Despite the fact that the PC component market still has a shortage of GeForce RTX 50 series graphics cards, the range and number of offers for models with GPUs with Blackwell architecture is gradually increasing. The increase in inventory allows to calm the initial hype demand and leads to a gradual price normalization. Usually this is how it worked. But this time the situation in the discrete graphics segment is quite specific, and long-term forecasts remain uncertain. We continue to get acquainted with new products, assessing their capabilities. This time the focus is on the MSI GeForce RTX 5070 Ti VANGUARD SOC Launch Edition.
NVIDIA GeForce RTX 5070 Ti
The GeForce RTX 5070 Ti uses the same Blackwell architecture GPU as the older GeForce RTX 5080, the GB203. The GB203-200 chip has 8,960 active compute units, compared to 10,752 in the GB203-300 version of the RTX 5080. The GPU has an area of 378 mm² and contains 45.6 billion transistors. The chip has a base clock speed of 2,295 MHz with an average boost of 2,452 MHz.
| GeForce RTX 5070 Ti | GeForce RTX 4070 Ti SUPER | GeForce RTX 4070 Ti | |
| Graphics processor | GB203-300 | AD103-275 | AD104-400 |
| Architecture | Blackwell | Ada Lovelace | Ada Lovelace |
| Technology | 5 nm, TSMC | 5 nm, TSMC | 5 nm, TSMC |
| CUDA computers | 8960 | 8448 | 7680 |
| TMU/rasterizers | 280/128 | 264/96 | 240/80 |
| Tensor kernels | 280 (5th generation) | 266 (4th generation) | 240 (4th generation) |
| RT blocks | 70 (4th generation) | 66 (3rd generation) | 60 (3rd generation) |
| GPU clock speeds | 2295/2452 MHz | 2340/2610 MHz | 2310/2610 MHz |
| Memory capacity | 16 GB GDDR7 | 16 GB GDDR6X | 12 GB GDDR6X |
| Memory bus | 256 bits | 256 bits | 192 bit |
| Effective memory frequency | 28,000 MHz | 21,000 MHz | 21,000 MHz |
| Memory bandwidth | 896 GB/s | 672 GB/s | 504 GB/s |
| Graphics card power consumption (TGP) | 300 W | 285 W | 285 W |
| Starting recommended price | $749 | $799 | $799 |
| Release date | February 2025 | January 2024 | January 2023 |
The GeForce RTX 5070 Ti retained the 256-bit bus and received 16 GB of memory. The video card uses GDDR7 chips with an effective operating frequency of 28,000 MHz. In this configuration, the memory bandwidth is 896 GB/s. The declared total power consumption level is 300 W, and for a system with this video card it is recommended to use power supplies with a capacity of 750 W or more. As for the recommended price, the MSRP for the basic modifications of the GeForce RTX 5070 Ti is $749. A potentially attractive value, but it is worth considering that this is an indicator for the American market without taking into account taxes, and even in that region the original versions of the GeForce RTX 5070 Ti are initially offered significantly more expensive.
MSI GeForce RTX 5070 Ti VANGUARD SOC Launch Edition
The launch of the next generation of graphics cards is a great opportunity for manufacturers to review and optimize their range, to make certain accents in an effort to attract the widest possible contingent of potential buyers. With the advent of the GeForce RTX 50, MSI has diversified its offers as much as possible with original options. It even went so far as to launch a whole new line. So, we are talking about the VANGUARD series models, which use virtually the same platform as the premium SUPRIM solutions, but implement visually bolder ideas regarding the external design of the devices.
Today we are reviewing the MSI GeForce RTX 5070 Ti VANGUARD SOC model in the Launch Edition version, which will be available in limited quantities after the launch of the new series, complete with a Lucky dragon figurine.
The video card is offered in a huge box with specific packaging. In addition to the video card itself, the kit includes an adapter from three 8-pin PCI-E 12V connectors to one 12VHPWR, a height-adjustable stand, and the aforementioned dragon figurine, which has one of nine images.
