INNO3D GeForce RTX 5070 TWIN X2 graphics card: an organic transition to 1440p

GeForce xx70 series video cards traditionally arouse increased interest among rational users who are probably not satisfied with the capabilities of the basic models, but they are not ready to pay full price for top-end graphics adapters. So, the appearance of another "seventy" from NVIDIA is a good reason to study its capabilities in detail in search of the best solution for their tasks. We finally got a GeForce RTX 5070 for review. This time we are talking about the INNO3D GeForce RTX 5070 TWIN X2 - an original two-fan and two-slot model (!), which adds intrigue and practical interest for owners of compact systems (SFF). Do we really have the best option for 1440p games?

NVIDIA GeForce RTX 5070

Like all RTX 50 series graphics cards, the GeForce RTX 5070 uses a processor with the Blackwell graphics architecture. In this case, it is a GB205 chip in the GB205-300 modification with 6144 active compute units. The GPU is manufactured by TSMC using the 5 nm process standard, has a die area of 263 mm² and contains 31.1 billion transistors.

The available processor configuration also includes 192 texture units and 80 rasterization units. The GPU is also equipped with 192 tensor cores and 48 RT cores for ray tracing calculations. The base operating frequency of the chip is set at 2325 MHz, while the average dynamic acceleration value is 2512 MHz.

If we quantitatively compare the performance of the available functional blocks with those of the GeForce RTX 4070, at first glance, the changes are relatively small. The number of processors has increased by only 4.3% - from 5888 to 6144. The same increase applies to texture blocks (192 vs. 184), as well as tensor and RT cores. Although in the latter case, you should take into account the different generation of these blocks. Is this really the whole difference? Fortunately, no. The first thing you notice when comparing the specifications is that the number of rasterizers (ROPs) has increased by 25% - from 64 to 80. A nice option that will be especially useful in higher resolution modes.

Another interesting nuance is that the GB205 processor in the GeForce RTX 5070 version has 48 MB of L2 cache, while the AD104-250 from the RTX 4070 only had 36 MB of L2 enabled. Access to the available 48 MB was only available to the upgraded GeForce RTX 4070 SUPER.

The GeForce RTX 5070 inherited a 192-bit bus and a total memory capacity of 12 GB from its predecessor. However, the new model is equipped with GDDR7 chips with an effective frequency of 28,000 MHz, while the RTX 4070 uses GDDR6X chips (effective 21,000 MHz). So the RTX 5070 ended up with a higher memory bandwidth - 672 GB/s versus 504 GB/s in the previous generation model. The actual PSP increased by 33% with the same bus width.

As for power consumption, with the same GPU manufacturing process, you should not expect significant improvements. Increasing the number of functional blocks and increasing the operating frequencies of the graphics processor require additional energy costs. The declared total power consumption level of the GeForce RTX 5070 is 250 W, while the TGP for the GeForce RTX 4070 was 200 W.

The recommended price of the GeForce RTX 5070 12 GB for the US market is $549. Recall that the GeForce RTX 4070 was offered from $599 at the start of sales in April 2023, and later, after the launch of the GeForce RTX 4070 SUPER with an identical price tag, the recommended price of the base RTX 4070 was officially reduced to $549. No matter how nice it would be to operate with MSRP for the US market without taking into account taxes, retail price tags in Ukraine are obviously quite far from similar values. However, we will return to this topic later.

NVIDIA has introduced its own reference model, the GeForce RTX 5070 Founders Edition. The circulation of this version is quite limited, so developers rely more on original modifications from partners. This time we will get to know one of these versions in more detail.

INNO3D GeForce RTX 5070 TWIN X2

We are dealing with an interesting version of the GeForce RTX 5070 from INNO3D. The manufacturer offers several modifications of this series at once.

The special feature of the INNO3D GeForce RTX 5070 TWIN X2 model is its dual-slot design and compliance with the NVIDIA SFF-ready specification, which allows it to be used in fairly compact systems.

The GPU frequency formula corresponds to the base specification - 2325/2512 MHz. The GDDR7 memory chips also operate at the recommended frequency of 1750 MHz (28,000 MHz effective).

In general, the manufacturer also offers a slightly accelerated modification of the INNO3D GeForce RTX 5070 TWIN X2 OC in an identical design, which provides +30 MHz to the average GPU overclocking value - 2325/2542 MHz.

Package contents

The video card is offered in a fairly compact cardboard box that has a bright design with a stylized photo of the adapter on the front panel.

The INNO3D GeForce RTX 5070 TWIN X2, in addition to basic paper instructions, also comes with an adapter from two 8-pin PCI-E connectors to one 12V-2x6 (12VHPWR) for connecting additional power. A similar accessory will be needed if a PSU without support for the corresponding power line is used.

