GeForce RTX 4070 Ti is a very interesting representative of the new line of NVIDIA video cards based on chips with the Ada Lovelace architecture. This model has a rather intriguing appearance history, a good balance of technical specifications and a decent level of performance, as for a representative of the RTX xx70 series. We will familiarize ourselves with the capabilities of the new line using the example of ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC – a forced version with an original cooling system.
NVIDIA GeForce RTX 4070 Ti
The history of the appearance of the GeForce RTX 4070 Ti is full of drama, controversy, and sharpening of consumer passions. Initially, the model was announced at the end of September last year, simultaneously with the flagship GeForce RTX 4090 24 GB and GeForce RTX 4080 16 GB and was supposed to go on sale in November under the name GeForce RTX 4080 12 GB. However, the significant difference in technical specifications from those for the older modification rightly raised a lot questions and even outrage towards the developers.
The wave of dissatisfaction turned out to be so powerful that NVIDIA even decided to refuse from the appearance of the GeForce RTX 4080 12 GB on the market, canceling the announcement. At the same time, the company assured that the graphics card is excellent, but has an inappropriate name. So it was obvious that the sudden “unlaunch” of the adapter is a temporary phenomenon.
Indeed, at the beginning of January during the CES 2023 exhibition, the GeForce RTX 4070 Ti 12 GB graphics card was announced, which is technically identical to the previously canceled GeForce RTX 4080 12 GB model. The difference is only in the name and recommended price. Instead of the initial $899, after the renaming, the developers decided to lower the starting price to $799.
|GeForce RTX 4090||GeForce RTX 4080||GeForce RTX 4070 Ti||GeForce RTX 3070 Ti||GeForce RTX 3090 Ti|
|Architecture||Ada Lovelace||Ada Lovelace||Ada Lovelace||Ampere||Ampere|
|Technology||4 nm, TSMC||4 nm, TSMC||4 nm, TSMC||8 nm, Samsung||8 nm, Samsung|
|CUDA cores||16 384||9728||7680||6144||10 752|
|Tensor cores||512 (4-generation)||304 (4-generation)||240 (4-generation)||192 (3-generation)||336 (3-generation)|
|RT-cores||128 (3-generation)||76 (3-generation)||60 (3-generation)||48 (2-generation)||84 (2-generation)|
|GPU clock rates||2.24/2.52 GHz||2.21/2.51 GHz||2.31/2.61 GHz||1.575/1.77 GHz||1.56/1.86 GHz|
|RAM||24 GB GDDR6X||16 GB GDDR6X||12 GB GDDR6X||8 GB GDDR6X||24 GB GDDR6X|
|Memory bus||384 bit||256 bit||192 bit||256 bit||384 bit|
|Effective memory frequency||21 000 MHz||23 000 MHz||21 000 MHz||19 000 MHz||21 000 MHz|
|Memory bandwidth||1008 GB/s||736 GB/s||504 GB/s||608 GB/s||1008 GB/s|
|Graphics card power consumption (GCP)||450 W||320 W||285 W||290 W||450 W|
As for the technical equipment, unlike the GeForce RTX 4080 based on the AD103 chip, the GeForce RTX 4070 Ti graphics card uses a more compact AD104 graphics processor, which of course also uses the Ada Lovelace architecture. The GPU is manufactured by TSMC using a 4-nanometer process, has a crystal area of 295 mm2, and contains 35.8 billion transistors.
The full AD104-400 modification includes 7680 CUDA cores, 240 texture units, and 80 rasterizers. In addition, the GPU received 60 RT cores for processing ray tracing and 240 tensor cores for vector computations. As in the case of older models of the RTX 4000 series, the new GPU has a rather capacious L2 buffer of 48 MB. The GPU has a base frequency of 2310 MHz with an average dynamic acceleration of up to 2610 MHz. Claimed computing performance is at 40 TFLOPS (FP16/FP32). The flagship on the Ampere architecture – GeForce RTX 3090 Ti – has the same indicators.
The GeForce RTX 4070 Ti comes with 12GB of GDDR6X memory clocked at an effective 21,000MHz. The bit rate of the bus is 192 bits, so the total bandwidth of the memory subsystem is 504 GB/s. The indicator is not particularly impressive, especially against the background of the values of the previous generation models. For example, even the GeForce RTX 3070 Ti has a 256-bit bus and RAM at 608 GB/s. However, this is only one of the indicators that can affect the overall performance of the video card.
