New Article: Radeon RX 9070 XT vs GeForce RTX 5070 Ti

Blackwell GPU Family Specifications and Benchmarks GeForce RTX 5080 made it clear that all 50-series positions, with the exception of the GeForce RTX 5090, are little more than a soft upgrade of the corresponding models of the previous generation. The situation was aggravated by the shortage of new video cards and, as a result, sky-high prices.

For most buyers, the top-end devices that NVIDIA monopolizes are of only theoretical interest. But one step down, the technological stagnation of "green" graphics has created a situation where AMD could make up for the lag in such problematic aspects of the RDNA architecture as ray tracing and upscaling in one go. However, we have been hearing the prophecy that AMD will restore parity with its competitor (even without claims to the flagship sector) for a long time, and each time it sounds less and less convincing, so the announcement of the Radeon RX 9000 video cards did not attract much attention against the backdrop of NVIDIA's vigorous activity. But AMD has carried out extensive work on errors and, if not eliminated, then at least compensated for the shortcomings of past GPUs.

We present the Radeon RX 9070 XT review in an unusual comparison format for 3DNews, since the “red” new product and its main competitor, the GeForce RTX 5070 Ti, fell into our hands at the same time, and one device cannot be considered in isolation from the other, since in Russia they are sold for about the same money.

⇡# Navi 48 GPU and RDNA 4 architecture

Only the lazy have not mentioned that NVIDIA has become a monopolist in the high-performance GPU market and behaves accordingly. Less often, attention is paid to the fact that all companies designing large processors are forced to use the services of one contractor for their production - Taiwanese TSMC. The demand for TSMC's capacity was probably one of the reasons why AMD stopped developing large consumer GPUs, and in general, in the line of "red" graphics processors of the new generation, there is only one crystal - Navi 48.

Among the older chips, it most closely resembles the second-tier model, Navi 32, in its core specs, but contains almost as many transistors as Navi 31 (53,9 and 57,7 billion, respectively). Navi 48 is built using TSMC's N4P technology, which belongs to the same 5 nm photolithographic node that was mastered two years ago. However, Navi 48 takes up only 68% of the area of ​​Navi 31, thanks to AMD's abandonment of the chiplet design, since MCD chiplets, which contain the third-level cache and memory controllers, are manufactured using the coarser N6 process technology.

The architecture of the new graphics processor, RDNA 4, is structurally no different from previous iterations. Navi 48 contains 64 Compute Units, or 4 FP096-compatible shader ALUs, which is a fairly typical formula for a second-tier GPU - as is the 32-bit video memory bus. However, AMD also compares Navi 256 in performance estimates to the flagship Navi 48, which has 31 standard-precision real-number shader ALUs. The difference should be more than compensated for by multiple logic improvements, and to understand them, we recommend refreshing your memory of the RDNA 6 architecture, which we discussed in detail in the review Radeon RX 7900 XTX.

Compared to green architecture Blackwell RDNA 4 is a larger and deeper upgrade of previous developments. However, the description of all the innovations surprisingly folds into a short list due to the fact that NVIDIA traditionally pays increased attention to complex software or semi-software solutions that require detailed comments, while AMD's innovations, as a rule, begin and end in hardware.

Fundamental changes have occurred in the Compute Unit device — an indivisible computing block, which is an analogue of SM in NVIDIA and CPU cores. Thus, the instruction scheduler now supports the so-called divided barriers, which allow instructions to be issued for execution during the action period. Prefetching of instructions from the cache has also been optimized, and scalar ALUs, which are used for economical execution of conditional jump and branching operations, have inherited the ability to work with real-valued data from the RDNA 3.5 mobile architecture. At the same time, the volume of scalar registers has also been increased: from 10 to 16 KB.

One of the most important innovations of RDNA 4 is the doubled execution rate of WMMA matrix instructions, which are used in machine learning tasks and data processing using neural networks. In addition, the RDNA 4 instruction set contains commands for calculations on structurally sparse matrices and 8-bit floating-point data (FP8 and BF8).