Design and layout

INNO3D GeForce RTX 5070 TWIN X2 has a rather laconic exterior design. It is an almost regular-shaped bar with dimensions of 250x116x41 mm and a weight of 890 g.

The embodiment of brutal simplicity. The upper protective casing has a combined design with an asymmetrical segment in the central part. Externally, the fragment is stylized as a metal surface, but it is still a combination of plastic parts. The screw fastening of the elements provides a "steampunk on minimal" styling, and a certain dynamism is added by the light ribbing of a certain area of the upper panel. The combination of the techniques used in the design made it possible to get rid of the boring design.

The overall dimensions of the INNO3D GeForce RTX 5070 TWIN X2 allow it to be used in compact systems. The model meets the recommendations of "SFF-Ready Enthusiast GeForce Cards" from NVIDIA and is included in the list of compatible video cards that allow you to implement the corresponding concept.

Despite the fact that even 2.5-slot models formally meet the specifications, in certain circumstances "pure" 2-slot video cards may have advantages.

The INNO3D GeForce RTX 5070 TWIN X2 uses a dual-fan cooler. The radiator array consists of three cassettes of aluminum plates, which connect six composite heat pipes with a nickel-plated coating.

The heat sink platform is based on a combined solution with a copper core in the GPU area and an aluminum profile for contact with memory chips and power assemblies.

The manufacturer's website contains an unexpectedly detailed description of the parameters of the radiator block, which states that the total length of the heat pipes is 1584 mm, and the total surface area of the entire cooling structure is 398,833 mm2.

The complex design "geography" of the radiator plate profiles shows that even in such limited conditions of a two-slot model, the developers tried to use the available free space under the protective casing of the video card as efficiently as possible.

A pair of 88mm axial fans are used to blow air through the radiator. The blades have a slightly longer length, a specific profile curve, and a connecting rim on the outer radius. This configuration usually allows for better airflow.

On the back side, the video card is covered with a protective metal plate. Since the length of the INNO3D GeForce RTX 5070 TWIN X2 printed circuit board is only 145 mm, while the total length is 250 mm, there was almost no doubt that the developers would offer a design with through-flow of the radiator.

The expectations were met. In the corresponding area of the protective plate, perforation is provided for the passage of air masses pumped by the fan. Such a cooler design has recently become widespread and really allows to increase the overall efficiency of the cooler.

The presence of a bracket in the GPU location area deserves special attention. The additional fastening element allows you to balance the pressing force of the radiator array against the graphics processor crystal.

To connect additional power, there is one 12V-2x6 (12VHPWR) connector on the top edge of the video card. The presence of an adapter with two 8-pin PCI-E connectors in the kit indicates that the adapter has enough power for a 300 W channel. This is true, considering the declared TGP (Total Graphics Power) for the GeForce RTX 5070 is 250 W.

As for the power supply capacity, the developers' recommendations coincide with those from NVIDIA engineers - a 650+ W PSU is required to operate the platform with the GeForce RTX 5070.

INNO3D GeForce RTX 5070 TWIN X2 did not receive additional RGB lighting. The layout of the adapter is quite dense, so there are no large free areas on the panels that require decorative filling. The slogan "Brutal by Nature" on the back panel does not seem to imply any petty annoying flickering. So lovers of bright illumination will have to rely on the capabilities of other components in this matter, or take the initiative into their own hands and decorate the internal space of the case with RGB/ARGB strips.

The interface panel is completely standard – three DisplayPort 2.1b and one full-size HDMI 2.1b. Approximately two-thirds of the steel cover is perforated with through holes to exhaust heated air outside the system unit.

Test bench configuration

• Motherboard: MSI MAG Z890 TOMAHAWK WiFi
• Processor: Intel Core Ultra 7 265K (8+12; 3.9/5.5 GHz + 3.3/4.6 GHz)
• Cooling: MSI MAG CORELIQUID A15 240
• Memory: Kingston FURY Renegade DDR5-7200 2×16GB (KF572C38RWK2-32)
• Drive: Kingston KC3000 1 GB (SKC3000S/1024G)

In operation

The video card supports hybrid operation mode, so at minimum load the cooling system fans stop.

Under such conditions, the GPU temperature was maintained at 42–43C, the memory chips heated up to 54C. The GPU supply voltage was 0.8 V, and the idle power consumption was 11.8 W.

Under prolonged gaming load, the chip temperature at its peak rose to 73.6C, the memory chips to 64C. The fan speed increased to ~2300 rpm. Subjectively, the noise level was average. This is probably not an ideal option for lovers of absolute silence, but the question that arises during testing, which is noisier - the SRO with fans at 1500 rpm or the video card cooler, is already a certain reassuring argument.

