The introduction of GeForce RTX 50 series graphics cards has certainly left no one indifferent. Graphics adapters based on Blackwell architecture evoke the widest possible range of feelings - from irreconcilable indignation at the excessive prices due to the initial shortage to admiration for the quality of DLSS 4 and fps after activating Multi Frame Generation. Gradually understanding the features of the new products, we continue to explore specific implementations of the top models. Video card manufacturers have put a lot of effort into surprising us with their new designs. And it's not just about the recommended price. Today we are reviewing MSI GeForce RTX 5080 16G SUPRIM SOC, a sturdy, statuesque representative of the new line with a tangible base overclock and a powerful cooling system.
NVIDIA GeForce RTX 5080
As a reminder, GeForce RTX 5080 is the second highest-end graphics card in NVIDIA's new graphics lineup. The model is equipped with a GB203 GPU based on the Blackwell architecture with 10,752 active cores. The GPU is manufactured using TSMC's 5-nm process technology with certain additional optimizations based on NVIDIA's requirements.
| GeForce RTX 5080 | GeForce RTX 4080 SUPER | GeForce RTX 4080 | |
| Graphics processor | AD104-400 | AD103-350 | AD104-250 |
| Architecture | Blackwell | Ada Lovelace | Ada Lovelace |
| Technology | 5 nm, TSMC | 5 nm, TSMC | 5 nm, TSMC |
| CUDA computers | 10 752 | 10 240 | 9728 |
| TMU / rasterizers | 336/128 | 320/112 | 304/112 |
| Tensor kernels | 336 (5th generation) | 320 (4th generation) | 304 (4th generation) |
| RT blocks | 84 (4th generation) | 80 (3th generation) | 76 (3th generation) |
| GPU clock speeds | 2295/2617 MHz | 2295/2550 MHz | 2205/2505 MHz |
| Memory capacity | 16 GB GDDR7 | 16 GB GDDR6X | 16 GB GDDR6X |
| Memory bus | 256 bits | 256 bits | 256 bits |
| Effective memory frequency | 30,000 MHz | 23,000 MHz | 22,400 MHz |
| Memory bandwidth | 960 GB/s | 736 GB/s | 717 GB/s |
| Video card power consumption (TGP) | 360 W | 320 W | 320 W |
| Starting recommended price | $999 | $999 | $1199 |
| Release date | January 2025 | January 2024 | November 2022 |
In the case of the GeForce RTX 5080, the GB203-400 processor has a frequency formula of 2295/2617 MHz. The video card uses a 256-bit bus and is equipped with 16 GB of memory of the new GDDR7 standard. Interestingly, the RTX 5080 has a maximum speed mode with an effective frequency of 30,000 MHz (RTX 5090 has 28,000 MHz). Combined with a 256-bit bus, this results in a bandwidth of 960 GB/s. This is 30% more than the same GeForce RTX 4080 SUPER. The declared power consumption of GeForce RTX 5080 is 360 watts.
NVIDIA has offered a reference model of the GeForce RTX 5080 Founders Edition, but the vast majority of offers on the market will be original models from partners.
After the announcement of the new line of graphics cards, MSI seems to have offered perhaps the widest range of GeForce RTX 5080. There are models with liquid cooling systems, top-of-the-line SUPRIM versions, adapters of the new VANGUARD line, GAMING TRIO series, mysterious SHADOW, “professionalized” INSPIRE, more affordable VENTUS versions, as well as variations with white exterior design for assembling systems in a similar range. In total, the manufacturer offered about 20 RTX 5080 modifications. The basic features of all versions can be found on the manufacturer's website, and this time we will take a closer look at one of the modifications - MSI GeForce RTX 5080 16G SUPRIM SOC.
The model received a significant factory overclocking of the GPU, the frequency formula of which has been changed from the standard 2295/2617 MHz to 2295/2745 MHz. GDDR7 memory chips operate at the recommended 30,000 MHz (effective).
MSI GeForce RTX 5080 16G SUPRIM SOC package contents
The video card is offered in a large cardboard box with a light design, a large GeForce RTX 5080 line logo, and the SUPRIM series in the center of the front panel.
Design, construction and layout
SUPRIM graphics cards are easily recognizable by their distinctive exterior design. It features a design with recognizable geometric accents that vaguely resemble diamond cuts.
Dynamic asymmetry, directional grinding of surfaces, a combined protective cover with sharp corners of the elements. All this adds a certain aesthetic integrity and completeness to the design. This is all without regard to the fact that the video card is quite massive. The model has dimensions of 360x150x76 mm, so it will surely require a fairly spacious case and four free slots for installation. To complete the overall impression, we note that, in addition to its “stout build,” the video card weighs 2.614 kg. Therefore, the included support is not a bonus freebie on the key, but an important element of protection for the PCI-E x16 slot of the motherboard.