However, the 1 FLOPs per clock means that Navi chips are only on par with the previous green architecture, Ampere. In addition, while NVIDIA and Intel chips use separate ALU arrays for matrix operations, in the RNDA 024 architecture this work still falls on shader SIMDs.

However, let's not forget that in real applications, RDNA 3-based GPUs never even come close to full load. The declared teraflops in general-purpose FP32 calculations are achievable only if the VOPD (Vector Operation Dual) format is used, which allows two independent vector instructions to be packed together. And this format itself imposes a lot of limitations, so half of the 32-bit ALUs inside the Compute Unit are most often idle. In a mixed load (such as a game with upscaling), matrix operations will, of course, take away some computing resources from shaders, but these losses will not be as great as it seems at first glance.

Frame scaling with a neural network is no longer a theoretical situation for Radeon accelerators. The new, fourth version of FSR finally relies on machine learning, including for frame generation, and the models used are a hybrid version of convolutional networks and transformers. Alas, unlike previous versions of the upscaler, which worked on almost any hardware, FSR 4 only supports RDNA 4 architecture GPUs (due to the expansion of the WMMA instruction set to FP8 data). A universal mode with simplified computing cores, like in Intel's XeSS, is not provided.

Another area of ​​the RDNA architecture reform was ray tracing. The speed of testing ray intersections with the BVH box and with a triangle has doubled per Ray Accelerator (and therefore one Compute Unit): from 4 and 1 per clock to 8 and 2, respectively. However, the RT block of Intel Battlemage chips works at a rate of 18 and 2, and in Blackwell graphics processors it finds four ray intersections with a triangle per clock and an unpublished number of ray intersections with the BVH box. Moreover, competitors perform two types of testing (against the BVH box and triangle) simultaneously, while AMD still does not. To some extent, parallelism is possible, because each Ray Accelerator now contains two independent "tester", but not at full speed.

Additionally, the Ray Accelerator has been given the ability to process rays out of order. Rays that do not require access to slow levels of the memory stack are processed first, and the test results are saved so that the shader program receives them in the expected order.

The ray tracing logic now uses a denser BVH8 structure instead of BVH4 (eight instead of four nodes for each branch of the volume hierarchy), which increases the speed of passage and simultaneously reduces the pressure on video memory. The size of the BVH in memory has also decreased due to the new primitive compression algorithm. Finally, AMD proposes to untie the orientation of the BVH boxes from the absolute coordinates of the scene. Rotating the boxes in accordance with the geometry of objects allows reducing the number of steps in the search.

It's a shame that despite all these changes, RDNA 4 still does BVH on shader ALUs rather than dedicated logic. According to Imagination Technologies' classification of hardware ray tracing tools, AMD's solutions are stuck at level 3, Intel's GPUs are at level 4, while NVIDIA's hardware is already moving from level XNUMX to level XNUMX thanks to SER, a dynamic instruction stream grouping feature designed to improve memory access coherence and, as a result, reduce latency.

There is no analog of SER in the RDNA 4 architecture either, but there is a function called Out of Order Memory, which helps solve the same problem in a different way. Previous versions of RDNA, working with several groups of instructions (wavefronts in AMD terminology), return data from memory strictly in the same order in which the requests were made. When the data requested by a shader is in a high-speed cache, the return can still be delayed if another shader made its request earlier, and its data is in comparatively slow VRAM. To be honest, this limitation seems completely meaningless in the context of multi-threaded computing, but it certainly appeared in order to simplify the organization of the GPU and for the time being (read: in rasterized games) it remained a reasonable compromise. RDNA 4 allows groups of instructions to return data regardless of the order of requests.

Out of Order Memory is just one of the changes to the memory stack in the RDNA 4 architecture. Previously, Compute Unit register allocation to shader programs was static, based on maximum consumption during the lifetime of the shader, which meant that a certain portion of the register file was effectively empty. RDNA 4 allows shaders to request and free registers dynamically. This allows the register file to be used more economically.

