MSI GeForce RTX 4080 Suprim X review: piling on the pounds

Jump to: Architecture | Performance | Conclusion

Notice how the launch of GeForce RTX 4080 feels different to most other recent releases? The GPU’s blistering performance doesn’t come as too much of a surprise – it is, after all, a chip off the incredible RTX 4090 block – and pricing, while lofty, aligns with what most folk may have expected feared. The distinction is that RTX 4080 hasn’t sold out.

On the contrary, while the Founders Edition may be hard to come by, partner cards are in plentiful supply at all the major retailers. Predicting stock levels in this volatile age is a fool’s errand. Crypto-fuelled demand has fallen off a cliff, chip shortages are abating, and with folk tightening their belts amid rampant inflation, the opportunities for unscrupulous scalpers are few and far between. That could all be true, but maybe Nvidia simply aimed too high.

£1,269 for an x80 Series GPU hasn’t sat well with the masses, and remember, this isn’t a full implementation of Nvidia’s second-rung AD103 chip. Premium positioning ultimately puts partners in a quandary. Undercutting the Founders Edition isn’t feasible, and the very best OC models attract fees that are hard to fathom.

MSI is no stranger to the best graphics card launches and meets RTX 4080 head-on with half-a-dozen day-one variants. Entry-level Ventus starts at £1,350, mid-range Gaming X Trio fetches £1,420, and top-end Suprim X lands at £1,500. Familiar segmentation, and it is customary for flagship models to attach such a premium, yet £1,500 for an RTX 4080 is bordering on madness and far too close to flagship RTX 4090 (£1,679) for comfort.

A shame, really, as of all the partner cards we’ve handled, Suprim X comes closest to matching the prestige of Nvidia’s Founders Edition. The silver-and-black colour scheme works really well and lends itself to almost any style of build, the brushed metal fascia and octagon-shaped fan cutouts offer something a little different, and whereas Gaming X Trio ships with a horizontal support bracket that’s a pain to install, Suprim X is bundled with a metal, vertical stand that’s quick to setup and easy on the eye.

There’s also a token mouse mat included, alongside the customary three-way, eight-pin-to-12VHPWR power adapter. We’ve experienced zero issues with the ATX 3.0 connector on all 40 Series cards tested to date, but concerns over reliability remain, and all users are encouraged to check cables are fully inserted with minimal bending.

As you might expect, MSI’s prettiest card also features the Taiwanese firm’s best cooling setup. Dubbed a Tri-Forzr 3S thermal design, we get Torx Fan 5.0 blowers promising a 23 per cent increase in airflow, a giant vapour chamber encompassing both GPU and VRAM, a vast full-length heatsink, dual BIOS, and metal backplate lined with thermal pads.

It is a monster of a unit measuring 336mm x 142mm x 78mm, and a weight of 2,364g makes it one of the heaviest graphics cards we’ve ever tested. Though built like a tank, Suprim X is tuned to operate like a stealth bomber. The three fans switch off at low load, spin-up gently when temperature exceeds 60°C, and never surpass 35 per cent (1,000RPM) throughout testing.

Such cooling lends itself to a factory overclock of 2,625MHz (up from 2,505MHz) while the 16GB of GDDR6X memory hums along at a default 22.4Gbps. MSI’s card routinely ran at just over 2.8GHz inside our real-world test platform, and can go higher if you choose to enable the optional Extreme Performance (2,640MHz) turbo clock through the MSI Center companion app.

Looks rather handsome in situ, wouldn’t you say? MSI’s stylish lighting is backed by enough LEDs to create seamless transitions, but you will need to install the aforementioned software for control and customisation; there’s no aRGB header for motherboard synchronisation. Benchmarks await, but first, a recap on everything you need to know about Nvidia’s Ada Lovelace architecture.

A Chip Off The Old Block

Those of you craving benchmark numbers can skip right ahead – they’re worth the wait – but for the geeks among us, let’s begin with all that’s happening beneath RTX 40 Series’ hood.

What you’re looking at is one of the most complex consumer GPUs to date. Nvidia’s 608mm² die, fabricated on a TSMC 4N process, packs a scarcely believable 76.3 billion transistors. Putting that figure into perspective, the best GeForce chip of the previous 8nm generation, GA102, measures 628mm² yet accommodates a puny 28.3 billion transistors. Nvidia has effectively jumped two nodes in one fell swoop, as Samsung’s 8nm process is more akin to 10nm from other foundry manufacturers.