MSI GeForce RTX 5080 16G SUPRIM SOC uses a variation of the Hyper Frozr cooler to cool the MSI GeForce RTX 5080 16G. This time we are talking about a very massive radiator block consisting of aluminum plates with a wave-like edge profile (Wave Curved 4.0) to increase the airflow area.
At the heart of the cooler is a large evaporation chamber that accelerates heat transfer from the GPU chip and memory chips. Heat pipes accelerate heat dissipation to remote areas of the radiator structure. One of the interesting technological solutions is the square formation of the tubes in the contact area with the evaporation chamber to increase the contact area and reduce the gap between the “heat pipes”.
Three 100 mm StormForce fans are used to blow the radiator unit. Their specificity is that the seven blades are connected along the outer radius in a continuous circular arc. There is also a specific finning on the surface. The developers assure that this combination allows to increase the generated airflow and make it more accentuated. To increase durability, the fans are equipped with double ball rollers.
Despite the large cooling system, the PCB of the video card is much more compact in length. The power subsystem has a 19-phase configuration in general. The GPU uses a 16-phase VRM configuration based on DrMOS elements (Monolithic MPS MP87993) with an operating current of up to 90 A. The stabilizer parameters are controlled by the MPS MP29816 controller.
To stabilize the power supply to the memory chips, three phases are used in the circuit. The power assemblies are the same as for the GPU - MP87993, and the parameters are controlled using the Monolithic MP2988 controller.
As for the GDDR7 memory chips themselves, Samsung's K4VAF325ZC-SC32 chips are used here, which can operate at an effective 32,000 MHz according to the specification. We immediately note the potential for additional overclocking, given that the RTX 5080 has a standard value of 30,000 MHz. The video card has a 256-bit bus, and the total memory capacity is 16 GB. So, there are 8 GDDR7 memory chips on the PCB, which are located in a U-shaped manner near the GPU.
Unlike the reference GeForce RTX 5080, which has a combined layout, the MSI GeForce RTX 5080 SUPRIM SOC uses a single PCB with an interface comb and video output ports.
One 16-pin 12VHPWR connector is provided near the top edge of the PCB to connect additional power to the video card. The declared power consumption level of GeForce RTX 5080 is 360 watts. For the overclocked version of the MSI video card, an identical figure has been announced, but the developers still recommend using a PSU with a capacity of 850 W or more, while NVIDIA's recommendations are 750 W. During practical experiments, we will check the actual consumption figures.
If your existing power supply does not have a suitable connector, you will have to use the included adapter from three 8-pin PCI-E 12V connectors to one 12VHPWR.
To simplify visual inspection of the connection quality, the developers suggested painting the connector guides yellow (MSI Dual Color 16-pin Cable). 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. This is a fundamentally important condition for 12VHPWR, which allows you to transmit up to 600W.
As expected, MSI GeForce RTX 5080 SUPRIM SOC is equipped with two BIOS chips with firmware that provide different cooling system algorithms. The basic Silent offers minimal fan speeds to reduce noise under load. While Gaming provides a more aggressive algorithm that allows for lower operating temperatures.
MSI GeForce RTX 5080 SUPRIM SOC has an additional backlight. The SUPRIM name is clearly visible on the top edge of the adapter with ARGB illumination. In addition, decorative segments with light guides are provided in the gaps between the cooling fans. These “fireflies” can be seen if the video card is installed in a case with a vertical positioning.
This is a single light accent - the series logo at the junction of the back and side surfaces of the adapter. The profile of the faceted diamond will complement the overall decorative composition.
The back of the video card is covered with an aluminum plate. It is an unchangeable attribute of high-end video cards. Usually, this element adds strength to the overall structure and protects the PCB from accidental mechanical damage. An additional option is to use the plate to dissipate heat from the elements on the back of the PCB.
A significant difference between the length of the PCB and the overall dimensions of video cards recently hints at the use of a cooler that involves blowing the radiator through. MSI GeForce RTX 5080 SUPRIM SOC also uses this approach. There is a perforated area on the rear panel that allows you to organize the corresponding principle.
The interface panel has a classic set of video outputs - three DisplayPort 2.1b and one full-size HDMI 2.1b. Almost half of the steel cover is occupied by a segment with through holes for improved ventilation.