It is interesting that Shader Engine — a large GPU structure that combines the CU array, rasterizer, primitive block and rasterization operation blocks — has lost its own first-level cache in RDNA 4. But each segment of L2 cache contains twice as many banks. Thus, not only its volume has increased, but also its throughput, and in both directions — to Shader Engine and to Infinity Cache (in other words, L3). The volume of Infinity Cache itself remained the same as in Navi 32 — 64 MB.

Navi 48 has 16 PCI Express fifth-generation bus lines, but in gaming tasks, of course, it cannot fully occupy such a fast interface.

Finally, AMD says that the hardware video codec in the next-generation GPU delivers improved encoding quality at the popular low bitrate for streaming, but nothing more. That means there's still no support for YUV 4:2:2 color subsampling, which NVIDIA and Intel GPUs have.

In terms of image output, the Navi 48 display controller is compatible with the latest DisplayPort and HDMI interfaces: 2.1a and 2.1b, respectively.

⇡#Technical characteristics, prices

The Radeon RX 9070 XT is equipped with a fully functional Navi 48 crystal with 64 Compute Units, which means 4 standard-precision real-valued ALUs. It is not easy to find a formal predecessor of the new product among the “red” accelerators of the previous generation, because both the model grid and the GPU numbering principle have changed (Navi chips have never reached eight before). In terms of the formula for computing units and the characteristics of the memory stack, the Radeon RX 096 XT is similar to the Radeon RX 9070 XT, but AMD positions it as a replacement for the Radeon RX 7800 XTX.

The GPU's Game Clock and Boost Clock are significantly higher than the previous-gen flagship, at 2,4GHz and 2,97GHz, respectively. The Radeon RX 7900 XTX's theoretical peak clock speed is 26% higher, but AMD is looking to make up for that with improvements to its RDNA 4 architecture.

The Radeon RX 9070 XT features 16GB of GDDR6 memory with a bandwidth of 20Gbps per pin on a 256-bit bus. AMD was rumored to have considered increasing the VRAM capacity to 32GB, but ultimately deemed such a configuration excessive for consumer solutions. It’s a shame, because gaming appetites are steadily growing, and the amount of video memory was an important competitive advantage for many of AMD’s previous devices.

The Radeon RX 9070 XT has a reference power consumption of 304W and a suggested retail price of $599 ($100 more than the Radeon RX 7800 XT).

In turn, the GeForce RTX 5070 Ti is based on the same GB203 GPU used in the GeForce RTX 5080, but the younger model has lost 14 streaming multiprocessors, that is, 1 FP792-compatible shader ALUs, and the gaming clock frequency has been reduced. As a result, the GeForce RTX 32 is separated from the RTX 5080 Ti by 5070% in theoretical performance, and the GeForce RTX 27 Ti itself does not differ in this parameter from the RTX 5070 Ti SUPER. As in the case of the Radeon RX 4070 XT, the lack of raw computing power is intended to be compensated for by the new GPU architecture, and first of all, the function of generating multiple frames using DLSS (up to three generated frames in the interval between two "real" ones).

The GeForce RTX 5070 Ti is rated for 300W of power consumption and, with an MSRP of $749, is $50 cheaper than the GeForce RTX 4070 Ti SUPER.

The $150 difference in recommended currency value between the Radeon RX 9070 XT and GeForce RTX 5070 Ti is not noticeable at current Russian prices: we found the most affordable partner models in DNS for 99 and 999 rubles, respectively. These are the ones we will test in benchmarks. The Radeon RX 104 XT is represented by the GIGABYTE GAMING OC graphics card with a factory GPU overclocked to 999/9070 GHz, and the GeForce RTX 2,52 Ti is a Palit GamingPro accelerator with reference specifications.

⇡#GIGABYTE Radeon RX 9070 XT GAMING OC: design

The Radeon RX 9070 XT by GIGABYTE has rather modest dimensions for a video card with a power consumption of no less than (and, as tests will show, much more than) 300 W: 288 mm in length and three slots in width. The only LED backlight element is a short strip on the side. The nameplate with the manufacturer's logo is movable: if you move it to the LED strip, the logo starts to glow.

