A few days ago we unboxed the new RX 6800 XT from AMD, if you are interested, but haven’t seen it yet, we have the link down below. Now it’s time to dig deeper into the features and performance – are we finally at a point where we can have an all-AMD system with no compromises? Well… we are going to find out.
So what are the new features for this Big Navi card? Through architectural improvements AMD claims to have achieved up to 54% improvement per watt over the last generation and is now able to compete with NVIDIA. While they kept the 7nm manufacturing process, they doubled the die size and now we have more than double the transistors. With this improvement they were able to put 72 compute units as well as 72 ray accelerators in their latest creation – RX 6800 XT. This is up from 40 compute units on the RX 5700 XT.
They have also improved GPU frequencies with up to 30% improvement. We are now seeing these cards sometimes hitting 2400MHz speeds.
With the increased CU count and clock speeds there is a need to improve memory bandwidth to feed this GPU monster. All of this already increases power consumption so they had to innovate when it comes to memory. AMD has integrated new cache design into infinity fabric and called it Infinity Cache. This is the first of its kind and it works slightly differently from a typical cache on the CPU.
In servers the cache is broken up into banks and these are normally set up in serial, but with RDNA 2 infinity cache is placed in parallel to increase bandwidth. With peak bandwidth hitting almost 2TB per second making it 4 times the speed of 256bit GDDR6 interface while keeping it much more power efficient. Infinity cache sits as part of infinity fabric and is dynamically clocked, when there is a high load it will clock up to provide more memory to the GPU engine, and when the process is engine bound – infinity fabric will clock down to save power. With this in mind, we should see much higher use in 4k gaming and it will likely clock down while in 1080p as that is normally more engine or CPU bound.
An interesting part is latency, even though there is a larger die which in turn increases latency, with introduction of infinity cache – AMD claims there is an average 34% reduction in latency compared to RX 5700 XT which should improve performance at lower power.
With all of these implementations AMD says BIG NAVI has 20% improvement per clock per compute unit, almost double the compute units and also increased clock frequency by about 30%. To ensure RX 6800 XT hits the 300W power limit, AMD had to optimise power consumption and make it 30% less energy demanding per cycle.
Another large development is Smart Access Memory, SAM for short. This allows the CPU to have direct access to all VRAM on the GPU, thus improving performance. It is a function currently only available between new Zen 3 CPUs and RDNA 2 graphics cards, but it seems like it will be available across all brands in the future.
That is a whole lot of tweaks, but how does it stack up against other cards out there?
Let’s find out! We used our new AMD test bench with Ryzen 9 5950X. Here we tested the older RTX 2080 Super, new RTX 3070 and RTX 3090. We are keeping all of these cards at stock and for the new AMD card we have a separate benchmark where we enabled Smart Access Memory, this will be indicated on the graphs. Unfortunately we were not able to retest RTX3080 just yet so we are including a few game results from our previous tests while running Intel i7 10700k test bench. Please note that due to this change – results are not directly comparable especially at 1080p so we will not focus on them too much.
Starting with Shadow of the Tomb Raider – at 1080p we see all the cards evenly spaced out with RX 6800 XT only about 10% behind the RTX3090 who’s taking the lead here. Enabling SAM improves Radeon’s performance by about 5%.
In 1440p the gap between RX 6800 XT and RTX3090 widens to 16% and enabling SAM provides just 4% improvement. What is interesting – at this resolution RTX3080 performance is much higher as it is no longer CPU limited.
In 4k we see a similar gap between AMD and NVIDIA cards, while enabling SAM provides us a moderate 2% improvement. It is still free performance – so we can’t really complain there.
In Horizon Zero Dawn at 1080p we see very similar average FPS scores on all three high end cards with the only difference in 1 percentile performance.
Moving to 1440p we have RX 6800 XT only about 8% behind RTX3090, considering this card is less than half the price – not bad at all!
Setting resolution to 4K the gap between RTX3090 and other cards increase to 17% and RTX3080 has caught up, it does have low 1 percentiles when comparing to RX 6800 XT and that is actually where new AMD card has a winning feature 16GB of VRAM – Horizon Zero Dawn at 4k maxed out settings needs over 12GB of VRAM and RTX3080 only has 10GB.
Next we have CS:GO and at 1080p the RTX cards perform about 10% better on 1 percentiles but at 200 FPS mark I feel it doesn’t really make a difference.
Bumping up the settings all the way to 4K we see RTX3090 is still the best performer but it’s nice to see that RX 6800 XT is matching it at least on average FPS.
So far results are very varied, but there’s no doubt that RX 6800 XT is competitive and with SAM enabled it appears to be performing really well.
