The uniqueness of the new 3DMark is that it makes it possible to directly compare the performance of dissimilar systems. Futuremark assures that this approach is correct, the company managed to develop a methodology that allows you to reduce single system coordinates results obtained on different devices.
Since yesterday (04.02.2013) the application for Windows has been available to users, versions for Windows RT, Android and iOS will be presented soon.
As for 3DMark for OS from Microsoft, the application has certain requirements to the system configuration.
Please note that to run 3DMark you will need Windows Vista or more late version(Windows 7/8). In this case, XP is no longer supported. Otherwise minimum requirements pretty ordinary: a dual-core processor with a frequency of 1.8 GHz, 2 GB of memory and 3 GB free space on the drive.
ice storm
Ice Storm is designed to evaluate the capabilities of mobile products, as well as systems entry level. It will be common for all types of devices. For this test, the DirectX 11 engine is used, however, the API capabilities are limited to DirectX 9.0 functions, therefore this stage can be used to test not the most modern systems. Android and iOS will use OpenGL ES 2.0.
Space battles have always been a plot that deserves the closest attention. Representatives of two irreconcilable factions demonstrate the wonders of piloting and tactical maneuvers. Ice Storm is rendered at a single resolution of 1280x720 (720p) while being displayed specific device with a native aspect ratio for its display and the number of dots. This method (off-screen rendering) allows you to more accurately compare the results in the future.
The developer emphasizes that the performance of Ice Storm and 3DMark 06, which was also developed using the DirectX 9.0 API, cannot be directly compared.
Cloud Gate
To assess the capabilities of mid-range laptops and PCs, the developer suggests focusing on the readings of the Cloud Gate stage. More intense and resource-intensive scenes that use DirectX 10 resources for processing.
The space theme formed the basis of the second test. It turns out that even such banal things as hyperjump gates can be done very effectively. The developers of EVE Online now have a clear idea of what to strive for when preparing the next expansion.
fire strike
To run the Fire Strike test, you need a graphics adapter with full support for DirectX 11 capabilities. This stage is heavily flavored with various post-processing effects that require significant GPU computing resources, and therefore it is expected to be used to evaluate the capabilities of charged gaming platforms.
It takes about 15 minutes to complete the full test cycle, including demo scenes.
Basic free version 3DMark Basic Edition allows you to evaluate performance in all tests, however, they will be run with default settings, and you will need a network connection to view the results. Capabilities fine tuning appear in version 3DMark Advanced Edition, an upgrade to which will cost $24.99. In addition, the extended version has a whole list of features that make it easier to use the test package. For example, you can get detailed results in the 3DMark application itself and save them locally.
A function with interactive graphs that show CPU / GPU temperatures, the number of frames / s at any time during which the test took place, as well as processor power consumption (we will check with the developers on the method of calculating it) can also be useful.
Of course, the advantage of the Advanced Edition is also the ability to selectively select a specific stage and skip demos to reduce the time it takes to pass tests. The advanced version also allows access to Fire Strike's Extreme mode and the ability to run certain steps in a loop to test system stability.
Version 3DMark Professional Edition worth $995 provides a license to use the application for commercial purposes, allows you to run from command line, and provides more flexibility in managing the results (saving in XML). In addition, this edition provides access to the Image Quality Tool, which allows you to evaluate and compare the rendering quality of specific frames on different adapters.
For owners previous version 3DMark 11 Advanced Edition Test Suite is 25% off. The same opportunity to save when buying within one week after the announcement of 3DMark will be available to users Steam service. What is curious, due to an oversight or with the intention to attract more users, but at the time of preparation of the material, the cost enhanced version on Steam was $11.24, i.e. The 25% discount was made from $15, not the planned $25. A curious feature of the Steam version is the presence of a number of funny "achievements".
Development partners new version 3DMark became MSI and Galaxy, so some products of these manufacturers will be equipped with keys for free Advanced activation edition.
Test results
We did some research into 3DMark's capabilities using the test platform and graphics cards available in the lab at the time of the application's announcement. The current list of participants is not indicative, in the future the list will be expanded.
Core [email protected]/MSI Z77A-GD65/2x4 GB DDR3-2133, Windows 7 (x64), NVIDIA ForceWare 310.90 drivers, AMD Catalyst 13.1.