There are no additional silent modes for the INNO3D GeForce RTX 5070 TWIN X2. If you wish, you can try to manually adjust the cooler's operation algorithm. However, it is worth considering that 250 W is 250 W that must be diverted and dissipated. Therefore, given the obtained indicators, the range for additional settings is quite limited.

Productivity

For the initial performance assessment of the GeForce RTX 5070 12 GB, we used the components of the 3DMark test suite. Despite the "synthetic" nature of the component levels, their final indicators allow us to draw the first evaluation conclusions and well complement the testing stage in real games.

For comparison, the results of the junior model of the new line – the GeForce RTX 5060 Ti 16 GB, as well as previously obtained indicators for the GeForce RTX 4060 Ti 8 GB, GeForce RTX 5060 8 GB and GeForce RTX 4060 8 GB were used.

For clarity, we have also added to the 3DMark charts the averaged performance of the predecessor GeForce RTX 4070 12 GB, a more productive version with a GPU with Ada Lovelace architecture, as well as the older model of the current line - GeForce RTX 5070 Ti 16 GB.

The first stages of Time Spy show the GeForce RTX 5070 leading its predecessor by ~25%. In both cases, the performance of the GB250 model came very close to that of the GeForce RTX 4070 Ti.

In the Steel Nomad and Speed Way stages, the GeForce RTX 5070 model even slightly increased its advantage over the RTX 4070 and nominally even surpassed the average performance of the GeForce RTX 4070 Ti. At the same time, the video card also demonstrates a very significant advantage (~40%) over the younger GeForce RTX 5060 Ti 16 GB.

In the Port Royal subtest with active ray tracing, the INNO3D GeForce RTX 5070 TWIN X2 also had a 24% advantage over the RTX 4070 and about 37% over the RTX 5060 Ti 16 GB.

It should be noted that in all cases the GeForce RTX 5070, in turn, is significantly inferior to the older model of the new line - the GeForce RTX 5070 Ti 16 GB. The performance of these video cards differs by about a third, which is indirect evidence that in the future we will see the conditional RTX 5070 SUPER, which will fill this "gap".

In computational tasks, the results we obtained during previous reviews of the respective graphics cards remained. This is a clear example of how much more productive the GeForce RTX 5070 12 GB is than the younger GeForce RTX 5600 Ti 16 GB model, not to mention the even slower graphics cards of the previous generation.

Gaming in 1080p mode

The GeForce RTX 5070 12GB series graphics card is more of a 1440p gaming solution, but 1920 × 1080 screens are still the most popular among users, so at least in express mode we will also evaluate the capabilities of the graphics card for Full HD entertainment. During gaming tests, settings with maximum graphics quality were used without ray tracing and without the use of scaling technologies. Pure raster as it is.

The top three "light" projects, which even Cyberpunk 2077 already falls into with the specified settings, demonstrate that there is no limit to perfection. Despite the expectation that video cards more powerful than the GeForce RTX 5060 Ti no longer make much sense for Full HD games, we see how the RTX 5070 additionally improves the performance of the younger model by 20–25%.

Judging by the performance in relatively new resource-intensive games, a logical question arises - is the RTX 5070 really excessive for Full HD? An average of 60-70 frames/s with maximum settings is already very close to the limit of comfortable perception of the gameplay.

However, the video card still managed to pull out an average of 60 fps in Black Myth: Wukong at maximum settings, which is already a decent level. The younger RTX 5060 Ti 16 GB with identical settings managed to get only 47 fps. This is a case where additional memory capacity does not give any advantages, but a more powerful GPU helps a lot.

Gaming in 1440p mode

The 1440p mode for the GeForce RTX 5070 can be considered a profile mode. The rapid growth in the popularity of screens with a resolution of 2560x1440/1600 is an indicator that an increasing number of players have managed to appreciate the advantage of a "gaming window" with such parameters, so over time it is 1440p that may become dominant in the entertainment niche. The more frequent use of the mode also indicates a gradual increase in the performance of video cards, which are primarily subject to additional load when switching from Full HD.

In moderately resource-intensive projects, even in the mode with maximum quality settings on the GeForce RTX 5070, we get an average of over 100 fps, justifying the use of monitors with an increased refresh rate. The advantage over the RTX 5060 16 GB of 30–35% once again proves that these are models of different classes.

Avoved, Alan Wake 2 and STALKER 2: Heart of Chernobyl somewhat restrain excessive optimism. Despite a significant advantage over the younger model, even the GeForce RTX 5070's maximum quality settings in these projects look excessive. The average 47 fps is a sign that the game needs additional settings.