The cooling system is served by three fans with an impeller diameter of 87 mm. Air passes through a small section of the radiator thanks to a cutout in the backplate.

The graphics processor and video memory chips give off heat to the evaporation chamber, but the former directly, and the latter through a U-shaped metal spacer. Separate plates are attached to the radiator lamellas, which serve as heat sinks for the VRM components. There are only seven heat pipes, their diameter is 6 mm. Note that GIGABYTE preferred liquid thermal pads to regular ones. Perhaps this is the reason why the VRAM chips heat up under load (see the empirical section of the review).

The backplate is involved in heat dissipation from the printed circuit board using several thermal pads (here they are regular, not liquid).

⇡#GIGABYTE Radeon RX 9070 XT GAMING OC: PCB

Some Radeon RX 9070 XT variants draw power from a 12V-2×6 connector, but GIGABYTE has opted for three universal eight-pin connectors. In light of the graphics card’s high power consumption and the ongoing complaints about the new connector, this is a more reliable solution, albeit not as convenient or aesthetically pleasing. Of the 17 power phases, 14 are dedicated to the GPU, controlled by a Monolithic Power Systems MP2868A PWM controller, and three are dedicated to video memory. Both VRMs are equipped with MP87993 power stages with an estimated current rating of 90A.

VRAM is composed of SK hynix chips, their marking H56G42AS8DX-014 encodes a bandwidth of 20 Gbps.

In addition to the BIOS copy that is active by default, the video card has a backup “silent” firmware.

⇡#Palit GeForce RTX 5070 Ti GamingPro: Design

The Palit GeForce RTX 5070 Ti GamingPro measures 331,9 mm in length, but only has three expansion slots. Under the GamingPro logo and the adjacent segment of the front panel are the RGB LEDs. The LEDs can be synchronized with the motherboard via a standard ARGB connector, which is located next to the 12V-2×6 power input.

The side surfaces of the case are formed by a cast aluminum frame with ventilation slots on the long sides. The protective plate on the back of the PCB has a familiar grill for through-blowing the radiator. The fan impellers have a diameter of 94 mm.

At the base of the cooler is a vapor chamber large enough to cover the GPU die and the surrounding video memory chips, and the VRM components are provided with separate plate heat sinks. The radiator lamellas are strung on eight heat pipes with a diameter of 6 mm.

The metal backplate serves only a protective and decorative function, since there is not a single thermal pad between it and the PCB.

The video card comes with an adapter from two eight-pin power connectors to 12V-2×6 and a small fabric mouse pad.

⇡#Palit GeForce RTX 5070 Ti GamingPro PCB

Unlike a video card Palit GameRock, which we reviewed in our recent GeForce RTX 5080 review, the next-oldest model in the series is assembled on a spacious PCB, but its power supply system is simpler. The GPU has a 15-phase VRM controlled by a Monolithic Power Systems MP29816 PWM controller. The GDDR7 memory chips are powered by a three-phase power supply based on the MP2988 controller. Both regulators are equipped with MP87993 transistor assemblies with an estimated current rating of 90 A.

The marking of the Samsung K4VAF325ZC-SC28 video memory chips indicates a throughput of 28 Gbps.

Like the Radeon RX 9070 XT from GIGABYTE, the Palit accelerator has two BIOS options - "performance" and "quiet".

⇡#Test stand, testing methodology

In most games, the average and minimum (we specify the 1st percentile of the distribution) frame rates are derived from an array of individual frame rendering times or instantaneous frame rates obtained using a built-in benchmark. The exceptions are games that do not have a built-in benchmark and tests that use frame generation: in these cases, we use OCAT to capture interframe intervals.

The power of the graphics cards is recorded separately from the CPU and other PC components using the NVIDIA PCAT device. The load for the power and noise level tests is the Cyberpunk 2077 game at a resolution of 3840 × 2160 and maximum graphics quality settings (without ray tracing), as well as the FurMark stress test with the most aggressive settings (resolution 3840 × 2160, MSAA 8x). All parameters are measured after the graphics card has warmed up, when the GPU temperature and clock frequencies have stabilized.