Almost 1500 frames / s in Ice Storm provide an indescribable smoothness of the picture, but, alas, the human eye cannot appreciate all the delights of such a frame rate. The 1.5-fold difference in system performance between Radeon HD 7770 and HD 7970 is an obvious sign that this stage is of little use for evaluating the capabilities of desktop systems.
Cloud Gate, although it generates a noticeably higher load, it is obvious that this stage does not allow you to appreciate the benefits of powerful components. Although the spread of results is higher than in the previous case, more than 200 fps (Radeon HD 7970) in the test scenes indicate that this subtest will be relevant for comparing systems with integrated graphics and entry-level video cards.
Fire Strike is a completely different matter, where a platform with an overclocked Corei7-3770K processor and a Radeon HD 7970 video card was able to deliver 25-30 frames / s to the mountain, and for acceptable performance in Extreme mode, a dual-chip GeForce GTX 690, or better a couple of these video cards.
Note that in addition to a serious load on the graphics adapter, during the passage of subtests with calculations of the physical interaction of particles, serious resources are also required from CPU. As an engine for calculating physics, the open Bullet Physics and a multithreaded algorithm are used that generate, if not the maximum possible load, then at least capable of well loading a quad-core CPU with Hyper-Threading.
Results
The previous version - 3DMark 11 - was introduced at the end of 2010. It took the developers a little over two years to prepare the next modification of the test application. Each subsequent one is traditionally an extraordinary event in the world of benchmarking and a serious challenge for current hardware. However, in this case with a new 3DMark Futuremark has tried to take into account the interests of both owners of portable systems and gadgets who want to have a certain tool for comparing dissimilar devices, and serious overclockers and enthusiasts who need a new high-tech application for measuring PC performance. At first glance, at least with the last part, Futuremark did a good job. Well, we are waiting for versions of packages for Windows RT, Android and iOS, and we include the already released application in our pool of mandatory synthetic utilities for testing video cards and processors.
Those who wish to test their system in combat conditions can already download free version 3DMark Basic Edition(967 MB). The results left in the comments will become the basis for the formation of the "people's" rating.
(Update. 08.02.2013) Thanks to all participants who left comments. The obtained additional data allowed us to understand the test results in more detail. With the exception of a few anomalies, the overall situation is fairly predictable. The Ice Storm and Cloud Gate tests are highly processor-dependent, however, the capabilities of video cards also have an impact on the final result. In Fire Strike, there is a clear hierarchy and the dependence of the result on the performance of the graphics adapter, and in the first approximation, adapters from AMD have some advantage. We have yet to study in more detail the capabilities of the new benchmark, as well as the influence of various parameters on the results obtained.
With the release of the 3DMark 11 test suite, it's time to test the most popular desktop graphics cards from AMD and NVIDIA, belonging to the lower, entry-level, mid-range and high-end segments with support for DirectX 11 technology.
The Expreview resource used 8 GeForce accelerators and 9 Radeon video cards for this purpose. Seeing 17 different DX11 accelerators on one table is very unusual and this picture is impressive.
Testing was carried out on two platforms. Processor selected for medium/powerful system Intel Core i7-920, and the work of the lower / initial was provided by the chip AMD Phenom II X2 550.
Medium/Powerful Platform Accelerators:
- GeForce GTX 580/570/480/470/465/460 and GeForce GTS 450;
- Radeon HD 5870/6870/6850/5770/5750.
Lower/Initial Platform Accelerators:
- GeForce GT 430;
- Radeon HD 5670/5570/5550/5450.
Accelerators of the middle and higher segments were tested in the Performance and Extreme settings of the benchmark, while entry-level and lower-end solutions were tested in Performance and Entry modes. The three 3DMark 11 settings (Entry, Performance, and Extreme) differ in the following settings:
As a driver for accelerators of the GeForce GTX 500 series, GeForce 263.09 WHQL was used, for other NVIDIA cards - 260.99 WHQL; for the Radeon HD 6800 series - Catalyst 10.10e Hotfix, for other AMD accelerators - Catalyst 10.11 WHQL. When changing the video card, the previous driver was carefully removed from the system using DriverSweeper.
Medium / powerful cards, installationPerformance
In this mode, the GeForce GTX 580 accelerator turned out to be the undisputed leader, scoring P6144, while the GeForce GTX 570 showed almost the same performance as the GTX 480. In the AMD camp, the P4456 received the highest score for the Radeon HD 5870, whose performance is slightly higher than the GeForce GTX 470.