Next I wanted to check its performance in more synthetic scenarios so we ran a few 3DMark benchmarks. Starting with DX11 test – Firestrike, which is running at 1080p we see RX 6800 XT actually beating the RTX3090 and surprisingly – enabling SAM actually made the score lower!
When we go up to Fire Strike Ultra which is a 4k test, the gap between AMD and NVIDIA cards shrinks but AMD is still on top. In this one enabling SAM does not make much of a difference.
Moving to DX12 test – Time Spy – which is running at 1440p, the tables have turned and RX 6800 XT gives its lead back to RTX3090.
Going up to Time Spy Extreme, which is a 4K test we see the same story with no changes after enabling SAM.
Lastly we have ray tracing test Port Royal and here RTX3090 just blows RX 6800 XT out of the water, the difference here is almost 50%. At the same time the new AMD card does beat RTX3070 by a considerable amount.
Since we are on the topic of ray tracing, for our tests AMD has provided a range of games to try out and two of them had built in benchmarks – please note that both of these set-ups are still in beta so performance here may not represent finished product. It does however represent the relative performance to other cards right now.
First we have Dirt 5 with Ray tracing options turned on, in 1080p we have RX 6800 XT beating out RTX3090 by about 25% while SAM is enabled.
When we bump resolution up to 1440p we see RTX3090 getting closer but still lagging behind.
Turning the settings all the way up to 4k, the gap shrinks to just 6%.
While all the settings were exactly the same, this game supports Variable Rate Shading which allows certain scenes to be rendered at lower rate to reduce the required processing power. This way things like corner frames or distant objects may be rendered at lower quality while elements that are in the middle of users focus area will be rendered at full quality resulting in optimised performance and higher frame rates without noticeable visual differences.
Next game is Riftbreaker and here I am really impressed with the ray tracing effects for both lighting and shadows as well as overall performance. At 1080p we have RTX3090 in the lead by about 20% but if you look at the frame rates, those are still at around 200FPS which is awesome.
In 1440p the gap actually increases to 35% and what is interesting here – RTX3070 and in fact RTX 2080 Super are also not that far behind.
Moving to 4K we see RTX3090 once again ahead by a long margin and its CPU is still at max usage. It is very unusual to see CPU being leveraged in a GPU intensive application, there is clearly still a lot of optimisation to be done before this game goes live, it was an interesting observation nevertheless.
That’s a lot of gaming and synthetic tests… but raw performance paired with 16GB of VRAM should also be good in productivity tasks, right? Well, kind off. In Blender while doing the shorter BMW test, this card is actually performing worse than RTX3070, but in the longer Classroom test it slots in nicely between RTX3090 and 3070, this is where that fast cache is going to come in very handy.
We also ran Blackmagic Davinci Resolve benchmark from Puget Systems and found RX 6800 XT actually pulling its weight rather well, scoring way above RTX3070 as we would expect, and landing about 10% behind RTX3090 while SAM is enabled. Considering its price and 16GB of VRAM this card may actually be very good for video editing, we will have to do some more testing on this in the future.
With direct performance numbers out of the way, let me cover a few other important parameters, starting with thermals. A quick note – our ambient studio temperature is 26.5C. In gaming RX 6800 XT would stay in the low to mid 70s while under heavy load. In Blender it hit around 70C and stayed there and it seems enabling SAM does not provide any noticeable impact on thermals.
This leads us well to the noise levels and I was completely blown away here. This card is incredibly quiet. In fact, while doing benchmarks I have not even noticed it. Only when we came to testing noise performance, I noticed some sound. To be honest, the testing was done on the open bench so it got more fresh air and needed less cooling.
At idle fans actually turn off and it becomes completely silent, to test it further we had to turn on Furmark, set it to 4k and leave it for 10minutes to warm up, after that we had a light hum at 42dBa with fan speed reaching only 45%. For those interested in the maximum noise that this card can generate – we set fan speeds to 100% and at peak it had 58dBa. This is loud but unlikely it will ever reach this in normal conditions.
AMD has achieved this performance by including three redesigned fans and full board-length fin array as well as using graphite thermal interface for optimum heart transfer to the vapour chamber and results speak for themselves.
There are many more features and options I would love to talk about, such as support for DirectX12 Ultimate, Radeon FidelityFX and overclocking options, but that will have to be another article.
To sum things up and clarify if we are finally at the point where we can have an all AMD system with no compromises: the short answer is no and I don’t think we will ever be. It is very difficult to make something good for everyone. There will always be trade offs. But damn they have made a great piece of kit and it can certainly compete with Nvidia – all I hope is that they keep up with support and work with developers to optimise game and productivity performance.