AMD Radeon RX 6800 XT and RX 6800 Review: Competing at the Top Again

AMD is back and there has been plenty of excitement building over the last few months. We have seen many unofficial leaks floating around the internet showing RX 6800 GPUs, also known as “Big Navi” hitting new highs and trading blows with Nvidia’s RTX GPUs. This is good news for Radeon graphics as it’s been some time since AMD has had a serious contender to take the GPU performance crown from team green.

Today we’ll take a close look at the Radeon RX 6800 as well as the RX 6800 XT. AMD claims the new GPUs are as good as or better than the RTX 3080 in performance and power consumption. If AMD has decent availability at launch, they just might convince some consumers to switch teams and stop waiting around for the elusive RTX GPUs with their artificially high price tags to become more readily available.

The RX 6800 XT and RX 6800 we have here today are aimed at the RTX 3080 and RTX 3070 enthusiast-class GPUs, and I think we’re all excited to see just how they measure up so let’s get to it.

Specifications and Features

The new Radeon RX 6000 series GPUs from AMD are based on RDNA2 which is an overhauled and fine-tuned version of their original RDNA architecture. AMD is claiming 50% of the power usage at the same frequency and a 30% increase in the maximum attainable frequency pushing the GPUs up to 2.2 GHz and beyond with a 54% performance/watt gain over RDNA(1).

This in itself is quite an accomplishment considering RDNA2 is still based on the same TSMC 7 nm node as its predecessor – meaning this all comes from architectural improvements. AMD concentrated on three key areas to accomplish this: more energy-efficient CUs, higher frequencies at the same power levels, and the Infinity Cache which offers an uptick in IPC.

High-Speed Design

AMD is in a unique position in the industry: housing CPU and GPU development under the same roof. During the development stages of RDNA2, AMD was able to incorporate design methodologies from the ZEN CPUs to streamline the Navi microarchitecture. This joint effort resulted in 1.3X the speed for the same perceived power per CU or 50% of the power per CU at the same frequency. The end result of which is double the CU count with a modest power increase and very high operating clocks.

Infinity Cache

Having such a high-speed graphics processor also requires a high-speed memory interface. On the surface, the 256-bit GDDR6 interface seems a bit anemic and it would be if not for the addition of AMD’s infinity cache. The 128 MB infinity cache acts much like a third level (L3) cache system but it has some unique features. A typical high-bandwidth GPU cache would take up a lot of die space but so does a 512-bit memory interface, not to mention the power cost for accessing off-die memory.

AMD turned to their CPU branch, specifically the EPYC server CPU which has a high-density L3 cache and these power-efficient designs are also 4x the density of RDNA L2 cache. The Infinity Cache runs on the Infinity Fabric bus (hence the name) on 16 channels moving 64 bytes each per clock at up to 1.94 GHz. This provides up to four times the bandwidth of a 256-bit GDDR6 interface and 2.4 times the bandwidth per Watt of a 512-bit GDDR6 interface. AMD claims on average a 34% reduction in overall memory latency by keeping data close to the CU and it provides the bandwidth needed to feed their High-Speed core design.

Overhauled Compute Units with Ray Tracing

AMD has made many improvements to the graphics engine: with microarchitecture, logic design, and a physical redesign streamlining for higher speeds and efficiency. According to AMD, this results in a 30% performance improvement at the same power levels. New this time in each CU is a Ray Accelerator unit for hardware-backed ray tracing which has a significant performance improvement over software backed ray tracing. There’s also dedicated logic included for ray tracing, sampler feedback, and texture shading for more efficient rendering by removing unnecessary operations and data movement.

The AMD team has realized that game engines are taking advantage of machine learning for many graphics-related tasks. So they have added more floating-point and integer mixed-precision operations in each CU to speed up these types of tasks. The Render Backing units have been redesigned from the ground up to hit power and frequency targets while doubling the 32 bpp color rate and now process eight 32-bit pixels per cycle. As you can see this isn’t just a refresh of the Navi core, AMD’s GPU and CPU engineering teams have been very busy over the last year.

This slideshow requires JavaScript.

DirectX 12 Ultimate Support

In the new RDNA2 pipeline AMD has made many architectural improvements adding four new capabilities. Aside from DirectX Raytracing they also included Variable Rate Shading, mesh shaders, and sampler feedback which will be used in many new gaming titles that support DX12 Ultimate.

Most of these new features will improve visuals and GPU performance. Take Variable Rate Shading (VRS). They take an 8×8 block of pixels and now instead of filling each pixel individually, VRS allows for a unique shading rate for each 8×8 pixel block improving performance without affecting visual quality.