We’ve gone from flyweight to heavyweight in the space of a generation, and a 170 per cent increase in transistor count naturally bodes well for specs. A full-fat die is home to 12 graphics processing clusters (GPCs), each sporting a dozen streaming multiprocessors (SMs), six texture processing clusters (TPCs) and 16 render output units (ROPs). Getting into the nitty-gritty of the block diagram, each SM carries 128 CUDA cores, four Tensor cores, one RT core and four texture units.

All told, Ada presents a staggering 18,432 CUDA cores in truest form, representing a greater than 70 per cent hike over last-gen champion, RTX 3090 Ti. Plenty of promise, yet to the frustration of performance purists, Nvidia chooses not to unleash the full might of Ada in the first wave. The initial RTX 40 Series duo shapes up as follows:

GeForce	RTX 4090	RTX 4080	RTX 3090 Ti	RTX 3080 Ti	RTX 3080 12GB
Launch date	Oct 2022	Nov 2022	Mar 2022	Jun 2021	Jan 2022
Codename	AD102	AD103	GA102	GA102	GA102
Architecture	Ada Lovelace	Ada Lovelace	Ampere	Ampere	Ampere
Process (nm)	4	4	8	8	8
Transistors (bn)	76.3	45.9	28.3	28.3	28.3
Die size (mm2)	608.5	378.6	628.4	628.4	628.4
SMs	128 of 144	76 of 80	84 of 84	80 of 84	70 of 84
CUDA cores	16,384	9,728	10,752	10,240	8,960
Boost clock (MHz)	2,520	2,505	1,860	1,665	1,710
Peak FP32 TFLOPS	82.6	48.7	40.0	34.1	30.6
RT cores	128	76	84	80	70
RT TFLOPS	191.0	112.7	78.1	66.6	59.9
Tensor cores	512	304	336	320	280
ROPs	176	112	112	112	96
Texture units	512	304	336	320	280
Memory size (GB)	24	16	24	12	12
Memory type	GDDR6X	GDDR6X	GDDR6X	GDDR6X	GDDR6X
Memory bus (bits)	384	256	384	384	384
Memory clock (Gbps)	21	22.4	21	19	19
Bandwidth (GB/s)	1,008	717	1,008	912	912
L1 cache (MB)	16	9.5	10.5	10	8.8
L2 cache (MB)	72	64	6	6	6
Power (watts)	450	320	450	350	350
Launch MSRP ($)	1,599	1,199	1,999	1,199	799

Leaving scope for a fabled RTX 4090 Ti, inaugural RTX 4090 disables a single GPC, enabling 128 of 144 possible SMs. Resulting figures of 16,384 CUDA cores, 128 RT cores and 512 Tensor cores remain mighty by comparison, and frequency headroom on the 4nm process is hugely impressive, with Nvidia specifying a 2.5GHz boost clock. The madness of Nvidia’s flagship is reflected in peak teraflops, which more than doubles from 40 on RTX 3090 Ti to an incredulous 82.6 on RTX 4090.

Front-end RTX 40 Series specifications are eye-opening, but the back end is noticeably less revolutionary, where a familiar 24GB of GDDR6X memory operates at 21Gbps, providing 1,008GB/s of bandwidth. The needle hasn’t moved – Nvidia has opted against Micron’s quicker 24Gbps chips – however the load on memory has softened with a significant bump in on-chip cache.

RTX 4090 carries 16MB of L1 and 72MB of L2. We’ve previously seen AMD attach as much as 128MB Infinity Cache on Radeon graphics cards, and though Nvidia doesn’t detail data rates or clock cycles, a greater than 5x increase in cache between generations reduces the need to routinely spool out to memory, raising performance and reducing latency.

Heard rumours of RTX 40 Series requiring a nuclear reactor to function? Such reports were wide of the mark. RTX 4090 maintains the same 450W TGP as RTX 3090 Ti, while RTX 4080 scales down to 320W. Almost 10 per cent lower than the last-generation x80 equivalent.