GeForce RTX 4070 Ti received a TDP of 285 W. This is a fairly moderate indicator for solutions of this class. We will remind you that for the older GeForce RTX 4080 model, power consumption is declared at the level of 320 W, the conditional predecessor GeForce RTX 3070 Ti – 290 W, and the GeForce RTX 3090 Ti with which the developers compare the new product – 450 W.
In the case of the GeForce RTX 4070 Ti, NVIDIA does not offer reference models, relying solely on offers from partners. The latest hobbyists have offered many variants of the GeForce RTX 4070 Ti, anticipating the increased demand for devices of this particular line. One of these graphics cards made it to us for review.
ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC
At the start, ASUS offered four versions of the GeForce RTX 4070 Ti, two models each in the STRIX and TUF Gaming series. We tested the OC modification of the TUF line with a small regular acceleration of the graphics processor.
The graphics card is delivered in a fairly large box. The manufacturer supplies an adapter from 12VHPWR to two 8-pin connectors for additional power, as well as an additional rack that can be transformed into a compact screwdriver with a cross-shaped tip.
TUF series components need no further introduction. Although the devices of the TUF Gaming line have somewhat moved away from the original brutal military style in design, the graphics adapters of this series are easy to recognize. It’s a block with a rather restrained external design. A graphite-colored metal casing covers the cooling system.
Judging by the name ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC, it is not difficult to predict that the graphics card received factory acceleration. Indeed, instead of the recommended 2310/2610 MHz, the graphics processor received a formula of 2310/2730 MHz. That is, the average overclocking rate has been increased by a significant 120 MHz.
The video card has a 3.25-slot design. However, in general, it is not perceived as a giant, as is the case with the top adapters of the new series. The dimensions of the model are 305 x 138 x 65 mm, the weight is almost 1.4 kg. Smaller dimensions reduce the requirements for the case, but it doesn’t hurt to make sure that there will be enough internal space beforehand, especially when it comes to compact mATX cases.
The graphics card cooling system includes two massive cassettes with sets of aluminum plates. Six heat pipes are provided in the design, helping to accelerate the transfer of heat to remote areas of the radiator.
Additional cooling is used both for power elements and for memory chips with which the thermal plate contacts through thermal conductive pads.
Three 90-mm axial-type fans are provided for blowing the structure. Models with double ball bearings are made according to the Axial-tech technology with the combination of blades along the outer radius. In addition, the blades themselves have a rather specific profile.
A multidirectional rotation system is technologically provided. The side fans rotate counterclockwise, while the central fan rotates clockwise. In this way, the effect of the possible emergence of turbulence zones is reduced. For the cooler, a hybrid mode of operation is provided, with fans stopping at GPU temperatures below 50C and starting them when the chip heats up to +55C.
The ASUS TUF Gaming GeForce RTX 4070 Ti 12GB OC uses an original PCB with a length of about 240mm. According to the proposed configuration of the power subsystem for the GPU, 11 phases are allocated on Vishay SiC639 DrMOS elements with maximum currents up to 50A, which are controlled by the uPI controller uP9512R.
Only two phases are allocated for memory chips. The same power assemblies are used here, while the parameters are controlled by the uPI uP9529Q chip.
GDDR6X memory chips in this case are represented by Micron D8BZC chips (MT61K512M32KPA-21:U), which, according to the specification, support operation at an effective 21,000 MHz.
Interestingly, the PCB is designed with the potential for the adapter to have a 256-bit bus, and you can see empty areas for elements of three more power phases. One might assume that these are blanks for the GeForce RTX 4080, but for older models, ASUS uses boards of a completely different design. So, the question remains open.
Returning to the ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC, it should be noted that the length of the board is much smaller than the overall dimensions of the adapter. An additional “window” in the design allows you to implement through-blowing of the radiator block.
In the upper right corner of the PCB, there is a 12VHPWR connector for connecting additional power. Despite the fact that we are talking about models of an alternative design, all RTX 4000 solutions are offered with this power connector.
The range of power supplies with standard 12VHPWR is gradually replenished, but for other cases, it is quite possible to use a supplied adapter. In this case, two 8-pin connectors are sufficient for connecting power lines.
A signal LED is provided for the connector, which will warn about the lack of connection to the power line. In the case of GeForce RTX 4070 Ti, there will probably be no problems with overheating of 12VHPWR, however, when connecting, you should not neglect the recommendations by tightly attaching the connector.
