In this article, we’re comparing the two ‘real world’ flagships of the AMD and Nvidia GPU line ups: The RTX 3080 and the RX 6800 XT.
It’s only right that we start with an explanation both of the delay to this review, but also around the prevailing situation with GPUs.
We’re not party to normal press channels to obtain cards so we’ve been subject to all the same pressures as the general market. We’re therefore only too painfully aware of the scarcity of GPUs across the board right now, and it’s part of the reason for the delay to this comparison article.
Of course, the global pandemic is one of the driving factors behind both restricted supply and also heightened demand for GPUs. On the one hand, production lines have been hit by a supply shortage and global supply chains have been disrupted. On the other hand, demand has skyrocketed with more people working from home or looking for means to entertain themselves. AMD in particular is struggling to produce sufficient GPUs on the 7nm production lines alongside the console APUs and Zen 3 CPUs that all vie for the limited capacity at TSMCs’ foundries. And finally of course Cryptocurrency mining has again come to the fore with miners buying up ‘gaming’ GPU’s in bulk to mine Ethereum. We all know the results of this: The cards we’re reviewing here sell via side channels for two to three times their launch MSRP, and are near impossible to buy via normal retail channels.
We’re going to keep this review straight down the line: You might be looking for one or other card and be prepared to pay. You might want to know if you can reasonably substitute one for the other and not suffer any major performance loss. This review simply seeks to answer the question: ‘how do these cards perform’, because it’s impossible to answer the question ‘are these cards worth your money’ until such time as the market stabilizes and they become available anywhere close to their suggested retail pricing.
First, let’s remind ourselves of the key specifications of these cards:
RX 6800 XT vs RTX 3080: Specifications
| AMD RX 6800 XT | Nvidia RTX 3080 | |
|---|---|---|
| Design |  |  |
| GPU Architecture | AMD RDNA 2 | Nvidia Ampere |
| Process | TSMC 7nm | Samsung ‘8nm’ |
| Transistors | 26,800 million | 28,300 million |
| VRAM | 16Gb GDDR6 | 10GB GDDR6X |
| Memory Bus | 256Bit 512GB/s | 320bit 760 GB/s |
| Ray Tracing Cores | 72 – RDNA2 | 68 – Ampere ‘2nd Gen RTX’ |
| Tensor Cores | 272 | |
| ‘Enhanced Memory Access’ | Smart Access Memory | Resizeable BAR |
| PCIe Version | 4 | 4 |
| Power Draw (tested) | 300W | 320W (340W) |
| Availability | Amazon.com | Amazon.com |
It’s not possible to make direct comparisons between clock speeds or the number of shader units for these two GPUs – their architectural differences make the comparison meaningless.
We can directly compare memory specification and the RX 6800 XT has a massive 16GB VRAM, but it is just standard GDDR6. The RTX 3080 has 10GB of VRAM but it’s higher performance GDDR6X, produced by Micron under an exclusive partnership. This combined with a 320bit bus gives the Nvidia card higher memory performance but less capacity. In respect of additional features, Nvidia is obviously a step ahead in hardware ray tracing, but the RX 6800XT does also have hardware ray tracing cores and few games currently utilize this feature. Both have a mechanism to enhance memory access and boost performance – Smart Access Memory for AMD and ‘Resizeable BAR’ for Nvidia, but again this is specific to certain titles, and in the case of the RX 6800XT it can be detrimental to performance in some games. Finally, both have similar power draw requirements although the RTX 3080 is slightly more demanding, hitting 340W under peak load or when overclocked.
The test set-up:
Both GPUs were tested in the same system – our Ryzen Test bench which comprises a Ryzen 7 5800X CPU, with PBO enabled but no additional overclocking. It runs 16GB of 3600MHz CL16 RAM and has a 1Tb Sabrent Rocket PCIe 4.0 NVMe SSD. It is powered by a Fractal Design Ion 860W Platinum power supply to ensure sufficient power. SAM and ReBAR were not enabled for any testing.
Synthetic benchmarks: 3D Mark
Running through the 3D Mark suite of benchmarks, we can see that in both of the straight ‘gaming relevant’ tests the RX 6800XT has the upper hand.
In Fire Strike which is a 1080p rendered DirectX11 based test, the RX 6800XT turns in an absolutely stellar performance with a score of nearly 53,000 – that’s over 10,000 points or 20% ahead of the RTX 3080, an impressive achievement indicating very high performance in titles using DX11 and running at 1080p.
Time Spy, which uses the more recent DirectX12 API and renders at 1440p also shows a slight advantage to the RX 6800XT although the margin is reduced here to 1000 point or about 5%.