MSI GeForce RTX 5070 Ti VANGUARD SOC received a standard GPU overclock. Instead of the recommended frequency formula of 2295/2452 MHz, the chip has 2295/2588 MHz. The memory chips meet the specifications - effective 28,000 MHz.
At first glance, the video card has common features with the MSI GeForce RTX 5080 16G SUPRIM SOC model, which we recently reviewed. At least, it is about an almost equally chunky graphics adapter with a huge cooler and three fans. In terms of its dimensions (357x151x66 mm) and weight of almost 2 kg, it is only slightly inferior to the top model, which additionally evokes a feeling of similarity. However, in the details, the individual feature of the MSI GeForce RTX 5070 Ti VANGUARD SOC is also enough. The asymmetry of the protective casing with carbon-textured inserts adds aggressiveness to the appearance.
The graphics card is equipped with a modified Hyper Frozr cooling system with a very massive radiator block, a large evaporation chamber and heat pipes with a special shape to increase the contact area. The radiator cassette consists of aluminum plates with a specific wave-like shape of the Wave Curve 4.0 profile.
To blow the radiator, three 105 mm STORMFORCE fans with blades that are connected along the outer radius by a continuous circular arc are used. The surface also has specific fins. The developers assure that this combination allows you to increase the air flow, as well as make it more accentuated.
The graphics card uses a printed circuit board, which is very similar in layout to the PCB we saw in the recently tested MSI GeForce RTX 5080 SUPRIM SOC model. In general, such unification is a common phenomenon, especially when it comes to adapters on the same GPU. Here we are also dealing with GB203, although in a modification with a smaller number of computing units. As for the schematic simplification, it is minimal. A 15-phase circuit is used for the GPU power stabilizer, while the RTX 5080 had 16. But otherwise the parameters are almost identical.
The GPU circuit uses MP87993 (DrMOS) power assemblies with a maximum operating current of up to 50 A. The MP29816 controller is used to control the circuit. Identical elements (MP87993) are used for the power phases of the memory chips, which are controlled by the MP2988 controller.
As for the GDDR7 chips themselves, in this case the video card is equipped with Samsung K4VAF325ZC-SC28 chips, which according to the specification can operate at an effective 28,000 MHz.
To connect additional power, there is one 16-pin 12VHPWR connector on the top edge of the video card. The GeForce RTX 5070 Ti video cards have a declared total power consumption of 300 W. A power supply with a capacity of 750 W or more is recommended for PCs with such models. MSI developers have kept NVIDIA's general recommendations relevant for their tuned modification. Despite the GPU overclocking, the adapter's energy "appetites" remain unchanged. Here we only note that the MSI Dual Color 16-pin Cable adapter with a yellow connector is used for the MSI GeForce RTX 5070 Ti VANGUARD SOC. If the bright part of the connector is not visible after connection, it means that the cable is connected correctly and the contact is tight enough.
It has become a certain rule of good taste to have two BIOS chips in mid-range and high-end graphics cards. MSI GeForce RTX 5070 Ti VANGUARD SOC also offers a similar option. On the top edge there is a two-position toggle switch that allows you to select modes - Silent and Gaming. Judging by the name, one of them provides a lower noise level, the other offers a more aggressive algorithm for the cooling system, allowing you to further reduce the heating of the graphics card elements. We will return to these features during practical tests. We note in advance that the manufacturer initially offers the Silent mode. Usually this is an additional option, but if the developers are confident in the efficiency of the cooling system, then with unchanged performance, the noise level gets a higher priority.
MSI GeForce RTX 5070 Ti VANGUARD SOC is decorated with additional ARGB lighting. On the top of the video card there is a large trapezoidal nameplate with a transparent light guide in the thickness of which there is a laser engraving with the logo of the game series. Support for addressable illumination allows you to get non-trivial lighting effects. The silhouette of the dragon is also illuminated on the side panel, which can usually be clearly seen through the transparent wall of the case.
