Nvidia launches a neutered version of the RTX 4070 Ti graphics chip. And it has a lot of steam, as the test of the RTX 4070 shows.
With the Geforce RTX 4070, Nvidia is now introducing the fourth graphics chip from the RTX 4000 series developed under the code name “Ada Lovelace”. The test shows: Compared to the RTX 4070 Ti, the variant without Ti works slower – but it is also significantly cheaper. But first things first: Here in the test is one of the first finished graphics cards with the new chip – the Zotac Geforce RTX 4070 Trinity.
RTX 4070 chip size and number of transistors
The internal model designation of the RTX 4070 is identical to that of the Ti version: AD 104. Nvidia uses the identical chip, but disconnects entire groups of computing units. The RTX 4070 chip, which is 295 square millimeters in size, also contains a whopping 35.8 billion circuits (transistors), but not all of them are activated. Because in contrast to the Ti version, the RTX only has 5,888 instead of 7,680 CUDA computing units. And Nvidia distributes them to 46 streaming multiprocessors (SM) with 128 Cuda cores each.
Lower clock speeds
The Cuda cores work on the RTX 4070 with lower clock rates: Instead of 2,310, the non-Ti version works with only 1,920 megahertz. In addition, there are 14 fewer ray tracing units available for calculating reflections and fine shadows. At 184, the number of tensor cores is also significantly lower. You are responsible for Deep Learning Super Sampling (DLSS) version 3.0. This technique can self-generate entire frames in customized games such as ”
A Plague Tale: Requiem “. To do this, the RTX 4070 analyzes the content of existing frames, including movements, particles, reflections and light. An artificial intelligence (AI) uses this to create the extra image.
Nvidia Geforce RTX 4000 Series Specifications
| Geforce RTX 4090 | Geforce RTX 4080 (16GB) | Geforce RTX 4070 Ti | Geforce RTX 4070 | |
| production | 4 nanometers at TSMC | 4 nanometers at TSMC | 4 nanometers at TSMC | 4 nanometers at TSMC |
| GPU type | AD102 | AD103 | AD104 | AD104 |
| chip size | 608mm² | 379mm² | 295mm² | 295mm² |
| transistors | 76.3 billion | 45.9 billion | 35.8 billion | 35.8 billion |
| shader models | 128 | 76 | 60 | 46 |
| cuda cores | 16,384 | 9,728 | 7,680 | 5,888 |
| base clock | 2,230MHz | 2,210MHz | 2,310 MHz | 1,920 MHz |
| boost clock | 2,520MHz | 2,510MHz | 2,610 MHz | 2,475 MHz |
| tensor cores | 512 (4th Gen.) | 304 (4th Gen.) | 240 (4th Gen.) | 184 (4th Gen.) |
| Ray tracing cores | 128 (3rd Gen.) | 76 (3rd Gen.) | 60 (3rd Gen.) | 46 (3rd Gen.) |
| texture units | 512 | 304 | 240 | 184 |
| graphics memory | 24GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | 12GB GDDR6X |
| memory interface | 384 bits | 256 bits | 192 bits | 192 bits |
| TDP | 450 watts | 320 watts | 285 watts | 200 watts |
Same memory and interface
The memory, on the other hand, is identical: As with the RTX 4070 Ti, it is quite generous with 12 gigabytes for the fast swapping out of calculated data. Data is exchanged between memory and chip via 192 conductor tracks (192-bit memory interface). This also corresponds to the Ti variant – as does the memory clock of 1,313 megahertz.
20 percent faster than RTX 3070
Even if many key data are identical to those of the Ti version, the tested Zotac Geforce RTX 4070 Trinity cannot make up for the significantly lower number of computing units. Elaborate calculations can be spread over fewer shoulders, the pace was 17 percent slower than the Ti model (comparison below). But the Zotac Geforce RTX 4070 Trinity was by no means sluggish: it still showed games smoothly even in the exhausting 4K format with 68 frames per second, when gaming with ray tracing effects turned to the stop it was still almost jerk-free 44 frames per second. Compared to the predecessor RTX 3070 (at the time of testing from 550 dollors), the new Zotac Geforce RTX 4070 Trinity was a whopping 20 percent faster.
Works pleasantly quiet
Also good: The Zotac worked completely silently with simple internet and office applications. Under full load in performance-hungry game scenes with transparent smoke, many reflections and a sea of flames, the three fans on the 34 centimeter long graphics card whispered to themselves with a tolerable volume of 1 sone. The Zotac’s trio of fans could only really be brought out of the reserve with synthetic benchmark programs such as 3DMark: Then it produced a hell of a noise with more than 10 sone. But that doesn’t matter in practice.
Moderate energy requirement
Much more important in practice: the energy requirement. And with 195 watts it was at a tolerable level. For comparison: The top model of the RTX 4000 series, the RTX 4080, consumes more than twice as much at 440 watts.
In the test, the Zotac Geforce RTX 4070 Trinity offered enough performance to smoothly play current games in 4K with full detail reproduction. Compared to the larger RTX 4070 Ti, it was 17 percent slower, but at the time of the test it was 26 percent cheaper at a price of 660 dollars- a real selling point. Because the Zotac Geforce RTX 4070 Trinity currently offers the best price-performance ratio of the RTX 4000 series.