Port Royal is a new benchmark that uses DirextX12 Ultimate to bring the GPUs hardware ray tracing to bear, and both cards run this benchmark fine but the RTX 3080 with its’ second-generation RTX cores has clearly better performance, by about 1,800 points or 20%. I’ve included the score of the RTX 2080ti here as well for reference, showing how closely the RX 6800XT matches the performance of Nvidia’s last-gen Ray tracing tech – although this score is an aggregate of rasterized and ray tracing performance and without a score break down it’s impossible to ascertain how much of that is the core rendering performance of the RX 6800XT and how much is the ray-tracing component. Nonetheless, it’s not fair to say the RX 6800XT is ‘poor’ at ray tracing, simply that it can only use DirectX12 Ultimate, and that it’s not at Amperes performance levels.
Blender
As a quick test of the relative rendering performance, we ran Blender 2.83s benchmark renders, BMW27 and classroom. Blender has a range of rendering options depending on the GPU used, and we tested cards in all compatible modes to get an idea of how they perform. The RTX 3080 can use CUDA or ‘OptiX’ acceleration, whilst the RX 6800XT can only use Open CL. The results aren’t favorable for the RX 6800 XT: The BMW render takes it 39 seconds to the 3080’s 11 seconds, whilst the classroom scene takes 74 seconds to the RTX 3080’s 41. The OPTIX rendering pipeline uses tensor cores to accelerate the process, and it’s dramatically faster. Even an RTX 2060 KO matches the RX 6800 XT in BMW27 using the OPTIX renderer vs Open CL, although the RX 6800XT is much faster at the Classroom render. This highlights how specific these benchmarks can be to one or other aspects of a GPUs performance.
Overall, if you’re intending on using these cards for productivity or 3D rendering, it does pay to look closely at performance in metrics specific to your needs and software. It may be a more readily available or cheaper Nvidia card that can serve the purpose as well as the RX 6800XT – as an example, the RTX 3060 combines an Ampere core with 12GB VRAM and would seem an excellent choice for a 3D content creator – if you can find one.
Games: First Person Shooters
We’ve grouped the results here in a per-game basis to simplify the comparisons, and to make it easy to focus in on the games or game types that interest you.
1. Call of Duty: Warzone
Firstly, looking at Call of Duty Warzone we can see that the two cards are relatively evenly matched: We run a full 5-6 minute Battle Royale, but against bots as this is the only way to make the benchmark a consistent and repeatable test. Settings are at the ‘High’ preset. The RX 6800XT has a very slight advantage at 1080p, but just 10 FPS at 250FPS isn’t a noticeable difference. At 1440p the two GPUs are evenly matched, and at 1440p Ultrawide they’re absolutely equal. At 4K the 3080 has a slight advantage, ahead 9 FPS at 128 FPS. We haven’t included other metrics to keep the chart legible, but the Minimum, Maximum, and 1%, and 0.1% lows are almost identical too. CoD Warzone doesn’t appear to have any particular preference and you’ll have an equivalent experience on either card. At 1080p you would be better looking at a lower-tier card and achieving nearly as high performance – an RTX 2060Ti achieves 220FPS at 1080p high settings, for example.
2. Rainbow 6 Siege
Rainbow 6 Siege is a much faster-paced title and it shows in the benchmarks. Here, the RX 6800 XT takes a commanding lead at both 1080p and 1440p – it’s around 10% faster at 1080p and nearly 20% faster at 1440p. Initially, we felt this might be down to the rendering pipelines of the RTX 3080 being optimized for higher resolutions, but Hardware Unboxed and others have found interesting discrepancies in the Ampere cards performance when CPU limited: It appears their drivers have a higher overhead and as such it can dent the performance potential of the GPU. This game (and Flight Sim 2020) presents the most CPU-limited test we run, and as such, it may be one or both of these effects preventing the RTX 3080 from performing. It might also go some way to explaining that stellar 3DMark Fire Strike result at the start of this review. Again, at 1440p ultrawide resolution, we see that performance is about equal, and at 4K the RTX 3080 has taken the lead with 304Fps to 239 for the RX 6800 XT.
3. Doom Eternal
Finally, Doom Eternal is a well-optimized title using the Vulkan API and scales very well with both hardware and settings. We see a slight lead for the RX 6800 XT at 1080p at 409 FPS to 384, but at all other resolutions, the two GPUs perform identically on average. However looking at the underlying statistics, 1% and 0.1% lows are significantly lower for the RX 6800 XT indicating a less consistent experience and this occurs at every resolution. This isn’t something we see in the other First-person shooters where the results for these metrics are closely matched.