⇡#Test participants

The following video cards took part in performance testing:

• AMD Radeon RX 9070 XT (1820/3060 MHz, 20 Gbps, 16 GB);
• NVIDIA GeForce RTX 5070 Ti (2295/2452 MHz, 28 Gbps, 16 GB);
• AMD Radeon RX 7900 XTX (1720/2499 MHz, 20 Gbps, 24 GB);
• NVIDIA GeForce RTX 4070 Ti SUPER (2340/2640 MHz, 21 Gbps, 12 GB);
• NVIDIA GeForce RTX 4070 Ti (2310/2610 MHz, 21 Gbps, 12 GB);
• NVIDIA GeForce RTX 4080 SUPER (2295/2580 MHz, 23 Gbps, 16 GB);
• NVIDIA GeForce RTX 4080 (2205/2505 MHz, 22,4 Gbps, 16 GB);
• NVIDIA GeForce RTX 5080 (2295/2617 MHz, 30 Gbps, 16 GB).

Note. The base and boost frequencies of the GPU are shown in brackets.

⇡#Clock speeds, power consumption, temperature, noise level and overclocking

The Navi 48 GPU loses the split clock domains into the frontend and CU array that AMD introduced in the previous generation of Navi chips. However, no matter which of the two clock rates of the outgoing line's flagship GPU (Navi 31) you take as a reference, Navi 48 has made a major step up to 2,9 GHz. The GeForce RTX 5070 Ti, on the other hand, is content with 2,8 GHz clock rate under gaming load.

Although the Radeon RX 9070 XT is not the flagship model of its generation (which it will not be this time), the power consumption of the new product in the GIGABYTE version reaches 374 W - almost like the Radeon RX 7900 XTX. This contrasts with the surprisingly low power of the "red" video card in idle mode - only 7 W. The GeForce RTX 5070 Ti from Palit consumes no more than 318 W - at the level of the 80 models of the 40th series.

The choice of firmware — “quiet” or “productive” — has a noticeable impact on the operation of the GIGABYTE Radeon RX 9070 XT GAMING OC cooling system. In the “productive” mode, the GPU temperature does not exceed 62 °C (and 84 at the hot spot), but the memory chips heat up to 94 °C. In the “quiet” mode, the temperature indicators change to 67, 90 and 98 °C, respectively.

In turn, the cooling system of the Palit GeForce RTX 5070 Ti GamingPro keeps the temperature of the graphics processor within 68 or 76 °C (again, depending on the active BIOS version), and the temperature of the VRAM chips stopped at a much more acceptable level of 70 or 78 °C. The GeForce 50 video card driver does not report the GPU hotspot temperature to monitoring utilities.

The noise level of the cooling system of both new products changes dramatically, obeying the BIOS switch. In the "quiet" mode, the Palit GeForce RTX 5070 Ti GamingPro turned out to be the quietest of the compared devices, but the GIGABYTE Radeon RX 9070 XT GAMING also distinguished itself with a moderate sound pressure adjusted for the significant power consumption. On the contrary, under the control of the "productive" BIOS, the GIGABYTE video card maintains an acceptable noise level, but the Palit GamingPro reaches 45 dBA - and this is without overclocking.

AMD's new GPUs are overclocked in the same way as NVIDIA chips: now, instead of the upper limit of the clock frequency, the user specifies the desired increase. The GPU supply voltage is regulated only downwards. Our video card sample remained stable when the target clock frequency was increased by 300 MHz, but due to the fact that the power reserve can only be increased by 10%, the actual core frequency increased by a symbolic 65 MHz. In turn, the video memory reached a bandwidth of 22 instead of the original 20 Gbps. The power consumption of the GIGABYTE Radeon RX 9070 XT GAMING in overclocking exceeded the 400 W mark, due to which the component temperature increased by a couple of degrees, and the noise level reached 41 dBA at a distance of 30 cm from the fans.