Special attention is paid to the side lighting. Additional segments of light guides located along the diagonal create backlighting, outlining the contours of the fans. So even if the video card is placed vertically in the case, the lighting will "work".
There is a protective metal plate on the back of the video card. Contact through a heat-conducting sticker also helps to partially remove heat from the GPU area. Special attention has been paid to the decor of the plate. The surface is decorated with a large dragon logo, the profile has a complex geometric shape and carbon inserts. Functional holes provide through-flow of the radiator block. The ratio of the length of the printed circuit board and the overall dimensions of the adapter allow for a high-quality implementation of such a concept.
The composition of ports on the interface panel is standard - three DisplayPort 2.1b and one HDMI 2.1b. A significant part of the cover is occupied by through holes to improve the removal of heated air outside the system unit.
Test system configuration
- Processor: Ryzen 7 9700X (8/16; 3.8/5.5 GHz; 32 MB L3)
- Motherboard: MSI MPG X870E CARBON WIFI
- CPU Cooling: MSI MAG CORELIQUID I360
- Graphics card: MSI GeForce RTX 5070 Ti VANGUARD SOC
- RAM: Kingston FURY Renegade DDR5-7200 2x16 GB (KF572C38RWK2-32)
- Drive: Kingston KC3000 (SKC3000S/1024G)
To test the video card, a platform based on the Ryzen 7 9700X processor (8/16; 3.8/5.5 GHz; 32 MB L3) and the MSI MPG X870E CARBON WIFI motherboard on the AMD X870 chipset were used. The eight-core chip is objectively a good option for a gaming system. Of course, for top-end solutions, X3D chips with an additional L3 cache buffer would not be superfluous, but the Ryzen 7 9700X is significantly more affordable, so it will not be a significant limiter, especially for 1440p and 4K modes. The current retail price of the Ryzen 7 9700X in Ukraine is from 14,500 UAH (~$350). However, it is worth recalling that even the BOX-package of these chips does not offer a standard cooler - the cooler will have to be purchased separately.
To increase the computing power of the Ryzen 7 9700X, we used a TDP of 105 W, instead of the standard 65 W. AMD officially confirmed that using such a mode is standard for this model, and MSI and other motherboard manufacturers have made it as easy as possible to switch between "65 W" and "105 W".
Switching to the appropriate mode requires changing only one parameter in the BIOS settings. In games, we are unlikely to see a significant difference after increasing the power limit, but under maximum multi-threaded load, the performance gain is ~15%. So during gaming tests, this is more of a safety net.
If we increase the TDP, then the cooling system should be relatively efficient. Here we have already relied on the liquid MSI MAG CORELIQUID I360. Probably a bit excessive for an 8-core processor, but there is never too much efficiency in the SRO.
In operation
In idle mode with minimal load on the GPU, the video card stops the fans.
In such conditions, the temperature of the graphics processor was kept at 43C, the memory chips at 54C. The adapter's silent operation during typical office tasks is already a familiar option for devices of this class. At rest, the graphics processor reduced its operating frequency to 195 MHz, and the memory chips operated at 50 MHz (800 MHz effective).
How does the cooler cope with cooling under load? First, let's evaluate the standard mode, and in this case it is Silent. During a long gaming session, the GPU temperature rose to 61C, while the GDDR7 memory chips heated up to 66C. At the same time, the fan speed increased only to ~1100 rpm. The video card worked very quietly, sometimes even unusually quietly for a model of this level.
We also checked the operation of the video card by switching the toggle switch to Gaming mode. After such manipulation, the fan speed at the peak increased to ~1380 rpm, while the temperatures of the elements decreased by an additional 3–4C. Under load, the graphics processor warmed up to only 58C, and the temperature of the memory chips did not exceed 62C. The noise level increased slightly, although the video card still remained in the "quiet" category.
Fans of silence can certainly stay in Silent mode. Given the small difference in temperatures, there is no particular point in using the Gaming profile. Unless the video card is installed in a case with poor ventilation, or there is a desire to experiment with maximum overclocking.