In the middle segment won the palm radeon card The HD 6870, whose P4360 score comes close to the GTX 470, while the HD 6850 scores P3641, is slightly ahead of the GeForce GTX 460 1GB (P3302). Finally, the HD 5750 scores slightly better than the GeForce GTS 450.
Medium/powerfulcards, installationExtreme
The extreme testing mode puts a lot of stress on the cards: even the flagship GTX 580 failed to reach the X2000 mark.
By the way, in this mode, the gap between Radeon accelerators increased: HD 6870 scored X1407 points, outperforming GeForce GTX 470 (X1342). The difference between the GeForce GTX 460 1 GB and the GeForce GTX 465 has been greatly reduced, but both cards are still far behind the Radeon HD 6850.
Performance
In 3DMark 11, in the Performance test mode, the results of Radeon accelerators are distributed in a "ladder", in accordance with the serial number, while the result of the GeForce GT 430 is identical to the Radeon HD 5550.
Entry and lower level cards, installationEntry
In the Entry mode, the result was also expected: the accelerators were distributed according to serial numbers, although the numbers are higher. The GeForce GT 430 slightly outperformed the HD 5550. The Radeon HD 5670 still takes first place with the E2410.
lab501 testing
Of some interest are the results of testing conducted by the resource lab501. He also compared 16 video cards, and included in the review the 2-chip accelerator Radeon HD 5970, whose result came out on top in the ranking:
The processors in the CPU test deserve special attention (the GTX 580 accelerator was used as graphics):
Interestingly, the performance of the 3DMark 11 benchmark in Performance and Extreme modes, even when using the GTX 580 accelerator, is almost independent of the processor frequency. The processor starts limiting the graphics only when Entry is set:
The number of processing cores affects the result more significantly, although here the points obtained in Extreme mode almost do not depend on the processor:
On the other hand, increasing the operating frequencies of the video card gives an almost linear increase in all modes:
But the influence of the GDDR5 memory frequency is not very significant, and is noticed mainly in the Entry and Performance modes:
So, from the above tests, we can conclude that the 3DMark 11 test suite, which uses the latest DirectX 11 technologies, really clearly reflects the difference in performance of the latest generation cards.
3DMark 11 has a certain headroom: today even the GeForce GTX 580 can't score more than 2000 points in Extreme mode. In some test scenes, even this most powerful accelerator shows a rather low number of frames per second.
We can also make a preliminary conclusion that at the moment AMD video cards show slightly better results than NVIDIA accelerators of the same level. It is worth noting that the SLI mode does not work now, although such problems are not observed in CrossFire. It seems that soon both AMD and NVIDIA will present new drivers optimized for 3DMark 11. Then, perhaps, the results will change somewhat.
As a rule, when testing new video cards, we select rivals based on their cost and market positioning. But sometimes you want to see the whole picture - how the performance of a full range of modern video cards compares, from the most budgetary to top-end solutions. This test is just dedicated to this issue.
We took 30 topical this moment video cards. In the photo below you can see what this magnificence looks like when you put them all together. For this group photo, the video cards had to be placed on the floor, since they simply did not fit on any office table.
Testing was carried out on a stand with the following configuration:
Since the graphics turned out to be quite voluminous, in this material we will consider the results obtained in only one application - 3DMark 11. Over the past decade, a test suite called 3DMark has become the de facto industry standard due to its simplicity, high repeatability of results and the use of the most modern technologies visualization of 3D applications.
Let's see how the "ranking table" looks like for modern video cards that support DirectX 11. And we'll start with the 3DMark 11 Graphics Score results obtained in the Extreme preset, that is, the maximum load on the computer's video subsystem.
Among the dual-core video cards, the Radeon HD 6990 looks the best, and in the single standings, NVIDIA's single-chip flagship GeForce 580 GTX is in the lead. Further, the results decrease almost linearly, and another qualitative leap occurs when switching to mid-range solutions using a 128-bit memory bus (Radeon HD 5770).
Of course, we also got results in the Performance and Entry presets, which are shown below. To avoid unnecessary confusion and make it easier to compare the results of different presets (if anyone needs it), the sorting of the list of video cards has remained the same.
The picture remains the same, except for the results NVIDIA graphics cards somewhat grown up in comparison with the nearest neighbors from the red camp.
When switching to the Entry preset, NVIDIA accelerators again slightly improve their performance in relation to the nearest neighbors manufactured by AMD. However, for top-end video cards, this does not make much difference. It is unlikely that a powerful video card is purchased for playing at a resolution of 1024x600 with low settings graphics quality. So the results in this mode may be interesting for graphics adapters of the middle and budget class.