As for the additional equipment of the graphics card, it should be noted that there is a switch on the upper edge of the PCB for selecting the operating mode of the cooling system. Performance Mode is used by default, and Quiet Mode position with reduced cooling fan speed is also provided.
On the reverse side, the graphics card is covered with an aluminum plate. It’s an appropriate element of decoration, protection against mechanical damage, and even an additional radiator for heat dissipation. As we have already noted, beyond the placement of the printed circuit board in the plate, a through hole is provided for blowing by a cooling system fan. Due to the additional air circulation under the metal casing, zones of increased heating are not created.
Three DisplayPort 1.4a ports are displayed on the interface panel, as well as a pair of new format HDMI 2.1a. An additional HDMI output expands the possibilities of the display configuration and will definitely not be superfluous.
A part of the panel is occupied by a perforation for the partial removal of heated air outside the system unit.
Those who love military style also sometimes do not mind additional lighting, so the developers have foreseen certain possibilities for illumination. On the top panel from the edge of the adapter, there is a small segment, as well as the TUF series logo with ARGB illumination. Support for addressing allows you to diversify the available effects. Of course, AURA Sync makes it possible to synchronize the backlight with that of other components of the platform.
In idle mode, the cooling system fans stop. In such conditions, on an open stand, the GPU temperature was maintained at 37C (45C hotspot), and the memory chips warmed up to 38C.
According to GPU-Z monitoring, ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC power consumption under minimum load was only about 8 W.
During gaming sessions, the graphics processor warmed up to 66C with a temperature of 77C at the hottest point of the crystal. A value of 66C was also fixed on the memory chips. In the regular Performance Mode under high load, the speed of rotation of the cooler fans increased to almost 1500 rpm. The overall noise level is below average, but the work of the adapter is felt, especially if a quiet processor cooling system is used.
The graphics processor was maximally accelerated to almost 2900 MHz, however, the average operating values under the game load are slightly lower – 2750-2830 MHz.
We also measured the parameters of the graphics card by selecting Quiet Mode. In this case, the speed of rotation of the cooling system fans was limited to 1250 rpm. How did it affect the temperature parameters of the elements? The graphics processor heated up to 68C with a peak of 80C. The temperature of the memory chips also rose by 2 degrees to 68C. Subjectively, the noise level has decreased somewhat, but the difference cannot be called fundamental. In both cases, the graphics card works quite quietly.
As for the difference in performance, there is none. As you can see, switching to Quiet Mode does not affect the frequency parameters of the GPU. The graphics processor was maximally accelerated to the same level as in the case of Performance Mode – 2895 MHz.
Test bench configuration
- Motherboard: ASUS TUF GAMING B650-PLUS WIFI (AMD B650)
- Processor: AMD Ryzen 9 7900X (12/24; 4.7/5.6 GHz)
- Cooling: ASUS ROG RYUJIN II 360
- Memory: Kingston Fury Beast RGB DDR5-6000 32 GB (KF560C36BBEAK2-32 2×16 GB)
- Video card: ASUS TUF Gaming GeForce RTX 4080 OC 16 GB (TUF-RTX4080-O16G-GAMING)
- Drive: Samsung 980 Pro 1 TB
- PSU: ASUS ROG STRIX 1000W GOLD (1000 W)
During the preparation of the review, a platform based on a 12-core processor AMD Ryzen 9 7900X with a motherboard ASUS TUF GAMING B650-PLUS WIFI on an AMD B650 chipset was used. It is this set that we recently reviewed in more detail. The system was also equipped with 32 GB (2×16 GB) of RAM with DDR5-6000 mode optimal for Socket AM5 solutions. The dual-channel Kingston Fury Beast RGB DDR5-6000 32GB (KF560C36BBEAK2-32) kit with support for the AMD EXPO profile was directly involved. The Samsung 980 Pro 1 TB drive was responsible for saving data, and the ASUS ROG STRIX 1000W GOLD block provided stable power.
For a direct assessment of the capabilities of the new graphics card, we chose the older model of the GeForce RTX 4080 16 GB series as a starting point – ASUS TUF Gaming GeForce RTX 4080 OC 16 GB. In this way, we will be able to confidently answer the question of maintaining a balance between performance and price for both models.
We started the productivity evaluation by checking the capabilities of the graphics card in the 3DMark test. A set of synthetic steps can be especially useful when you need to compare the performance of solutions with the same architecture. It was for this purpose that the resource-intensive stages of Time Spy Extreme were involved, as well as subtests using ray tracing – Port Royal and Speed Way.
ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC specifications
ASUS TUF Gaming GeForce RTX 4080 16 GB OC specifications
The final results are presented in the corresponding screenshots. As we can see, in the listed stages, GeForce RTX 4070 Ti 12 GB is inferior to GeForce RTX 4080 16 GB by about 21-25%. The difference is quite noticeable, so it is quite logical and justified that the younger model was not named GeForce RTX 4080 12 GB. In the latter case, it would be misleading users. Whereas for models that are on different steps in the graphical hierarchy, the difference is acceptable. But this is synthetic, let’s look at the situation with real games.
After testing a whole set of quite resource-intensive projects in 1080p mode with maximum image quality, certain generalizations can already be made. According to the results, the ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC is on average 12% slower than the ASUS TUF Gaming GeForce RTX 4080 OC 16 GB. The difference depends on the specific game and in some cases it can be about 20% (Cyberpunk 2077, Dirt 5, Dying Light 2), and in others, it does not exceed 5% (Far Cry 6, Shadow of the Tomb Raider).
When using the mode with a resolution of 2560×1440, the advantage of the older model increases to 17% on average. In certain projects, the difference grows quite actively, in others it remains almost at the same level as in the case of Full HD. Overall, the GeForce RTX 4070 Ti performs very well at 1440p. We got 120-150+ fps most of the time. Among the exceptions, only the already mentioned “heavyweight” Cyberpunk 2077 and Ghost Recon Breakpoint, which at the maximum quality settings lose significantly in frames/s. But even in the last case, we have 95 fps. With such indicators, monitors with an increased refresh rate (144-165+ Hz) will definitely not sit idle.
When switching to 4K, the lag of the younger model already increases to an average of 22%. Obviously, the RTX 4070 Ti has less memory bandwidth compared to the RTX 4080 – 504 vs. 717 GB/s. However, the performance of the graphics card in the vast majority of cases still allows you to get more than 60 fps at the maximum quality settings. The exceptions are the same – Cyberpunk 2077 and Ghost Recon Breakpoint, which manage to get “only” average 50-55 fps. That is, the 4K mode is quite suitable for the GeForce RTX 4070 Ti, you can just feel more comfortable with the RTX 4080. But here is the time to mention the one and a half times difference in the recommended price of graphics cards ($799 vs. $1199), so that you can agree on the need to periodically play with the settings for optimal performance.
If you try to compare the performance of the GeForce RTX 4070 Ti with those of the previous generation graphics cards, it turns out that the new product is on average 5% more productive in games than the GeForce RTX 3090 Ti 24 GB at 1080p, has very similar performance at 1440p and is 5-10% inferior the flagship of the RTX 3xxx series when using the 4K mode. In our opinion, it is precisely such indicators that allowed NVIDIA to significantly increase the prices of not only the top models but also models a step lower in the new series.
If you try to extrapolate the indicators to compare the GeForce RTX 4070 Ti with its conditional predecessor – the GeForce RTX 3070 Ti, the new product will be ~40% more productive in all modes. This is a very noticeable increase in fps, which allowed the model on the AD104 to actually rise in class. But the recommended price of the titanium “seventy” has increased by $200.
The confrontation between GeForce RTX 4070 Ti 12 GB and Radeon RX 7900 XT 20 GB will be very interesting. Each of these graphics cards has its advantages and disadvantages. A model with an AMD chip will almost certainly get a ~10% advantage when using classic rendering, while the NVIDIA adapter will be more agile in tracing and has support for DLSS 3. In addition, the recommended price of the GeForce RTX 4070 Ti 12 GB is $100 lower, and the models in the series are already widely sold, despite the later announcement which can’t be said about Radeon RX 7900 XT yet. However, such a “versus” definitely deserves more detailed, separate coverage.
In order to check the capabilities and the balance of power in modes with activated ray tracing, we used two projects – Metro Exodus EE and Dying Light 2. In general, when using the maximum RT settings without activating DLSS, the difference in indicators was 18-24% in favor of the RTX 4080. Moreover, the relative dependence on the distribution capacity was minimal here.
With intelligent scaling, the fps is noticeably increased, and the difference between the two video cards in Dying Light 2 is significantly reduced.
Another real test of the capabilities of video cards will be Portal with RTX, which is actually created to demonstrate the potential of RTX 4000 series adapters with DLSS support.