In contrast, the Palit GeForce RTX 5070 Ti, like the GeForce RTX 5080, overclocks superbly. Although the non-OC version of the card does not allow the factory TBP to be exceeded at all, we were able to add 375 MHz to the actual GPU clock in the gaming test and increase the GDDR7 memory bandwidth from 28 to 32 Gbps. Overclocking did not have a major impact on power consumption and component temperatures, and the noise level remained the same only because it was already quite high with the “performance” firmware.

⇡#Gaming tests (1920×1080)

At 1080p, both the GeForce RTX 5070 Ti and Radeon RX 9070 XT deliver frame rates in excess of 60 FPS in most of our test titles, and at least XNUMX FPS in Black Myth: Wukong.

Although some games prefer the AMD or NVIDIA architecture, from a practical point of view, the new products have equal performance: the GeForce RTX 5070 Ti is only 9070% slower than the Radeon RX 3 XT in FPS. Compared to the previous generation devices, the Radeon RX 9070 XT is 3% slower than the Radeon RX 7900 XTX. The GeForce RTX 5070 Ti is equivalent to the GeForce RTX 4080 and RTX 4080 SUPER, and the difference between the RTX 4070 Ti SUPER and the RTX 5070 Ti is 13% in frame rate.

⇡#Gaming tests (2560×1440)

In most benchmarks at 1440p, the new AMD and NVIDIA models achieve frame rates of at least 80 FPS, but the average frame rate in Black Myth: Wukong dropped to 51-52 FPS.

The Radeon RX 9070 XT and GeForce RTX 5070 Ti are again separated by a virtually insignificant 4% frame rate. The “red” new-generation graphics card is also only 4% behind the former flagship, the Radeon RX 7900 XTX, while the GeForce RTX 5070 Ti remains a replacement for the 80s of the previous generation. Compared to the GeForce RTX 4070 Ti SUPER, the new NVIDIA model has become 15% faster.

⇡#Gaming tests (3840×2160)

The Radeon RX 9070 XT and GeForce RTX 5070 Ti are generally suitable for gaming on a 4K screen at maximum graphics quality. Only in resource-intensive titles such as Black Myth: Wukong and Cyberpunk 2077 will the user have to put up with a frame rate significantly below 60 FPS or use upscaling.

High resolution hasn't changed the positions of the main competitors: Radeon RX 9070 XT still outperforms GeForce RTX 5070 Ti by 4% in average frame rate, but at the same time is 5% behind Radeon RX 7900 XTX. GeForce RTX 5070 Ti itself still replaces GeForce RTX 4080 and RTX 4080 SUPER, but the advantage over RTX 4070 Ti SUPER has grown to 18% FPS.

⇡#Ray Tracing Gaming Tests

Ray tracing games are the most important test that AMD's new accelerators will have to pass after years of lagging behind NVIDIA's solutions in this area. Fortunately, there are big changes for the better. Although in rasterized games the Radeon RX 9070 XT is an analogue of the Radeon RX 7900 XTX and on average even lags behind the former flagship, in ray tracing the Radeon RX 9070 XT is already ahead with a gap of 20-21% FPS.

True, there are games where the difference is small (Metro Exodus and Returnal), and in Indiana Jones and the Great Circle the frame rate on the Radeon RX 9070 XT dropped to 1 FPS even with FSR, which is apparently due to a lack of VRAM. However, this game only opened path tracing to owners of "red" GPUs in the latest add-on. The necessary driver optimizations are probably missing, so we did not take into account the results of Indiana Jones and the Great Circle when calculating the percentage ratios between the test participants. By the way, among all the test participants, only the Radeon RX 7900 XTX has the necessary amount of video memory to run Indiana Jones and the Great Circle on a 4K screen with maximum graphics quality. Other devices do not have this even with the most aggressive frame scaling settings.

In absolute frame rate terms, the Radeon RX 9070 XT has the performance needed for hybrid rendering titles at resolutions up to 1440p and, at a stretch, 4K.