AMD Radeon RX Series Comparison

AMD Radeon RX Series Specifications
Model AMD Radeon
RX 6900 XT
AMD Radeon RX 6800 XT AMD Radeon RX 6800 AMD Radeon RX 5700 XT
GPU Navi 21 Navi 21 Navi 21 Navi 10
Architecture RDNA2 RDNA
Process TSMC 7 nm
Transistors 26.8 Billion 10.3 Billion
Infinity Cache 128 MB N/A
Compute Units 80 72 60 40
Ray Accelerators 80 72 60 N/A
Stream Processors 5120 4608 3840 2560
ROPs 128 96 64
Game Clock 2015 MHz 2015 MHz 1815 MHz 1755 MHz
Boost Clock 2250 MHz 2250 MHz 2105 MHz 1905 MHz
Memory 16 GB GDDR6 8 GB GDDR6
Memory Speed 16 Gbps GDDR6 14 Gbps GDDR6
Memory Interface 256-bit
Throughput (FP32) 20.6 TFLOPs 18.6 TFLOPs 13.9 TFLOPs 9.75 TFLOPs
TDP (Watts) 300 W 300 W 250 W 225 W
Release Date 12/08/2020 11/18/2020 11/18/2020 07/07/2019
MSRP $999.00 $649.00 $579.00 $399.00

AMD Radeon RX 6000 GPUs

As you can see above the first release Big Navi GPUs all share the same Navi 21 die but with differing CU (compute unit) counts. They all have the 128 MB of Infinity Cache as well as 16 GB of 16 Gbps GDDR6 VRAM on a 256-bit bus (this gives all three 512 GB/s of memory bandwidth).

AMD Radeon RX 6900 XT

AMD’s flagship GPU is the Radeon RX 6900 XT which comes with the full 80 CUs of the Navi 21 die enabled and an MSRP of $1000. This price is steep for many, but the RX 6900 XT is also the most difficult die to harvest which in part justifies the heavy premium they charge over the RX 6800 XT. With the 6900 XT, we will also see a game clock of 2015 MHz and a boost clock 0f 2250 MHz. Surprisingly, it has the same TBP (total board power) as the RX 6800 XT which, to me, indicates some careful binning behind the scenes to squeeze the extra performance at the same power level. The RX 6900 XT will be available on December 12th, 2020.

AMD Radeon RX 6800 XT

Moving down the product stack we have the AMD Radeon RX 6800 XT which is a bit more reasonable price-wise. It’s still expensive at $650 but much better than the $1000 the RX 6900 XT is demanding. The 6800 XT has a pared-down Navi 21 die with 72 CUs enabled. It has identical clock speeds to the RX 6900 XT, 2015 MHz game, and 2250 MHz boost as well as the same 300 W TBP.

AMD Radeon RX 6800

That brings us to the AMD Radeon RX 6800 which is the last in the product stack at this time with a price tag of $580. Comparing to the RX 5700 XT released over a year ago at $400 that’s a significant increase but looking at the specifications it does appear justified. The Radeon RX 6800 uses the same Navi 21 die as the RX 6900 XT cut down to 60 CUs which is 50% more than the RX 5700 XT at 40 CUs. Considering the RDNA2 GPUs have an improved IPC and the RX 6800’s higher game and boost clocks of 1815 MHz and 2105 MHz respectively it should offer at least 50% more performance than the RX 5700 XT if not more. Lastly, the Radeon RX 6800 has a slightly lower TBP than it’s siblings coming in at 250 W, AMD is recommending a 650 W power supply for the 6800 and 750 W for either the 6800 XT or 6900 XT.

We’ll see the midrange and budget-oriented SKUs such as an RX 6700 or RX 6600 later this year or early next.

 

Retail Packaging and Accessories

The packaging for the Radeon RX 6000 cards are mostly black with a picture of the new cooler on the front. The card’s make and model information is located in the top right corner and both ends of the package. We also find AMD printed on the other two sides and the back has very little info aside from AMD’s address in California and a web address for downloading additional information.

Opening up the RX 6800 XT box, owners are greeted by a message from AMD welcoming them to “Team Red”. The two cards have slightly differing packaging. The RX 6800 comes in your typical container with an inner box that slides out from one end while the RX 6800 XT has a magnetic flap on the front side opening like a chest. Both cards have custom lift-out lids revealing them resting in a snug-fitting cut foam which keeps them snug neither card had anti-static packaging.

This slideshow requires JavaScript.

Meet the AMD Radeon RX 6000 GPUs

This time around AMD has done away with the blower-style cooler and moved to an updated triple-fan cooler for improved cooling performance. