
Table of Contents
Today, we present Overclockers.com’s first 5000 series Blackwell-based video card review. The new Blackwell architecture promises increased performance through various methods, including introducing DLSS4 (from 3) and neural rendering, where the GPU uses generative AI to create parts of a frame. This new technology comes from a huge GB202 monolithic die built on the same 5nm (TSMC 4N) and uses the first implementation of GDDR7 vRAM and a PCIe 5.0 connection. Additionally, you get 5th-Gen Tensor cores, updated streaming multiprocessors, and fourth-gen ray tracing cores. It’s a generational upgrade in many things, but rasterization doesn’t take a significant leap forward.
The card on our test bench is MSI’s flagship air-cooled variant, the RTX 5080 Suprim SOC. It’s clear the card has a premium design and is crafted from premium materials. With a multifaceted, almost diamond-like design, it’s sure to be the centerpiece of your PC. It even has some tasteful RGB lighting showing off the branding. You can read on to see details about this beast of a card and how it performed in our test suite.
MSI RTX 5080 Suprim SOC (Image Courtesy of MSI)
Blackwell Architecture and Technologies

Nvidia’s new Blackwell architecture is again fabricated using TSMC’s 4N manufacturing process. There wasn’t a process shrink, but that didn’t stop Nvidia from jamming in 45.6 billion on the GB203(-400-A1) die. Compared to the RTX 4080, they have remarkably similar die sizes, 378 mm² vs. 379 mm², transistors, while the RTX 5080 has slightly faster clock speeds. The Suprim SOC has a base core speed of 2,295 MHz, with a boost clock starting at 2,745 MHz (~200 MHz difference).
Power consumption is listed as 360W (Total Graphics Power) with partner cards like our MSI, potentially increasing that limit a bit. Compared to the 4080, that’s an increase of around 40W over the last generation—not what we like to see.
The GB203 GPU on the RTX 5080 comes equipped with 5 Graphics Process Clusters (GPCs), 40 Texture Processing Clusters (TPCs), 84 Streaming Multiprocessors (SMs), and a 256-bit memory interface (8x 32-bit memory controllers). Nvidia states the TFLOP rate for the FP64 cores is 879.3 GFlops (1:64), while FP32 sits at 56.2 TFLops. Worth noting is the Blackwell generation of SM features concurrent FP32+INT32 capability on all SIMD units. The pixel fill rate is 335 GPixel/s with a texture rate of 879 GTexels /s. A bit more than the RTX 4080 Super.

Below is a close-up view of a single Streaming Multiprocessor and everything it contains.