The same goes for the GeForce RTX 5070 Ti, and yet the average score for NVIDIA's device is 25-32% ahead, mostly in fully traced games. In its own camp, the GeForce RTX 5070 Ti confirms its status as the successor to the GeForce RTX 4080 and RTX 4080 SUPER: the latter's advantage lies in the range of 1 to 5% of frame rate. Compared to the GeForce RTX 4070 Ti SUPER, performance has increased by 13 to 19%.

⇡#Game tests with ray tracing and frame scaling

Thanks to Balanced upscaling, the new generation of AMD and NVIDIA devices guarantee comfortable frame rates in “hybrid” games at resolutions up to 4K. However, due to the 20-26% advantage, the GeForce RTX 5070 Ti copes much better with fully traced titles.

Compared to the Radeon RX 7900 XTX, the upgraded AMD chip architecture provided the Radeon RX 9070 XT with an 18-20% FPS performance boost. The GeForce RTX 5070 Ti, in turn, lags behind the GeForce RTX 4080 and RTX 4080 SUPER by up to 4% FPS, but outperforms the GeForce RTX 4070 Ti SUPER by 10-14%.

⇡#Gaming tests in overclocking

As you might expect from the meager GPU clock speed increase, overclocking the Radeon RX 9070 TX (at least in the GIGABYTE GAMING OC version) is a near-futile exercise, only allowing for a 4% frame rate gain. The Palit GeForce RTX 5070 Ti GamingPro, on the other hand, is 9% faster after overclocking.

⇡#Tests in production applications

In the Blender benchmark, the new AMD graphics card was able to extract a significantly greater advantage from hardware acceleration of ray tracing than the Radeon RX 7900 XT. However, the old flagship completed rendering faster due to its advantage in raw performance. As a result, there can be no talk of equal competition between the Radeon RX 9070 XT and the GeForce RTX 5070 Ti. The GeForce RTX 5070 Ti itself took an intermediate position between the RTX 4070 Ti SUPER and RTX 4080.

In transcoding tasks using Premiere Pro, the "red" accelerator of the new generation is on par with the Radeon RX 7900 XTX, but lags behind the "green" competitors in the overall assessment due to the relatively low speed of working with RAW files. GeForce RTX 5070 Ti, like the RTX 5080, uses an updated H.264 and HEVC decoder, so it confidently outperforms the 40-series video cards.

The Radeon RX 9070 XT takes the last place in the GPU effects rendering speed chart, while the GeForce RTX 5070 Ti follows the RTX 5080 and two versions of the RTX 4080 in this test.

DaVinci Resolve transcoding tests put the Radeon RX 9070 XT on par with the Radeon RX 7900 XTX, just behind the GeForce RTX 5080. The GeForce RTX 5070 Ti outperforms all older NVIDIA solutions and follows the Radeon RX 9070 XT.

In the GPU effects rendering benchmark, the Radeon RX 9070 XT again falls to the bottom of the chart. The GeForce RTX 5070 Ti, on the other hand, is ahead of even the Radeon RX 7900 XTX and is second only to the GeForce RTX 5080.

The GeForce RTX 5070 Ti is poorly suited for a number of CAD applications due to the lack of necessary driver optimizations. Here, the Radeon RX 9070 XT takes revenge.

⇡#Video encoding/decoding

The hardware decoder on the Navi 48 chip received a non-trivial increase in performance when working with AV1, but the “green” GPUs still cope with other video formats noticeably faster.

In turn, the video encoder has also gained speed and can now compete with Intel's QuickSync, although it lags behind the "green" NVENC.

⇡#Performance per Watt

Since our Radeon RX 9070 XT and Radeon RX 7900 XTX samples have almost the same power reserves, the difference in power efficiency between them is entirely determined by gaming performance. The old flagship leads with a 5% FPS advantage in specific frame rate in rasterization, but in traced games the Radeon RX 9070 XT made a 20% jump.