Shifting focus to RTX 4080, it is instructive to know Nvidia is not repurposing the same die across multiple high-end SKUs, as was the case with the previous generation. Instead, RTX 4080 calls upon newfound AD103 silicon but once again falls short of full implementation, featuring 76 of 80 possible active SMs.

Core count is cleaved by over 40 per cent from 16,384 on RTX 4090 to 9,728 on RTX 4080. RT and Tensor cores are slashed proportionally to 76 and 304, respectively, while onboard GDDR6X memory not only drops from 24GB to 16GB, bus width is also narrowed from 384 bits to just 256. Savage cuts on paper, and certain specifications intimate a sideward step alongside existing parts. You’ll have noticed RTX 4080 has fewer CUDA, RT and Tensor cores than RTX 3080 Ti, so what gives?

Nvidia has kept the door ajar for a potential RTX 4080 Ti with a full complement of 80 active SMs

In addition to a treasure chest of Ada Lovelace architecture optimisations, the 4nm process enables significantly higher clocks – RTX 4080 is binned to operate at effectively the same 2.5GHz as RTX 4090 – and such lofty frequencies ensure peak teraflops best the previous-generation champion, RTX 3090 Ti, to the tune of 22 per cent. Memory clock is also upped from 21Gbps on RTX 4090 to 22.4Gbps on RTX 4080, to help mitigate the tighter bus, and let’s not forget the 64MB of L2 cache, representing a generational increase of over 10x.

Leaving plenty of room to manoeuvre, Nvidia has kept the door ajar for a potential RTX 4080 Ti with a full complement of 80 active SMs, and as for the ill-fated RTX 4080 12GB, well, the less said the better. Word on the grapevine is the third-rung $899 GPU will eventually see the light of day under new guise as RTX 4070 Ti in 2023.

Outside of core specifications, display outputs remain tied to HDMI 2.1 and DisplayPort 1.4a – DisplayPort 2.0 hasn’t made the cut – and PCIe Gen 4 continues as the preferred interface. There’s no urgency to switch to Gen 5, says Nvidia, as even RTX 4090 can’t saturate the older standard. Finally, NVLink is conspicuous by its absence; SLI is nowhere to be seen on any RTX 40 Series product announced thus far, signalling multi-GPU setups are well and truly dead.

Ada Optimisations

While a shift to a smaller node affords more transistor firepower, such a move typically precludes sweeping changes to architecture. Optimisations and resourcefulness are order of the day, and the huge computational demands of raytracing are such that raw horsepower derived from a 3x increase in transistor budget isn’t enough; something else is needed, and Ada Lovelace brings a few neat tricks to the table.

Nvidia often refers to raytracing as a relatively new technology, stressing that good ol’ fashioned rasterisation has been through wave after wave of optimisation, and such refinement is actively being engineered for RT and Tensor cores. There’s plenty of opportunity where low-hanging fruit is yet to be picked.

Shader Execution Reordering

Shaders have been running efficiently for years, whereby one instruction is executed in parallel across multiple threads. You may know it as SIMT.

Raytracing, however, throws a spanner in those smooth works, as while pixels in a rasterised triangle lend themselves to running concurrently, keeping all lanes occupied, secondary rays are divergent by nature and the scattergun approach of hitting different areas of a scene leads to massive inefficiency through idle lanes.

Ada’s fix, dubbed Shader Execution Reordering (SER), is a new stage in the raytracing pipeline tasked with scanning individual rays on the fly and grouping them together. The result, going by Nvidia’s internal numbers, is a 2x improvement in raytracing performance in scenes with high levels of divergence.

Nvidia portentously claims SER is “as big an innovation as out-of-order execution was for CPUs.” A bold statement and there is a proviso in that Shader Execution Reordering is an extension of Microsoft’s DXR APIs, meaning it us up to developers to implement and optimise SER in games.

There’s no harm in having the tools, mind, and Nvidia is quickly discovering that what works for rasterisation can evidently be made to work for RT.

Upgraded RT Cores

In the rasterisation world, geometry bottlenecks are alleviated through mesh shaders. In a similar vein, displaced micro-meshes aim to echo such improvements in raytracing.

“The era of brute-force graphics rendering is over”

Bryan Catanzaro, Nvidia VP of applied deep learning research

With Ampere, the Bounding Volume Hierarchy (BVH) was forced to contain every single triangle in the scene, ready for the RT core to sample. Ada, in contrast, can evaluate meshes within the RT core, identifying a base triangle prior to tessellation in an effort to drastically reduce storage requirements.