Unfortunately, Portal with RTX does not have a built-in benchmark, and third-party utilities were unable to determine the average FPS during the game, so to compare the performance of video cards in different modes, we had to use the least accurate indicator – the current FPS in the Steam overlay. It adequately represents the difference on the same bench when used in the same scene, but we cannot compare our figures with those obtained by other resources in their own reviews, even with a similar configuration of the test bench.
As for indicators, in general, the balance of power remained unchanged. There were no surprises here. However, one notable exception should be mentioned. We are talking about 4K mode without activating DLSS technology. At Ultra and even Medium picture quality settings, we received a slideshow with a performance of 3-5 fps on the GeForce RTX 4070 Ti video card, while the GeForce RTX 4080 provided indicators that were also too low for a comfortable game, but significantly higher – 16-23 fps.
Given the identity of the GPU graphics architecture of both video cards, the five-fold difference can only be explained by the limitations and lack of local memory capacity on the RTX 4070 Ti. Indeed, the primary assumptions were confirmed during the monitoring of system parameters in GPU-Z. Switching to 4K mode radically increases the required memory capacity. Although the diagnostic utilities display the reserved, and not actually used, capacity, in the case of Portal with RTX the situation is quite clear – 12 GB for 4K is not enough here. The game can actually be called a big technological demo, but still, as we can see, such situations cannot be completely excluded.
When using DLSS, the urgent need for the memory capacity of the graphics card is reduced. This is logical because in such a case the GPU transforms the “picture” of a smaller size with subsequent scaling. So after activating DLSS, about 8.8 GB of memory was already reserved during the Portal game with RTX, and the difference in the performance of the video cards returned to the expected limits of 20-25%.
For GeForce RTX 4070 Ti 12 GB video cards, the declared TDP is 285 W. For comparison, the RTX 4080 has a 320W package. The top models of the previous generation have higher energy consumption. The GeForce RTX 3090 24 GB has a TDP of 350 W, and the flagship GeForce RTX 3090 Ti 24 GB, with which the new product competes on an equal footing in games, has a total of 450 W.
As for the direct performance of the ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC, the video card consumed up to 280 W in Performance Mode and about 285 W when switching to Quiet Mode.
The power consumption of the predecessor GeForce RTX 3070 Ti is about 300 W with a declared TDP of 290 W. Taking into account the fact that the RTX 4070 Ti has a 40% higher performance, it is easy to calculate how much the energy efficiency of video cards on GPUs with the Ada Lovelace architecture has improved.
But if you compare the indicators within the new line, the performance per watt ratio will obviously be better in the GeForce RTX 4080. The GeForce RTX 4070 Ti model is 11-12% more economical, while the difference in performance is twice as large, at least in 4K mode. For the ratio to even out, the TDP should be at the level of 250 W. And this is quite expected, taking into account the fact that the GeForce RTX 4070 Ti uses a more compact GPU, the bus bit rate is smaller, and, accordingly, the number of memory chips. At the same time, the working frequency of the graphics processor of the younger model is about 100 MHz higher. Let’s assume that this requires more energy consumption.
In order to check the overclocking capabilities of the video card at a basic level, we minimally experimented with the frequency limits for the GPU and memory chips. First of all, we note that in the case of ASUS TUF Gaming GeForce RTX 4070 Ti 12 GB OC, the Power Limit parameter can be increased by 10%.
Our sample video card showed quite good results. The base frequency of the GPU was increased from 2310 MHz to 2520 MHz, while the value of the average acceleration increased accordingly from 2730 MHz to 2940 MHz (+7.7%). The GDDR6X memory chips also accelerated from 1313 MHz to 1450 MHz without sacrificing stability. That is, effective frequencies increased from 21,008 MHz to 23,200 MHz (+10.4%).
After additional overclocking, the GPU was automatically accelerated under load to 3105 MHz. That is, the symbolic and previously unattainable 3 GHz remains far behind. The power consumption of the video card after afterburning increased from 280 W to 300 W.
Usually, additional amateur overclocking of top video cards does not bring a noticeable increase in performance. However, good acceleration of both the GPU and memory chips, in this case, ensured an increase in 3DMark indicators by 7%. This is already much more than the measurement error, but certainly not enough to even theoretically catch up with the older GeForce RTX 4080 16 GB model. Current RTX 4000 series adapters have a clear classification, excluding possible internal competition of solutions neighboring in the hierarchy.