The next test we paid close attention to was 3DMark 11 Combined. This test creates a load on both the central processor and the graphics accelerator. The CPU calculates the "physics" of rigid bodies, and the GPU calculates the movement of tissues using the Bullet engine, which runs through DirectCompute. We are interested in how exactly video accelerators cope with the computational load, since the same central processor was used during all tests. As before, let's start by reviewing the results in the "heaviest" mode - Extreme (the order of video cards on the diagram remains the same).
The overall picture remains the same as we saw in the Graphics test.
When switching to Performance mode radeon video card The HD 6990 unexpectedly loses its lead, and the single-chip GeForce GTX 580 overtakes the dual-chip Radeon HD 5970.
In the Entry mode, miracles are observed - the GeForce GTX 580 becomes the absolute leader. Apparently, dual-chip video cards are simply not able to reveal their potential with the most lightweight graphics quality settings. In general, NVIDIA solutions in this mode look somewhat preferable to their closest competitors.
Perhaps some readers will have a question - why are we not presenting the final results of 3DMark 11? The fact is that the final assessment includes the results of not only graphic tests, but also tests of "physics" carried out on the central processor. Therefore, directly compare the overall result of 3DMark 11 obtained on different systems turns out to be difficult. At the same time, the test of the graphics subsystem, especially in heavy mode, depends on the power of the central processor rather weakly, unless, of course, you are testing on a not very old CPU. For those who are interested, we inform you that the result of 3DMark Physics Score of our test bench was equal to 8750 points. And finally, below is a table with the results of all tests in various modes, including the total score (Total).
| Entry, 1024×600 | Performance, 1280×720 | Extreme, 1920×1080 | |||||||||
| 3DMark 11.ver. 101 Prof. | Total | Graphics | Combined | Total | Graphics | Combined | Total | Graphics | Combined | ||
| HD 6990 (830/830/1250) 4096 MB GDDR5 256-bit | 12195 | 15556 | 7274 | 8968 | 9335 | 7267 | 3295 | 3041 | 3472 | ||
| GTX 590 (613/1225/855) 3072MB GDDR5 384-bit | 12284 | 15426 | 7833 | 8806 | 8985 | 7654 | 3006 | 2836 | 2550 | ||
| HD 5970 (725/725/1000) 2048MB GDDR5 256-bit | 10125 | 11455 | 6899 | 7092 | 7018 | 5980 | 2646 | 2420 | 2803 | ||
| GTX 580 (772/1544/1002) 1536MB GDDR5 384-bit | 9518 | 9997 | 8074 | 6099 | 5740 | 6183 | 2015 | 1808 | 2369 | ||
| HD 6970 (880/880/1375) 2048 MB GDDR5 256-bit | 8039 | 8222 | 6091 | 5299 | 4941 | 5039 | 1794 | 1611 | 1991 | ||
| GTX 570 (732/1464/950) 1280MB GDDR5 320-bit | 8529 | 8725 | 7009 | 5340 | 4964 | 5233 | 1737 | 1557 | 1981 | ||
| GTX 480 (701/1401/924) 1536MB GDDR5 384-bit | 8461 | 8598 | 7083 | 5322 | 4928 | 5360 | 1717 | 1534 | 2021 | ||
| HD 6950 (800/800/1250) 2048 MB GDDR5 256-bit | 7541 | 7575 | 5790 | 4878 | 4499 | 4722 | 1599 | 1432 | 1810 | ||
| HD 5870 (850/850/1200) 1024MB GDDR5 256-bit | 6831 | 6772 | 4985 | 4491 | 4132 | 4189 | 1570 | 1408 | 1747 | ||