A smaller, compressed BVH has the potential to enable greater detail in raytraced scenes with less impact on memory. Having to insert only the base triangles, BVH build times are improved by an order of magnitude and data sizes shrink significantly, helping reduce CPU overhead.

The sheer complexity of raytracing is such that eliminating unnecessary shader work has never been more important. To that end, an Opacity Micromap Engine has also been added to Ada’s RT core to reduce the amount of information going back and forth to shaders.

In the common leaf example, developers place the texture of foliage within a rectangle and use opaque polygons to determine the leaf’s position. A way to construct entire trees efficiently, yet with Ampere the RT core lacked this basic ability, with all rays passed back to the shader to determine which areas are opaque, transparent, or unknown. Ada’s Opacity Micromap Engine can identify all the opaque and transparent polygons without invoking any shader code, resulting in 2x faster alpha traversal performance in certain applications.

These two new hardware units make the third-generation RT core more capable than ever before – TFLOPS per RT core has risen by ~65 per cent between generations – yet all this isn’t enough to back up Nvidia’s claims of Ada Lovelace delivering up to 4x the performance of the previous generation. For that, Team Green continues to rely on AI.

DLSS 3

Since 2019, Deep Learning Super Sampling has played a vital role in GeForce GPU development. Nvidia’s commitment to the tech is best expressed by Bryan Catanzaro, VP of applied deep learning research, who states with no uncertainty that “the era of brute-force graphics rendering is over.”

Third-generation DLSS, deemed a “total revolution in neural graphics,” expands upon DLSS Super Resolution’s AI-trained upscaling by employing optical flow estimation to generate entire frames. Through a combination of DLSS Super Resolution and DLSS Frame Generation, Nvidia reckons DLSS 3 can now reconstruct seven-eighths of a game’s total displayed pixels, to dramatically increase performance and smoothness.

Generating so much on-screen content without invoking the shader pipeline would have been unthinkable just a few years ago. It is a remarkable change of direction, but those magic extra frames aren’t conjured from thin air. DLSS 3 takes four inputs – two sequential in-game frames, an optical flow field and engine data such as motion vectors and depth buffers – to create and insert synthesised frames between working frames.

In order to capture the required information, Ada’s Optical Flow Accelerator is capable of up to 300 TeraOPS (TOPS) of optical-flow work, and that 2x speed increase over Ampere is viewed as vital in generating accurate frames without artifacts.

The real-world benefit of AI-generated frames is most keenly felt in games that are CPU bound, where DLSS Super Resolution can typically do little to help. Nvidia’s preferred example is Microsoft Flight Simulator, whose vast draw distances inevitably force a CPU bottleneck. Internal data suggests DLSS 3 Frame Generation can boost Flight Sim performance by as much as 2x.

Do note, also, that Frame Generation and Super Resolution can be implemented independently by developers. In an ideal world, gamers will have the choice of turning the former on/off, while being able to adjust the latter via a choice of quality settings.

More demanding AI workloads naturally warrant faster Tensor Cores, and Ada obliges by imbuing the FP8 Transformer Engine from HPC-optimised Hopper. Peak FP16 Tensor teraflops performance is already doubled from 320 on Ampere to 661 on Ada, but with added support for FP8, RTX 4090 can deliver a theoretical 1.3 petaflops of Tensor processing.

Plenty of bombast, yet won’t such processing result in an unwanted hike in latency? Such concerns are genuine; Nvidia has taken the decision to make Reflex a mandatory requirement for DLSS 3 implementation.

Designed to bypass the traditional render queue, Reflex synchronises CPU and GPU workloads for optimal responsiveness and up to a 2x reduction in latency. Ada optimisations and in particular Reflex are key in keeping DLSS 3 latency down to DLSS 2 levels, but as with so much that is DLSS related, success is predicated on the assumption developers will jump through the relevant hoops. In this case, Reflex markers must be added to code, allowing the game engine to feed back the data required to coordinate both CPU and GPU.

Given the often-sketchy state of PC game development, gamers are right to be cautious when the onus is placed in the hands of devs, and there is another caveat in that DLSS tech is becoming increasingly fragmented between generations.