The next evolution in neural rendering technologies utilizing the RTX Tensor Cores is DLSS 4. The latest iteration features Multi Frame Generation (75 games and apps support upon launch); DLSS Multi Frame Generation (MFG) generates up to three additional frames per traditional rendered frame, working with an entire ecosystem of DLSS technologies to multiply framerates up to an astounding 8x over more traditional ‘brute-force’ rendering. DLSS 4 also introduced the most significant upgrade to its AI models since DLSS 2 in 2020.
DLSS Ray Reconstruction/Super Resolution/DLAA will now be powered by the first real-time application of the ‘transformer model’, the same advanced architecture powering AI models like ChatGPT, Flux, and Gemini. The Transformer models improve image quality and temporal stability, less ghosting, and higher detail in motion. Any DLSS games with these features can be upgraded to the new DLSS transformer model for improved performance and IQ.
For streamers, Nvidia also updates its NVENC encode and decode accelerators (four each -3/4 NEVENC and 2/4 NVDEC are enabled). The new 9th-gen encode accelerations come with 4:2:2 AV1 and HEVC encoding support.
Details aside, the table below lists the specifications for the new ADA GPUs, including our review sample, MSI RTX 5080 Suprim SOC.
Specifications and Features
Nvidia RTX 5000 Series Specifications | ||||
Model | RTX 4090 (24GB) | RTX 4080 (16GB) | MSI RTX 5080 Suprim SOC (16GB) | RTX 5090 |
Architecture | Ada Lovelace | Blackwell | ||
Manufacturing | TSMC 4N (5 nm) | |||
CUDA Cores | 16,384 | 9,728 | 10,752 | 21,760 |
RT Cores (Gen 3) | 144 | 80 | 84 | 170 |
Tensor Cores | 576 | 320 | 336 | 680 |
Texture Units | 576 | 320 | 336 | 680 |
ROPs | 192 | 112 | 128 | 176 |
L2 Cache | 96 MB | 64 MB | 64 MB | 96 MB |
Base Clock (MHz) | 2,230 | 2,210 | 2,295 | 2,017 |
Boost Clock (MHz) | 2,520 | 2,510 | 2,745 | 2,407 |
Memory | 24 GB GDDR6X | 16 GB GDDR6X | 16 GB GDDR7 | 32 GB GDDR7 |
Memory Speed (Gbps) | 1008 | 736 | 960 | 1.79 (TB/s) |
Memory Bus | 384-bit | 256-bit | 256-bit | 512-bit |
Supplementary Power | PCIe 5.0 16-pin 3x 8-pin | PCIe 5.0 12-pin 2x 8-pin | PCIe 5.0 16-pin 3x 8-pin | PCIe 5.0 12-pin 3x PCIe 8-pin |
Standard Display Connectors | HDMI (2.1) 3x DisplayPort (1.4a) | HDMI (2.1a) 3x DisplayPort (2.1b) | ||
Max Resolution | 8K (7680 x 4320) | |||
TDP | 450 W | 320 W | 360 W | 575 W |
Release Date | 10/12/22 | 11/15/22 | 1/30/25 | 1/23/25 |
MSRP | $1599 | $1,199 | $999 | $1,999 |
The MSI RTX 5080 Suprim SOC
MSI’s Suprim series cards have been welcomed warmly since their inception a couple of generations ago. Their flagship SKU (if you’re not counting the liquid-cooled version) boasts premium build quality and appearance, a more robust power delivery, the Hyper Frozr thermal design, and factory overclocks, which sets it apart from the Founder’s Edition.
The card gets its power from a single 12VHPWR connector capable of supplying up to 675W of in-spec power. MSI recommends an 850W power supply, which is enough, even with high-end processors and some overclocking. It also has two BIOS, Gaming and Silent, which adjusts the fan curve. All of our testing was with the BIOS set to the Gaming option, and it was as quiet as to be expected, removing up to 360W of heat through the massive cooling structure. We can’t provide a decibel reading, but it was folding away about four feet from my ears while I wrote this article. I could hear the fans whirring, but the tone wasn’t hard on the ears. When fully ramped up, they were loud.
MSI offers several differnt RTX 5080 SKUs. From the top down is the Suprim Liquid SOC and Suprim. Then the Gaming Trio OC, Vanguard, You’ll recognize the Gaming Trio and Ventus 3X (three fan) SKUs, Vanguard, and the RGB-less Inspire 3X OC and Shadow 3X OC. Oh, and there’s even a Gaming Trio OC White. You’re too picky if you can’t find a card that suits you in that product stack.
There are a few RGB-illuminated areas on the card. On the front, two chevrons flank the middle fan, while the diamond design and Suprim branding all light up on top. RGB???? It’s a good-looking card with loads of brushed aluminum on the front and back.
Retail Packaging and Accessories
Retail packaging for the RTX 5080 Suprim SOC is a light grey box with crystalline/diamond-like shapes on the top. The Nvidia model number is at the bottom right, while the Suprim branding is reflected like a rainbow in the middle. Opening up the box and below a layer of padding, you’ll find the card and accessories securely tucked in the protective foam. Inside, you’ll find the latest 12VHPWR connector that uses three 8-pin PCIe connectors (Unique – Do not piggyback!) to feed the board. MSI also includes a small GPU support to hold up the massive, nearly 4-slot, card.
Meet the MSI RTX 5080 Suprim SOC