| GTX 560 Ti (880/1760/1050) 1024MB GDDR5 256-bit | 7354 | 7363 | 5480 | 4555 | 4190 | 4237 | 1521 | 1357 | 1771 | ||
| HD 6870 (900/900/1050) 1024MB GDDR5 256-bit | 6677 | 6595 | 4856 | 4355 | 3999 | 4036 | 1401 | 1251 | 1600 | ||
| GTX 470 (608/1215/837) 1280MB GDDR5 320-bit | 7109 | 6953 | 5798 | 4316 | 3956 | 4270 | 1376 | 1227 | 1590 | ||
| HD 5850 (725/725/1000) 1024MB GDDR5 256-bit | 5919 | 5705 | 4286 | 3822 | 3463 | 3586 | 1302 | 1162 | 1476 | ||
| GTX 560 (810/1620/1000) 1024MB GDDR5 256-bit | 6370 | 6192 | 4732 | 3839 | 3472 | 3659 | 1260 | 1118 | 1504 | ||
| GTX 460 (726/1451/900) 1024MB GDDR5 256-bit | 5812 | 5562 | 4263 | 3476 | 3124 | 3282 | 1136 | 1007 | 1354 | ||
| HD 6850 (775/775/1000) 1024MB GDDR5 256-bit | 5708 | 5470 | 4131 | 3664 | 3309 | 3446 | 1135 | 1007 | 1331 | ||
| HD 5830 (800/800/1000) 1024MB GDDR5 256-bit | 5508 | 5237 | 3993 | 3361 | 3012 | 3196 | 1080 | 961 | 1228 | ||
| HD 5770 (850/850/1200) 512MB GDDR5 128-bit | 4535 | 4204 | 3234 | 2789 | 2478 | 2584 | 862 | 770 | 915 | ||
| GTX 550 Ti (900/1800/950) 2048MB GDDR5 192-bit | 4279 | 3925 | 3065 | 2502 | 2214 | 2282 | 806 | 710 | 975 | ||
| GTS 450 (810/1620/902) 1024 MB GDDR5 128-bit | 3822 | 3477 | 2666 | 2211 | 1944 | 2026 | 704 | 623 | 809 | ||
| HD 6670 (810/810/1000) 1024MB GDDR5 128-bit | 3037 | 2719 | 2049 | 1832 | 1601 | 1670 | 576 | 511 | 630 | ||
| HD 5670 (775/775/1000) 512MB GDDR5 128-bit | 2860 | 2542 | 1950 | 1683 | 1463 | 1561 | 475 | 420 | 540 | ||
| HD 6570 (650/650/900) 1024 MB GDDR3 128-bit | 2306 | 2033 | 1505 | 1357 | 1181 | 1183 | 413 | 368 | 430 | ||
| GT 440 (810/1620/900) 1024 MB DDR3 128-bit | 2143 | 1878 | 1409 | 1220 | 1058 | 1071 | 377 | 334 | 408 | ||
| HD 5570 (650/650/667) 1024 MB GDDR3 128-bit | 2111 | 1856 | 1361 | 1212 | 1055 | 1031 | 352 | 315 | 348 | ||
| HD 5550 (650/650/900) 512MB GDDR5 128-bit | 2156 | 1878 | 1466 | 1275 | 1099 | 1183 | 349 | 309 | 383 | ||
| GT 430 (730/1460/900) 1024 MB DDR3 128-bit | 1976 | 1722 | 1304 | 1122 | 971 | 986 | 347 | 307 | 379 | ||
| GT 520 (810/1620/900) 512MB DDR3 64-bit | 995 | 854 | 618 | 550 | 473 | 467 | 177 | 158 | 174 | ||
| HD 6450 (625/625/667) 1024 MB GDDR3 64-bit | 905 | 772 | 571 | 517 | 444 | 443 | 148 | 133 | 143 | ||
| HD 5450 (650/650/650) 512MB GDDR3 64-bit | 686 | 566 | 508 | 379 | 318 | 386 | 111 | 98 | 128 | ||
As owners of the extended version of 3DMark 11, we decided to get acquainted with the new test package for graphics cards.
3DMark 11 is designed to evaluate computer performance, tests make extensive use of DirectX 11 features, including tessellation, shader calculations, and multithreading.
The program greets the user with a window for selecting a group of tests and a mode:
The Entry (E) group settings were created for testing laptops and netbooks, the resolution of 1024x600 was chosen in accordance with the purpose of the group.
The Performance (P) group is designed to evaluate the performance of most desktop computers, standard definition for this group 1280×720 (720p).
The third group, Extreme (X), sets the resolution to 1920x1080 (1080p). This is the heaviest standard set benchmarks designed for high performance computers.
In the lower half of the window, you must select the testing mode, and then run the package.