DLSS 3 now represents a superset of three core technologies: Frame Generation (exclusive to RTX 40 Series), Super Resolution (RTX 20/30/40 Series), and Reflex (any GeForce GPU since the 900 Series). Nvidia has no immediate plans to backport Frame Generation to slower Ampere cards.

8^th Generation NVENC

Last but not least, Ada Lovelace is wise not to overlook the soaring popularity of game streaming, both during and after the pandemic.

Building upon Ampere’s support for AV1 decoding, Ada adds hardware encoding, improving H.264 efficiency to the tune of 40 per cent. This, says Nvidia, allows streamers to bump their stream resolution to 1440p while maintaining the same bitrate.

AV1 support also bodes well for professional apps – DaVinci Resolve is among the first to announce compatibility – and Nvidia extends this potential on high-end RTX 40 Series GPUs by ensuring both launch models include dual 8^th Gen NVENC encoders (enabling 8K60 capture and 2x quicker exports) as well as a 5^th Gen NVDEC decoder as standard.

Performance

Our trusty test platforms have been working overtime these past few months, and though the PCIe slot is starting to look worse for wear, the AM4 rigs haven’t skipped a beat. RTX 4080 locked and loaded.

Appreciating that RTX 4080 is limited with regards to overclocking headroom, partner cards aren’t going to alter performance by any meaningful degree. A fraction quicker than FE is the order of the day.

Massive, heavyweight coolers do bode well for consistency. 99.4 per cent is an outstanding score.

RTX 4080 is quick enough to beat anything from the previous generation, but results also reinforce that RTX 4090 is on another level entirely.

Assassin’s Creed Valhalla

Almost 25 per cent quicker than RTX 3090 Ti at a 4K UHD resolution, RTX 4080 sets a high bar for a second-rung GPU.

Cyberpunk 2077

The choice between Nvidia GeForce and AMD Radeon becomes lopsided when raytraced games are added to the equation. Radeons prop up the bottom of the Cyberpunk charts across all three resolutions, and RTX 4080 is over twice as quick as Radeon RX 6950 XT in this title.

Far Cry 6

Already factored an RTX 4080 into your build plans? Consider a high-speed, high-res display to go with. 4K 144Hz ought to do it.

Final Fantasy XIV: Endwalker

RTX 4080 pricing absolutely needs to stay well clear of RTX 4090. Nvidia’s top dog barks loudest.

Forza Horizon 5

4K120? That’ll do nicely, thank you very much.

Marvel’s Guardians of the Galaxy

4K60 in Marvel’s Guardians of the Galaxy with ultra raytracing settings ought to be the benchmark for any next-gen GPU. RTX 4080 exceeds that target. No pressure, Radeon RX 7900 XTX.

Tom Clancy’s Rainbow Six Extraction

RTX 4080 tears up everything from the previous generation with ease, and RTX 4090 beats RTX 4080 by a further 38 per cent. Staggering.

Power, Temps and Noise

Any partner card choosing to ship with a factory overclock will naturally draw a little extra power from the wall. MSI plays it relatively conversative, resulting in system-wide power consumption increasing by four per cent.

Cooling Nvidia’s 320W GPU is child’s play for cards of this magnitude.

Effortless cooling translates to ultra-low noise levels. Our trusty decibel meter reckons the MSI is fractionally more vocal than the Founders Edition, but to our ear there’s nothing in it. Both are extremely quiet even when gaming.

Overclocking

RTX 4080 doesn’t have a great deal of headroom at stock voltages, and factory overclocked cards are already running close to the ceiling. It’s easy to push memory to over 23Gbps, but in truth, you’re not going to see much more performance using automated scanners.

Basic overclocking is hardly worth the effort. DLSS, on the other hand, is well worth pursuing.

DLSS 3

Having spent the past few decades evaluating GPUs in familiar fashion, it is eye-opening to witness a sea change in how graphics performance is both delivered and appraised.

Thunderous rasterisation and lightning RT Cores, yet DLSS 3 is the storm that’s brewing

RTX 40 Series may boast thunderous rasterisation and lightning RT Cores, yet DLSS 3 is the storm that’s brewing. Rejogging the memory banks after all those benchmarks, remember developers now have the option to enable individual controls for Super Resolution and/or Frame Generation. The former, as you’re no doubt aware, upscales image quality based on setting; Ultra Performance works its magic on a 1280×720 render, Performance on 1920×1080, Balanced works at 2227×1253, and Quality upscales from 2560×1440.