The card boasts a premium look with an aluminum faceplate and brushed aluminum finish. In between, you can spot the two RGB areas. The edges are plastic, as is the Suprim badging. The rear of the card sports a cool grated design where plenty of air can flow through the heatsink. The diamond design in the middle is a nice touch. The backplate is also made out of aluminum with a two-tone matte brushed aluminum finish.
A Closer Look…
On the I/O, we see Nvidia’s standard lineup of three DisplayPorts and one HDMI. With Blackwell comes a bump in spec with DP 2.1b and HDMI 2.1a. The former offers ~3x the bandwidth of 1.4a and supports up to 4K 480Hz (with DSC) or 8K 165Hz (DSC).
Power for this card goes through the updated (improved?) 12VHPWR connector (included), which uses three 8-pin PCIe power leads for the 12-pin configuration. Again, please use three unique PCIe cables with the connector, or buy an ATX 3,1/PCIe 5.1-spec’d power supply.
Once we break the card down, we’re greeted by a massive piece of engineering called a heatsink and a proportionately tiny PCB where all the magic happens. In the middle of the board is the GB203-400-A1 die with Samsung 16Gb, 32 Gbps GDDR7 ICs, a Monolithic Power Systems MP29826-A controller managing MPS MP87993 Dr MOS MOSFETs. I couldn’t find any data sheets on these, but I’m guessing 90A.
Below is a screenshot of GPU-Z showing the clocks our card ran with stock speeds. Our review card has a listed boost clock of 2,745 MHz, but ran just a little higher (~2,780 MHz+) throughout our testing, peaking at 2,820 MHz.

MSI RTX 5080 Suprim SOC on the test bed…

Test System and Benchmark Methods
Test System Components | |
Motherboard | Gigabyte Z790 Aorus Xtreme (F11b) |
CPU | Intel i9-14900K @ stock |
CPU Cooler | CoolerMaster Master Liquid PL360 Flux |
Memory | Kingston 2x16GB DDR5-6000 (36-38-38-80) |
SSD | 2x 1TB PCIe 4.0 x4 NVMe (OS + Applications) |
Power Supply | EVGA 850W P6 |
Video Card | @ Stock (572.02 driver as of 1/2025) |
Our test system is based on the latest (at the time of publishing) mainstream Intel z790 platform and uses the i9-14900K 8P,16E/32t CPU. We’ve updated it to the latest BIOS. The DRAM is in a 2×16 GB configuration at 6000 MHz with CL36-38-38-80 timings, a slower option by all accounts today, but still balances performance and cost. The CPU runs stock for the motherboard.
Since the last update, we have made some changes and updated titles. More details can be found in the most recent GPU Testing Procedure article, which we have updated with our latest benchmarks. Below is a quick summary for easy reference.
- UL 3DMark Time Spy – Default settings
- UL 3DMark Fire Strike (Extreme) – Default settings
- UL 3DMark Port Royal – Default Settings (Ray Tracing capable cards only)
- UL 3DMark Speedway – Defaul t settings
- Unigine Superposition – 4K Optimized
- Shadow of the Tomb Raider – DX12, “Highest” preset
- Assasin’s Creed: Mirage – Ultra High preset
- F1 2024 – Ultra High preset, Las Vegas track
- Far Cry 6 – Ultra preset, HD Textures enabled
- Avatar – Ultra preset
- CyberPunk 2077 – Ultra preset
Synthetic Benchmarks
Our first set of benchmarks hails from Underwriters Laboratories, which acquired Futuremark in 2014. Earlier in 2018, a rebrand occurred, and since that time, Futuremark is now UL. The benchmarks have not changed; it is just the name. We chose to use 3DMark Fire Strike (Extreme) and 3DMark Time Spy, 3DMark Speedway, and 3DMark Port Royal as these tests give users a good idea of performance on modern titles and include ray tracing.
3DMark Fire Strike (Extreme) is a DX11-based test that runs at 1080p resolution. UL says the graphics are rendered with detail and complexity far beyond other DX11 benchmarks and games. 3DMark Time Spy is a DX12 benchmark designed for Windows 10 PCs. It supports new API features such as asynchronous computing, explicit multi-adapter, and multi-threading and runs at 2560 x 1440. 3DMark Port Royal is the first Ray Tracing benchmark designed for Windows PCs and graphics cards with Microsoft DirectX Raytracing capabilities, while Speedway is another DX12 benchmark.
Results across our synthetic benchmarks put the card notably ahead of our RTX 4080. Across all titles, it averages ~15% ahead, with a low of 9% (Time Spy—the most CPU-dependent of these) and a high of 16% (Port Royal).
Gaming Benchmarks
We have updated our testing suite for gaming benchmarks to bring more modern titles into the mix. Gone are Assasin’s Creed: Odyssey, F1 2022, The Division 2, and Metro Exodus, replaced/updated by Assassin’s Creed: Mirage, F1 2023, Avatar, and CyberPunk 2077. It is worth mentioning that high-end cards aren’t made for 1080p gaming, so the gaps between them tend to be minimized as this is a wholly CPU-bound resolution with such powerful SKUs.
1920 x 1080 (1080p) Results
In 1080p gaming, our card pushed all titles to over 160 FPS+. It’s an overkill card for 1080p and is limited by our processor (a 9800X3D would yield better results, for example). Again we see the RTX 5080 beating our RTX 4080 by a significant margin. Across these titles, it is a 3.5% average in favor of the RTX 5080. We had an outlier where the RTX 4080 beat our card. If we take that result away, it’s over 5%. There is not a big difference in this CPU-bound resolution.
2560 x 1440 and 4K UHD Results
Below are the higher resolution results starting with 2560 x 1440 and the gaining-in-popularity 3840 x 2160 (4K UHD) resolution.