In general, it is not surprising that the developers chose Silent as the default mode. The noise level is lower, and there is no loss of performance. Note that in both cases the maximum operating frequency of the GPU was increased to the same value – 2880 MHz.
Test results
Previous tests of the GeForce RTX 5070 Ti 16 GB were conducted earlier, so we already have a general idea of the capabilities of this line of video cards.
In the "synthetic" tests from the 3DMark package, the new product significantly outperforms its predecessor, the GeForce RTX 4070 Ti, and in terms of results, it approaches the performance of the GeForce RTX 4080 SUPER. Moreover, in the Steel Way and Port Royal stages with ray tracing, the MSI GeForce RTX 5070 Ti VANGUARD SOC even managed to surpass the older model of the previous generation.
What will the balance of power be in real games? Initially, we use 1440p mode with maximum quality settings, without scaling and RT technologies.
According to the results of testing, the GeForce RTX 5070 Ti usually outperforms the GeForce RTX 4070 Ti SUPER by quite a bit (+10%), and in some cases even outperforms the GeForce RTX 4080 SUPER. As for the comparison with the older model from the new NVIDIA line, the RTX 5080 usually shows ~15% better performance.
GeForce RTX 5070 Ti generally feels very confident in 4K mode. Considerable computing power and 16 GB of memory often allow you to use the maximum quality settings without "accelerators" in the form of DLSS. But the available examples also demonstrate the opposite situation, when without scaling you will still have to somewhat quench the thirst for the original image quality, agreeing to additional optimizations. As a reassuring argument - even owners of the more expensive GeForce RTX 5080 would have to do the same.
DLSS 4 + MFG
One of the important advantages of the new GeForce RTX 50 line is support for DLSS 4 intelligent scaling technology with the Multi Frame Generation (MFG) mechanism, which allows you to radically increase the number of frames/s in games if necessary. The new generation of DLSS with Transfromer models requires more computing power, while allowing you to get a higher quality picture, largely compensating for image "artifacts" when scaling.
In turn, the MFG mechanism allows you to get up to three additional frames. Because of this number of generated frames, there is a problem of uneven intervals between frames, which affects the "smoothness" of the picture. GPU video cards with Blackwell architecture provide hardware Flip Metering, which transfers the logic of changing frames to the display engine, allowing the GPU to more precisely control the display time. This is one of the reasons why MFG is supported only by RTX 50 series video cards.
According to NVIDIA, over 100 games and applications already support DLSS 4, and the implementation of the new generation of scaling is only gaining momentum. It seems that with the advent of AMD's FSR 4, which is also based on AI algorithms, as well as the wider implementation of AMD Fluid Motion Frames (AFMF), the confrontation will move from the number of fps to the balance between performance and the quality of the final image.
PhysX Physics
After the announcement of the first GeForce RTX 50 series graphics cards, it was somewhat of a surprise that models based on Blackwell architecture GPUs lost support for the PhysX physics engine in 32-bit applications. The mechanism is quite actively used in games to organize the interaction of certain elements, which allows for realistic effects.
Although this is not about the general rejection of PhysX support, but only its implementation in 32-bit games, the list of "affected" projects includes many hit releases from 10-15 years ago - Borderlands 2, Batman: Arkham City, Mirror's Edge, Assassin's Creed IV: Black Flag, Mafia II and others.
Without PhysX support, you will either have to abandon the corresponding effects or rely on their calculation using the central processor. The latter option may be quite acceptable in certain cases, but there are also opposite examples, when calculating PhysX on the CPU leads to a radical decrease in performance.
We conducted a small experiment using the game Mafia II: Definitive Edition with PhysX effects enabled. The platform with GeForce RTX 5070 Ti in such conditions in 1440p mode and maximum quality settings managed to provide only an average of 26 frames/s. The thing is that all the interaction of particles with rich PhysX effects in the test scenes was calculated using the central processor, not the GPU. The indicators are completely sad. What can be done here?