In the second tab “Advanced”, you can select the necessary tests from the package, as well as customize to your own requirements:
The first graphics test is based on the Deep Sea scene, which can be seen in the video below. This test does not use tessellation, but the test is made heavier by lighting from multiple sources and rendering complex shadows:
In the second test, the same scene is used, but the average tessellation level has already been added, the lighting level has been slightly reduced:
The third test with medium tessellation and single source lighting is based on the High Temple episode:
The same scene is used in the fourth test of the video card, but this time the tessellation level is already complex, and the calculation of shadows from several sources has also become heavier:
The physics test simulates the physics of bodies with a large number of objects:
The combined test simulates the physics of bodies with a moderate number of objects; the DirectCompute API and the Bullet Physics libraries are used to calculate the physics. Together with the modeling of the physics of objects, the average level of tessellation is used:
We decided to test the available system. Test system configuration:
The EVGA GeForce GTX 470 SuperClocked video card acted as the first tester:
The operating system was installed on the system Windows system 7 64-bit, used for video card current version NVIDIA drivers GeForce/ION 260.99.
To begin with, the Performance test group was selected with default settings. Frankly, we were somewhat surprised by the "brakes" when passing the tests, as conceived by the developers intended for the bulk of the PC. What can I say, everything can be seen in the results:
The video card scored 4406 points, the average FPS did not rise above 26, and in the bulk it could not even reach 20 frames / s.
The result of passing tests with Extreme settings was already predictable:
The result is 1370 points, the number of frames per second in the tests resembled a turn-based strategy.
So the first acquaintance with 3DMark 11 took place. The complexity of the tests is quite understandable - the performance evaluation indicators are aimed far ahead, and the current generation of "average" video cards are not able to cope with the complexity inherent in the tests. We can only guess what we can expect in future games that use the full capabilities of DirectX.
NVIDIA will announce official details about the GeForce graphics card GTX 1070 Ti, which will occupy an intermediate position between the GeForce GTX 1070 and GTX 1080. The formal announcement and the start of accepting pre-orders for the GTX 1070 Ti in retail stores are scheduled for next Thursday, October 26, and the first reviews will be published a week later, on November 2 accelerator and the shipment of cards to customers will begin.
Meanwhile, enthusiasts continue to wool the Internet in search of information about the performance of new items. This was found in the 3DMark Achievements database, where there are four results for the GeForce GTX 1070 Ti: two in the Fire Strike Extreme subtest (9449 and 9546 Graphics points), one each in Time Spy (6777 Graphics points) and in 3DMark 11 Extreme (9002 Graphics points) .
Futuremark test suites are usually not a reliable tool for determining the characteristics of video cards, and this case is no exception. In three entries in the database, the core frequency was 1886 MHz, and in one it was 1607 MHz (recommended values for the GTX 1070 Ti are from 1607 to 1683 MHz, excluding boost mode). Frequency buffer memory GDDR5 in 2003 (8012) MHz was correctly defined, as was its 8 GB capacity.
Taking the 3DMark results table from the Radeon RX Vega 56 as a reference, we can roughly estimate what the upcoming new product is capable of. The average 3DMark Fire Strike Extreme score of 9498 points is 1% higher than the RX Vega 56, while the 3DMark Time Spy score of 6777 means a 7.9% improvement over the same graphics card. Given the possible overclocking limitation of the GeForce GTX 1070 Ti adapter, the winner in his confrontation with the representative AMD families Radeon is too early to name. Also note that the tested GTX 1070 Ti is closer to the GTX 1080 than the GTX 1070 (especially in Time Spy).
| 3DMark (Graphics Score) including GeForce GTX 1070 Ti preliminary results | ||||||
|---|---|---|---|---|---|---|
| Permission | NVIDIA GeForce GTX 1070 (1506/8008 MHz, 8 GB) | NVIDIA GeForce GTX 1070 Ti (?/8012MHz, 8GB) | NVIDIA GeForce GTX 1080 (1607/10008 MHz, 8 GB) | AMD Radeon RX Vega 56 (1471/1600 MHz, 8 GB), Turbo | AMD Radeon RX Vega 64 (1546/1890 MHz, 8 GB), Turbo | |
| fire strike | 1920×1080 | 17 519 | 21 694 | 19 848 | 22 503 | |
| Fire Strike Extreme | 2560×1440 | 8298 | 9498 | 10 264 | 9488 | 10 711 |
| Fire Strike Ultra | 3840×2160 | 4079 | 5001 | 4774 | 5 400 | |
| time spy | 2560×1440 | 5693 | 6777 | 7111 | 6281 | 7 079 |
As for the result of the GeForce GTX 1070 Ti in 3DMark 11 Extreme, given the low popularity of this benchmark, we turned to the database of overclocking achievements