Starting with the middling resolution of 2560×1440, the average across all titles is 190 FPS, with peaks over 270 FPS (SOTR) and lows of 119 FPS in Avatar. If you play other, less stressful titles, they’ll easily approach and surpass that 240 FPS/Hz mark. Obviously, that is title and settings dependent.
At 4K UHD, the RTX 5080 chews through our gaming suite, averaging just over 120 FPS across all titles. Avatar and Cyberpunk kill the averages, as both are in the low 70s, but that’s still well over the 60 FPS threshold. With first-person shooters and esports games (think Counter-Strike, Valorant, Apex, Fortnite, PUBG, etc.), it’s a solid high-hertz option, even at this high resolution.
Ray Tracing and DLSS/FSR Testing
The charts below show what the review card can do when using ray tracing and DLSS/FSR capabilities. In this grouping, we test at 2560 x 1440 and 3840 x2160.
Far Cry 6
F1 2024
Assassin’s Creed: Mirage
Cyberpunk 2077
Shadow of the Tomb Raider
Avatar
Ray tracing still takes its toll on today’s graphics cards, often cutting the frame rates in half. DLSS (and AMD’s FSR) can help with this, and NV’s new DLSS 4 and now Multi-Frame Generation can take up that slack while offering similar image quality (sometimes as good or better, fewer times worse). While we didn’t have a chance to dive into those technologies in this review, we’ll cover things in more detail in a future article now that we have access to a few more titles.
Outside of Cyberpunk at 4K with Ultra-Raytracing, our RTX 5080 proved to be a capable ray-tracing card, pushing at least 60 FPS at 1440p and 4K UHD. Enable DLSS, and that number will jump considerably (though variably by title).
Folding @ Home Performance
For all of the folding @ home community, we had a chance to let this run for a full day (actually, a few). Below is a screenshot that shows us around 21.4M PPD. During this full 24 hours, the card averaged 241W with a temperature of ~60C. Not too bad!

Overclocking the MSI RTX 5080 Suprim SOC
For overclocking, we used the latest version of MSI Afterburner. Unlike previous OC methods, I went manual this time and landed on +71 for the core and +100 MHz on the RAM. Which left plenty of meat on the bone. I could push close to a 3,200 MHz core and the RAM over +300 before the driver got in the way. As you’d expect, the scores/FPS responded well at higher GPU-bound resolutions. Power peaked at ~315W (compared to 306W at stock – F1 24 1440p w/RT) as we did not touch the voltage during this endeavor (nor the power limit). If needed, this specific card can raise the power limit by 11% over its 360W board power.