Having at our disposal a GeForce RTX 4070 Ti SUPER and an AMD X870E motherboard with a pair of PCI-E x16 slots, we decided to combine the platform, simultaneously using the resources of two video cards - the main GeForce RTX 5070 Ti and the auxiliary RTX 4070 Ti SUPER.
After installing the second video card, we make the necessary settings in the Control Panel, selecting the GeForce RTX 4070 Ti SUPER video card directly for PhysX calculations. After simple manipulations, we repeat the test in Mafia II: Definitive Edition.
As a result, we have an average of 168 frames/s with identical quality settings. The "hybrid" configuration really works. It is unlikely that anyone will use such a combination in practice, but we are talking about performance in quite old projects with specific requirements. However, the option with two video cards remains if you want to replay a cult 32-bit project from 10 years ago with all the visual effects implemented using PhysX. Of course, it is not necessary to use the GeForce RTX 4070 Ti SUPER for this, a much more modest NVIDIA model of previous generations will also be suitable, which will be used exclusively for PhysX calculations.
Overclocking
Additional overclocking of NVIDIA's new line of graphics cards, in addition to academic, has a completely practical interest. Despite the fact that the GPUs actually remained on the 5-nanometer process, certain optimizations allow for significantly higher operating frequencies. In addition, the new GDDR7 memory chips have considerable potential for additional acceleration.
MSI GeForce RTX 5070 Ti VANGUARD SOC allows you to increase the initial power limit by 16%, so in the end we are talking about ~350 W, instead of the initial 300 W. We expand the energy framework to the maximum and start experimenting.
As a result, the base frequency of the GPU was increased by 415 MHz - from 2295 MHz to 2710 MHz, as a result of which the average acceleration value increased from 2588 MHz to 3003 MHz. The memory chips worked stably at 2125 MHz (34,000 MHz effective) and an additional frequency increase here is limited by software. Overclocking GDDR7 by more than 21% allows you to get a bandwidth of 1088 GB/s instead of the standard 896 GB/s.
During overclocking experiments, we used Gaming mode, which involves a more aggressive cooling system algorithm. Under such conditions, the temperature of the graphics processor rose only to 60C, and the memory chips to 64C. At the same time, the fan speed increased to ~1500 rpm. Obviously, after overclocking, the power consumption of the video card increased, as well as heat generation by key elements. However, an effective cooler allows you to maintain a comfortable temperature regime without much effort. These are exactly the conditions in which the advantages of the increased dimensions of the radiator block and the general design of the cooler with an evaporation chamber and through-flow are felt.
GPU/memory frequency forcing makes perfect sense. For example, after overclocking in the 3DMark Time Spy Extreme test, the video card scored 15,176 points (Graphics Score) instead of 13,640 . So, we have an increase of 11% . In conditions where 10–15% is the difference between generations of video cards, an additional 10% looks like a very good bonus. Especially considering that the cooler easily copes with the additional load. At least it's nice to know that there is such an opportunity, even if overclocking the video card was not even considered initially.
Power consumption
The declared power consumption level for the GeForce RTX 5070 Ti 16 GB is 300 W. Despite the additional frequency tuning of the GPU, the MSI GeForce RTX 5070 Ti VANGUARD SOC actually maintained the set level.
During gaming load, the video card required up to 302 W. This figure is close to what the tuned GeForce RTX 4070 Ti SUPER requires (~300 W). While the GeForce RTX 4080 SUPER requires up to 320 W.
After additional overclocking with an increase in Power Limit, the actual power consumption of the MSI GeForce RTX 5070 Ti VANGUARD SOC in "heavy" games increased to 330–335 W. Considering the additional performance gain, even in this case, the energy efficiency of the video card remains at a good level.
In idle mode, the graphics adapter consumed about 13 W of energy. This is a typical situation for the top models of the RTX 50 line. And this is slightly more than in the previous generation models of the same class (8–10 W).
Although we have not yet had the opportunity to evaluate the power requirements of AMD's direct competitor - the Radeon RX 9070 XT 16 GB, according to numerous studies, accelerated versions of this series of video cards require about 350 W.