AAA Titles
Looking at the more demanding titles, this is where we expect to see both of these high end GPUs really shine.
Red Dead Redemption
Red Dead Redemption still ranks amongst the best looking and most demanding titles and again it uses the Vulkan API. Here we’ve run it at very high settings, near ultra, but with a couple of performance tweaks such as water being reduced to medium for a representative experience. Across the board these two GPUs perform near identically, exceeding 100 FPS at 1080p, close to 100FPS at 1440p, and exceeding 60FPS in 4K. Both provide a top tier experience, which is what you’d hope for flagship cards.
Shadow of the Tomb Raider
Shadow of the Tomb Raider always impresses us not only with its graphics but with its consistency when benchmarking – it reliably highlights very minor changes in system configuration or performance. At highest settings the numbers generated between these two cards are spookily similar, and not just the average frame rates, but the rest of the metrics as well. The experience on either card is identical, with one caveat: You can’t enable RTX Shadows or DLSS on the RX 6800 XT. However, in this title and with these GPUs this is a moot point as you don’t need either of these technologies for either visual improvements or performance. It’s a dead heat in this title.
Microsoft Flight Simulator 2020
Finally we’ll look at Microsoft Flight Simulator 2020.
This title continues to receive updates that fix bugs and improve performance, and it’s certainly a demanding test of a whole system. Our benchmark run is a four-minute AI piloted flight from LaGuardia over Manhattan at 2,500 – 3,500 ft – a worst-case scenario and highly demanding of the CPU as well as the GPU.
It’s the CPU that limits the ultimate frame rate here, and even at low settings, 65FPS average is the absolute best performance that can be obtained. At more realistic and rewarding high settings, we can see how little the FPS scale with such powerful GPUs; at 1080p, 1440p and even 1440p ultrawide we achieve the same results, at our around 60FPS. We’re mostly CPU limited here. At 4K we start to see the impact of a demanding resolution AND settings even on these cards: High settings see 45-47FPS on both, whilst ultra settings cut performance to just over 30 FPS for both cards. The inconsistencies at lower resolutions are caused by CPU limitations, and not indicative of the relative performance of the cards, but we did have trouble extracting performance from the 6800XT at ultra settings and lower resolutions – it was reluctant to exceed 45FPS despite driver updates and tweaking. Complex games and hardware sometimes deliver these anomalies. At higher resolutions, it performed fine, and it’s likely a minor setting adjustment would have resolved the issue but invalidated the controls to our benchmark.
The lesson here is how important balancing the whole system to the target resolution is in Flight sim 2020, as well as settings needing to be tweaked for optimum performance. Both of these cards are capable of playing MS Flight 2020 at high resolutions, a mix of high and ultra settings, and both are more than capable of an acceptable VR experience as well.
Related: Flight Simulator 2020 Performance Analysis: Get the most out of MS Flight 2020
Conclusions
As you’d hope and expect, both of these GPUs turn in an excellent performance. There’s nothing they won’t run well. Having spent over a month of tenting and general use with both of these cards, they are truly equal in performance in our opinion. There are some games and certain corner cases where one outperforms the other but on balance, the differences are infrequent and slight.
It’s really in feature set and versatility where we see the RTX 3080 offer more to a prospective buyer. RTX and DLSS may or may not be the next big thing for gaming, but it’s nice to have the option and in many titles, it just isn’t there if you’ve got the AMD card. We hope that developers will use the DirectX 12 ultimate API to bring hardware Ray tracing to the broadest audience possible, rather than sticking to Nvidias’ ‘RTX’ walled garden, whilst AMD need to develop and deliver their ‘Super Resolution’ functionality to the FidelityFX suite, to challenge Nvidia in dynamic upscaling.
Users who need a GPU for more than just gaming: for 3D content creation or data analysis, or to accelerate rendering – should look closely at benchmarks specific to their usage case. The RX 6800 XT performs better at some functions, and if you need a very high VRAM capacity then this or the 6800 is the cheapest way to get 16GB. It also offers better compatibility in Hackintosh and some Linux applications. However, for most tasks, the versatility of CUDA and Tensor cores combined with the higher bandwidth memory access make the 3080 a great choice. It will come down to what exactly you need to do.
Finally, the choice may well be out of your hands: If you see an opportunity to buy either of these GPU’s at a non-ridiculous price, and need a high-end gaming GPU, then either offers more than satisfactory performance at the highest resolutions and settings and will stand the test of time.