Temperatures and Power Use
We test power consumption by running through the game benchmarks of F1 2024 and Cyberpunk 2077. We monitor temperatures throughout this testing, with the peak temperature listed in the data below. The benchmarks are extended (time) to allow the card to settle and, more accurately, simulate extended gaming conditions. Most air-cooled cards will saturate and normalize around 10 minutes or so.

Temperatures on this card reached 64°C in F1 2024 and 65°C CP 2077 at stock settings. Once overclocked, temperatures increased slightly in both, peaking two degrees Celsius higher in F1 2024, and two degrees in CP 2077. Longer gaming sessions will increase the temperatures, but this gives you a general idea of the card’s cooling capability. When folding for over 24 hours straight, it peaked at 63C (average of ~60C), so the thick cooler and fans do the job quietly in most instances.
Power use for the RTX 5080 Suprim SOC peaked at 305W (video card power only) in F1 2024 and 320W in CP 2077. Overclocking didn’t increase use much as we didn’t (manually) increase the voltage. A high-quality 850W power supply will be plenty to support this card and a flagship-class processor.
Conclusion
Let’s talk about Blackwell in general first. As most of you have already seen, It’s been an interesting launch for the 5000 series. While there are notable technological improvements with DLSS 4 and Multi-Frame Generation, rasterization didn’t get the bump that many enthusiasts wanted. There are a lot of ways to dissect this. As a gamer who doesn’t play a lot of ray traced titles, we certainly could’ve used a bigger jump. A mere ~10% and not reaching RTX 4090 levels is a disappointment for many.
However, those who use DLSS (Nvidia says over 80% do) will see improvements with DLSS 4, and it’s not only Blackwell users. It even trickles down to Ada, Ampere, and Turing-based cards that support ray tracing and DLSS (performance improvements vary depending on the card). Nvidia now uses a transformer-based AI model that improves image quality. Here’s a great article from Techpowerup, complete with image quality comparisons and performance, so you can see how the various modes compare and judge for yourself.
By now, we know the RTX 5080 Founder’s Edition is a well-built card with capable cooling, decent noise profile, and overclocks well. But board partners can do it bigger, I mean, better. Or, both. MSI’s RTX 5080 Suprim SOC takes up four slots but does so gracefully. The clean premium finishes and subtle lighting, make you want to show it off inside your chassis. But it not only looks good, the Hyper Frozrz cooler is quiet unless you crank the fans, which isn’t remotely necessary. You get a minor factory overclock, but a lot of meat is left on the bone for overclocking. I wish I had more time with the card before getting it out.
Priced at $1,250, it’s towards the middle of aftermarket cards, with the Asus Astral sitting at a mind-blowing $1,500 and others as low as $1,200. It’s still a $250 premium over the Founder’s Edition, but not that you can find them for that price after the limited supply sells out almost immediately. If you can score one, you won’t be disappointed. To that, we’re awarding the card with an “approved” stamp on the merits of what MSI did and not by the ghost of significant rasterization increase in the past. Pricing is right among its peers, and it’s one that oozes the most premium vibes of all the partner cards. Just make sure you have room for it inside your chassis!

– Joe Shields (Earthdog)
- MSI RTX 4070 Gaming X Trio
- MSI RTX 4070 Ti Suprim X Review
- MSI RTX 4080 Suprim X Review
- MSI RTX 4090 Suprim Liquid X Review
- MSI RTX 3090 Ti Review
- MSI RTX 3080 Ti Suprim X Review
- XFX MERC319 BLACK RX 6700 XT Review
- MSI RTX 3060 Ti Gaming X Trio Review: Best Value 4k/60fps Gaming
- AMD RX 6800 XT and RX 6800 Review: Competing at the Top Again
- ASUS ROG Strix RTX 3080 OC Review
- MSI RTX 3090 Gaming X Trio Review
16 replies
Loading new replies...
Senior Golfer
Member
Member
Member
Senior Golfer
Member
The Printer Guru
Member
Join the full discussion at the Overclockers Forums →