have the constraint of battery life. You probably don’t want your battery to die out at the wrong time.
NVIDIA claims that an unnamed mobile GPU based on Ada can perform similarly to the RTX 3070 while consuming almost 3x less power. That is quite the generational leap but the graph below does not show us the true story. In any case, we should probably wait for independent reviewers to verify this information.
The 5th Generation of Max-Q Technology
NVIDIA has introduced the 5th generation of its signature Max-Q technology with the RTX 40 Mobile GPUs. They claim that the Ada GPUs using this technology have the lowest GDDR6 memory voltage in the market.
- NVIDIA® DLSS 3 has been optimized for laptops and built into Max-Q technologies, including Whispermode and BatteryBoost, which improve performance, acoustics, and battery life by up to 2x.
- The lowest voltage GDDR6 memory ever, shipped in partnership with leading manufacturers.
- Tri-speed memory control enables the GPU to switch to newer, lower-power memory states dynamically.
- Ada’s on-chip memory has been optimized for Max-Q, doubling the bandwidth, increasing the size by 16x and improving clock gating.
Conclusion
NVIDIA announced many products ranging from the controversial RTX 4070 Ti to the insanely fast but efficient Ada GPUs. Despite receiving bad critics, the RTX 4070 Ti at $799 may be the relief people wanted. Of course, many will argue that $800 for a 70-class GPU indicates something wrong with the market itself. But, then again, this is the best next-gen offering available.
NVIDIA Ada Mobile Lineup
SKU | Codename | Chip | FP32/CUDA | Max Clock | FP32 Compute | Memory | Memory Bus | TGP |
RTX 4090 | X21-X11 | AD103 | 9728 | 1.45 – 2.04 GHz | 38.9 TFLOPS | 16GB | 256-bit | 80-150W |
RTX 4080 | X21-X9 | AD104 | 7424 | 1.35 – 2.28 GHz | 33.8 TFLOPS | 12GB | 192-bit | 60-150W |
RTX 4070 | X21-X6 | AD106 | 4608 | 1.23 – 2.175 GHz | 20.0 TFLOPS | 8GB | 128-bit | 35-115W |
RTX 4060 | X21-X4 | AD107 | 3072 | 1.47 – 2.37 GHz | 14.6 TFLOPS | 8GB | 128-bit | 35-115W |
RTX 4050 | X21-X2 | AD107 | 2560 | 1.61 – 2.37 GHz | 12.1 TFLOPS | 6GB | 96-bit | 35-115W |