In turn, the GeForce RTX 5070 Ti here also turned out to be an analogue of the GeForce RTX 4080 SUPER, but this video card has a lower TBP compared to the Radeon RX 9070 XT and outperforms the "red" competitor in ray tracing. As a result, the energy efficiency of the GeForce RTX 5070 Ti is higher by 13 or 55%, depending on the rendering method.

⇡#Summary of gaming benchmark results without ray tracing

⇡#Ray Tracing Gaming Benchmarks Summary

⇡#Summary of gaming benchmarks with ray tracing and frame scaling

⇡#Conclusions

Of the two competing devices, the Radeon RX 9070 XT and the GeForce RTX 5070 Ti, the latter is the easiest to describe. The graphics card is a variant of the GeForce RTX 5080, cut down just enough to make it an analogue of the GeForce RTX 4080 or RTX 4080 SUPER, only with a lower MSRP. This in itself would be forgivable if it weren’t for the fact that the GeForce RTX 5070 Ti outperforms the RTX 4070 Ti SUPER by 19% FPS at best under the most favorable conditions, and officially costs only $50 less, not to mention the actual prices.

The Radeon RX 9070 XT takes the place of the Radeon RX 7900 XTX in rasterized games, despite a large gap in raw computing power, and most importantly, due to the improved architecture, it has pulled ahead by 21% FPS in ray tracing. It is nice to see that at least AMD's mid-range GPU has caught up and surpassed the old top-tier GPU. But alas, even this increase was not enough to put an end to the multi-year gap with NVIDIA. The Radeon RX 9070 XT has a symbolic victory over the GeForce RTX 5070 Ti in rasterization, but in games with ray tracing, the average advantage of the GeForce RTX 5070 Ti reaches 32%, and in fully traced games - even more. The recommended price of the Radeon RX 9070 XT is proportionally lower than that of the RTX 5070 Ti, which balances out the weaknesses of the RDNA 4 logic. However, in Russia, the minimum prices of the competitors are almost equal, so despite NVIDIA's stagnation and AMD's progress, the choice between the Radeon RX 9070 XT and GeForce RTX 5070 Ti is simple - at least for now.

Among the gaming functions of the new GPUs, upscaling takes first place — DLSS or FSR version 4. NVIDIA chips received the ability to generate multiple frames and transformable models, and AMD, in principle, began to use neural networks for this purpose. But the FSR XNUMX technology, like DLSS, is tied to the "native" hardware, so the frame scaling tools available to owners of "green" and "red" accelerators are divided, and it is no longer possible to say that GeForce video cards support both DLSS and FSR, and Radeon - only FSR. We will consider the topic of upscaling quality and related latency separately in the upcoming study.

The Radeon RX 9070 XT, like the GeForce RTX 5070 Ti, has 16 GB of video memory. This means that within the current generation, only the GeForce RTX 5090 offers more, while other models risk getting into a situation where the GPU could handle the maximum graphics quality in a game (even with the help of upscaling), but is limited by the amount of VRAM. So far, there is only one such game - Indiana Jones and the Great Circle, but this is just the first bell.

The Radeon RX 9070 XT is a pure gaming product, poorly suited to most of the workloads we tested (excluding CAD). The GeForce RTX 5070 Ti, on the other hand, performed well in video editing tasks thanks to its updated hardware codec, but again is unlikely to be of interest to professionals due to the lack of local memory.

Finally, let's say a few words about the video cards that represent the Radeon RX 9070 XT and GeForce RTX 5070 Ti in the review - and now these are the most affordable versions of both models on the Russian market. The "red" GIGABYTE accelerator and the "green" Palit have a higher power consumption than prescribed by the reference specifications (in the "Radeon" it has exceeded 360 W) and work quite quietly if the corresponding BIOS is active. At the same time, the Palit GeForce RTX 5070 Ti GamingPro overclocks much better, although it does not even allow you to increase the TBP, but the GIGABYTE Radeon RX 9070 XT GAMING OC unpleasantly surprised with a VRAM temperature above 90 ° C and liquid thermal pads, which make it difficult to service the video card.

Source: 3dnews.ru