On top of that, Frame Generation inserts a synthesised frame between two rendered, resulting in multiple configuration options. Want the absolute maximum framerate? Switch Super Resolution to Ultra Performance and Frame Generation On, leading to a 720p upscale from which whole frames are also synthesised.

Want to avoid any upscaling but willing to live with frames generated from full-resolution renders? Then turn Super Resolution off and Frame Generation On. Note that enabling the latter automatically invokes Reflex; the latency-reducing tech is mandatory, once again reaffirming the fact that additional processing risks performance in other areas.

Reviewers have been granted access to pre-release versions of select titles incorporating DLSS 3 tech. Initial impressions are that developers are still getting to grips with implementation. Certain games require settings to be enabled in a particular sequence for DLSS 3 to properly enable, while others simply crash when alt-tabbing back to desktop. There are other limitations, too. DLSS 3 is not currently compatible with V-Sync (FreeSync and G-Sync are fine) and Frame Generation only works with DX12.

Point is, while Nvidia reckons uptake is strong – some 35 DLSS 3 games have already been announced – these are early days, and some implementations will work better than others. Microsoft Flight Simulator is one of the first to officially support DLSS 3 as part of Sim Update XI (SU11, pictured above), and the results are dramatic to say the least.

Performance is benchmarked at a 4K UHD resolution with Ultra rendering quality. With DLSS disabled entirely, RTX 4080 is able to return an average of 66 frames per second in the Sydney Famous Landing Challenge. The game’s tendency to be CPU bound is reflected by small framerate boosts with Super Resolution (DLSS 2) set to either quality, balanced or performance. There’s little extra mileage to be had.

Frame Generation (DLSS 3) has a far more profound impact. Turning it on boosts performance by a whopping 55 per cent, climbing to 94 per cent when Super Resolution is added to the mix at maximum quality.

Flight Simulator is an excellent example of where Frame Generation holds merit. This is a tech unlikely to be widely used by competitive gamers who rely on maximum accuracy, but in the right scenario, the framerate gains are simply too large to ignore. With the right monitor, the difference between 4K60 and 4K120 can be profound, and when done well there’s little detriment to image quality.

That isn’t to say it’s perfect – we did notice occasional flickering of elements in the cockpit – but you really have to look for artifacts to know they’re there. Implementation is impressive, even at this early stage, and Nvidia does well to keep latency in check through Reflex. With DLSS disabled entirely, in-game Flight Simulator latency is recorded as 54.2ms. This rises to 64.5ms with Frame Generation enabled, and then drops back down to 54.9ms with Frame Generation and Super Resolution at maximum quality.

It is too soon to deem DLSS 3 a must-have feature, yet make no mistake, it is an important feather in the 40 Series’ cap and one that rivals will soon attempt to replicate. Next-generation AMD Radeon will employ Fluid Motion Frames in an effort to double 4K framerate, but when exactly the tech will be available, and how well it performs, remains unknown.

Conclusion

Following up on the blockbuster performance of GeForce RTX 4090, Nvidia extends its assault on the high-end with GeForce RTX 4080. Starting at £1,269, the new addition cements RTX 40 Series’ positioning as the preserve of deep-pocketed enthusiasts, and even when a third GPU reappears, don’t expect much change from a thousand pounds for an alleged RTX 4070 Ti.

Founders Editions lead the charge, and with the in-house design evolving dramatically in terms of size and performance, add-in-board partners are granted little room to manoeuvre. Producing cooler and quieter cards is no longer a given, and the focus instead shifts to factory-overclocked variants that stand out through unique aesthetics.

MSI’s best effort takes all the essentials and wraps them in an ultra-refined package that looks the part, is built like a tank, and runs quicker than reference. The caveat, and it’s a big one, is that pricing balloons to £1,500, and a near-20 per cent premium puts an already expensive GPU beyond the reach of the gamers RTX 4080 ought to be targeting.

Suprim X is stuck between a rock (RTX 4080 FE) and a hard place (RTX 4090 FE), and we wonder what it’ll take for Nvidia to budge on pricing. Here’s looking at you, Radeon RX 7900 XTX.