Black Myth: Wukong and The Elder Scrolls IV: Oblivion Remastered would also benefit from some image optimizations to improve gaming comfort. Undoubtedly, the GeForce RTX 5070 is significantly better suited for such modes than the GeForce RTX 5060. The presence of a larger memory capacity in such conditions does not yet provide tangible advantages. 12 GB is mostly enough for games at 1440p. At least, when ray tracing and frame generation are not activated, which usually requires an increase in memory capacity.

Let's also check the situation with the "rays". Still, owners of expensive video cards of the GeForce RTX 5070 level will probably want to use all the advantages of their new acquisition.

High-quality ray tracing is a very resource-intensive task. Even in Dying Light 2, activating RT (HQ) more than halved performance. While the RT Ultra mode in Alan Wake 2 dropped fps by as much as three times. In such cases, you have to rely on DLSS scaling and frame generation. The methods are really very effective, but in the pursuit of maximum fps it is important to maintain a balance of settings.

DLSS 4 support with multiple frame generation allows you to experiment with RT Overdrive mode in Cyberpunk 2077 and see the benefits of Path Tracing in DOOM: The Dark Ages without any special reservations. Owners of the GeForce RTX 5060 Ti 16 GB also have these modes available, but in all cases the performance of the younger model is about a third lower.

4K gaming

We also checked the situation with the transition to 4K mode. It is probably initially clear that with 12 GB of memory, the GeForce RTX 5070 graphics card is not the best option for such conditions, at least when it comes to maximum image quality settings.

Even regardless of memory capacity, an additional increase in resolution significantly increases the requirements for the graphics card's processing power. An average of 20–40 frames/s is not a level that will satisfy a gamer.

Video cards with 8 GB of memory in such conditions may not be suitable for running games in 4K at all. In our list of test projects, we did not encounter a situation where the presence of 16 GB in the RTX 5060 Ti would be a "lifeline" and help to gain an advantage over the GeForce RTX 5070 12 GB. However, when it comes to 4K at maximum settings, such cases are not excluded.

Looking at the results, it seems that ray tracing is out of the question. But, again, DLSS 4 and FG/MFG can be very helpful, especially in a balanced combination.

Even in a resource-intensive game like Alan Wake 2, you can get decent performance with RT Ultra enabled. Well, the 10x fps increase in Cyberpunk 2077 still feels like a fantastic transformation that you can witness with your own eyes.

Overclocking

Despite the fact that the INNO3D GeForce RTX 5070 TWIN X2 uses the recommended frequency formula for GPU/memory, the video card probably has additional hidden reserves, which we tried to explore during practical experiments.

We would like to point out right away that the video card does not allow you to increase the power limit (Power Limit). The dual-slot configuration of the GeForce RTX 5070 with a compact cooler still imposes certain limitations. However, does this significantly affect the frequency potential?

Despite the lack of changes in the power tolerance, we managed to significantly boost the video card. The graphics processor worked stably even after increasing the base frequency by +410 MHz - from 2325 MHz to 2735 MHz. At the same time, the GDDR7 memory chips (Samsung) "stuck" with NVIDIA's software limitation, which allows them to be increased to 2125 MHz (34,000 MHz effective). Compared to the standard value of 1750 MHz (28,000 MHz effective), we have an increase of +21.4%. The total memory bandwidth has also increased accordingly - from 672 GB/s to 816 GB/s.

After additional overclocking, the temperature regime of the video card changed somewhat. The graphics processor warmed up to 77C, the memory chips - to 66C. At the same time, the fan speed increased to ~2500 rpm. The GPU-Z utility records maximum bursts of the GPU operating frequency up to 3255 MHz. Considering the relative frequency increase, the temperatures increased very moderately, although we will not ignore the increase in fan speed.

Are the extra megahertz just "corn"? What about the real performance gain? Everything is pretty decent here. The 3DMark Time Spy Extreme test improves the Graphics Scrore score by almost 11% - from 10,467 points to 11,585 points.

In real games – Cyberpunk 2027 and Black Myth: Wukong – average fps in 1440p mode also increased by 7–11%.

Power consumption

The claimed power consumption level of the GeForce RTX 5070 is 250 W. During testing, we actually obtained the figures from the specification.

During gaming, the maximum level of power consumption was observed in Cyberpunk 2077. During peak loads on the INNO3D GeForce RTX 5070 TWIN X2, 252 W was obtained.

While testing the platform in various projects, we also recorded the overall level of PC consumption. Depending on the game, the figure fluctuated in the range of 340–430 W. Let's recall that we are talking about a system with a 20-core (8+12) Core Ultra 7 265K processor. So, a 650 W PSU for a configuration with a GeForce RTX 5070 will probably be more than enough.