Test system configuration
- Processor: Ryzen 7 9700X (8/16; 3.8/5.5 GHz; 32 MB L3)
- Motherboard: MSI MAG X870E CARBON WIFI
- CPU cooling: MSI MAG CORELIQUID I360
- Video card: MSI GeForce RTX 5080 16G SUPRIM SOC
- RAM: Kingston FURY Renegade DDR5-7200 2x16 GB (KF572C38RWK2-32)
- Storage device: Kingston KC3000 (SKC3000S/1024G)
To test the video card, we used a platform based on a Ryzen 7 9700X processor (8/16; 3.8/5.5 GHz; 32 MB L3) and an MSI MAG X870E CARBON WIFI motherboard based on the AMD X870 chipset. An eight-core chip is objectively a good option for a gaming system. Of course, X3D chips with an additional L3 cache buffer would be nice for top-end solutions, but Ryzen 7 9700X is much more affordable, so it won't be a significant limitation, especially for 1440p and 4K modes.
To increase the computing power of Ryzen 7 9700X, we used a 105W TDP mode instead of the standard 65W. MSI and other motherboard manufacturers have made it as easy as possible to switch between 65W and 105W. 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 the maximum multi-threaded load, the performance increase is ~15%. So during gaming tests, this is more of a safety net.
If we increase TDP, we need a relatively efficient cooling system. Here we have already relied on the MSI MAG CORELIQUID I360 liquid cooling system. It's probably somewhat excessive for an 8-core, but you can't remember when the efficiency of the SRO was too much.
In operation
First of all, we checked the temperature conditions of the video card. At rest, the GPU heated up to 44C, and the memory chips heated up to 54C.
In such conditions, the cooling fans stop, and the graphics adapter operates silently. GPU-Z monitoring records power consumption at 15 watts.
MSI GeForce RTX 5080 16G SUPRIM SOC comes with the Silent mode activated as standard. In this way, the developers obviously want to emphasize the low noise level of the video card. Under similar conditions, the GPU heated up to 66C and the memory chips to 68C. At the same time, the fan speed increased only up to 1200 rpm. The adapter really works very quietly, even to some extent unusually quiet, as for a GeForce RTX 5080 with a TGP of over 300 watts.
We also tested the Gaming profile by switching to the appropriate mode using the toggle switch on the top edge of the video card. In this case, under gaming load, the GPU temperature rose only to 58C, while the memory chip heated up to 62C. The fan speed increased to 1450 rpm. The overall noise level was obviously higher compared to the conditions we got with Silent, but overall it remains below average. Excellent performance for a model of this class.
Test results
Since this is not the first time we have seen GeForce RTX 5080, we have some idea of the capabilities of this line of graphics cards. Testing of the next model actually repeats the previous results. Here are the screenshots with the results of MSI GeForce RTX 5080 16G SUPRIM SOC, which were obtained during the research.
“Synthetic” benchmarks and certain computing tasks demonstrate the new product's superiority over its predecessor, the GeForce RTX 4080 SUPER.
For ease of evaluation, we add the performance of GeForce RTX 5070 Ti 16 GB and GeForce RTX 4070 Ti SUPER 16 GB graphics cards to the charts. In all cases, the results were obtained on models with relatively low factory GPU overclocking, which usually adds 1-3% of performance. For this reason, the corresponding positions are labeled “OS”.
When testing the capabilities of MSI GeForce RTX 5080 16G SUPRIM SOC in 1440p mode, we get the expected ~10% advantage over GeForce RTX 4080 SUPER. If we compare the results with those of the GeForce RTX 4070 Ti SUPER, which we usually recommend as a starter for advanced gaming configurations, the advantage is ~25% on average, but it largely depends on the specific project.
Games in 4K mode
In 4K mode, GeForce RTX 5080 series graphics cards gain more advantages. The average difference with the GeForce RTX 4080 SUPER increases to ~15+%. Obviously, in addition to the architectural features of the GPU, the significantly higher bandwidth of GDDR7 memory is at work.
The Dying Light 2 benchmark shows a 19% advantage over its predecessor, and the GeForce RTX 4070 Ti SUPER lags behind by 44%. The situation is rather exceptional, but there may be such cases.
In the Forza Motosport car simulator, the RTX 5080 was only 10% faster than its predecessor and almost a third faster than the GeForce RTX 4070 Ti SUPER.
Cyberpunk 2077 gets a very significant acceleration when using models on GPUs with Blackwell architecture. The 27% advantage of the RTX 5080 over the RTX 4080 SUPER is the other extreme, which is generally atypical. Another indicator of the game's “favor” for the new GPUs is that the RTX 5070 Ti managed to catch up with the RTX 4080 SUPER.
Assassin's Creed Mirage can also be classified as a project that reacts moderately to innovations. Of course, there is a 10% difference, but this is definitely not a reason to even think about a possible upgrade. Especially when we are talking about 100+ fps in general.
The owner of a GeForce RTX 5080 will watch the adventures in Indiana Jones and the Great Circle with ~8% more frames than a user of an RTX 4080 SUPER with identical settings. Is this a reason to be happy? It depends on the price.
In Alan Wake 2, we have +10% more frames per second compared to its predecessor and a quarter higher performance than the RTX 4070 Ti SUPER 16 GB.
The Ukrainian long-in-the-works game S.T.A.L.K.E.R. 2: Heart of Chornobyl, which has already been played by more than 6 million players, prefers the Blackwell model, offering 13% higher performance when the platform is charged with GeForce RTX 5080.
An even bigger increase (~16%) is seen in Black Myth: Wukong. By the way, only the GeForce RTX 5080 was able to provide more than 60 fps in 4K mode with maximum picture settings without additional picture scaling.
DLSS 4 and Multi Frame Generation
When the frame count in games does not please with comfortable performance, the well-known DLSS and frame generation technologies come to the rescue. Both mechanisms have been improved with the introduction of the GeForce RTX 50 line.
Some of the functionality of the new DLSS 4 version is available to models of previous generations, but the mechanism of multiple frame generation is an exclusive feature of the RTX 50.
Is it possible that even a GeForce RTX 5080 is not enough for gaming? It depends on the settings. If you rely on the modes with the highest quality ray tracing, then yes, even such a video card may not be enough. Good examples are Cyberpunk 2077 and Alan Wake 2 in RT Overdrive and RT Ultra modes, respectively.
The available 15-20 fps in native mode is hardly the result you'd expect from a graphics card of this level. It is for such cases that DLSS + frame generation is the combination that allows you to “stretch” the game and reassure owners of even top-end graphics adapters.
Multi Frame Generation is capable of improving quantitative indicators many times over. Judging by the results obtained in games that already have its support, MFG is able to generate as many fps as needed. Of course, it is not without issues of latency and visual artifacts, but these are issues for optimization of specific projects. With the right effort from developers, the results are impressive.
Overclocking
Габаритна система охолодження, яка легко справляється з роботою відеокарти у штатному режимі якоюсь мірою провокує на дослідження "понаднормових" можливостей графічного адаптера. Розробники дозволяють підвищити енергетичний ліміт відеокарти до 400 Вт (+11%). Використовуючи такі можливості, намагаємось знайти граничні частотні режими.
By experimenting with modes that maintain the stability of the video card, the base frequency of the GPU was increased by 385 MHz - from 2295 to 2680 MHz. At the same time, the average Boost frequency increased from 2745 MHz to 3130 MHz.
The GDDR7 memory chips also operated at significantly higher than recommended values - 2250 MHz (36,000 effective) instead of 1875 MHz (30,000 effective). This resulted in a 20% increase in bandwidth, from 960 GB/s to 1152 GB/s. As we noted, the video card is equipped with Samsung chips, which according to the specification should operate at an effective 32,000 MHz, so a certain reserve is provided initially. Unfortunately, the value of 2250 MHz is still a purely software limitation at the driver level, which does not allow us to raise the frequency bar even higher.
Після додаткового розгону температура графічного процесора під навантаженням підвищувалась до 62С, і мікросхеми GDDR7 також прогрівались до такої ж позначки. При цьому оберти вентиляторів підвищувались до ~1550 об/хв. Пікова частота GPU реєструвалась на рівні 3240 МГц.
And what about the actual performance increase after overclocking? Given the significant increase in GPU/memory frequencies, the response is appropriate. In the 3DMark Time Spy Extreme test in 4K mode, the Graphics Score increased from 16,570 points to 18,411 points. That is an increase of 11%. In games, the increase in fps was also appropriate. In a situation where 10-15% is the difference between generations, an additional +10% generally significantly improves the perception of new graphics cards. In the case of the GeForce RTX 5080, this acceleration usually allows you to reach the performance of the GeForce RTX 4090 24GB.
Power consumption
GeForce RTX 5080 16GB series graphics cards have a claimed power consumption of 360 watts. During our experiments in games, we recorded values for the MSI GeForce RTX 5080 SUPRIM SOC at 332-337 watts.
Under similar conditions, the GeForce RTX 4080 SUPER needed about 320 watts. The natively overclocked GeForce RTX 5070 Ti consumed ~300 watts. So, against the background of these values, the original version of the GeForce RTX 5080 has good overall energy efficiency.
After additional overclocking of MSI GeForce RTX 5080 16G SUPRIM SOC with an increase in Power Limit in games, the power consumption of the video card increased to ~365-370 watts. In fact, we have an increase in consumption proportional to the increase in performance. This further justifies additional overclocking experiments, especially in cases where the efficiency of the standard cooling system allows it.
Additionally, let's pay attention to the load distribution on the power lines. The video card received only about 5 watts of power through the PCI-E x16 slot, the rest was transmitted through the 12HVPWR connector.
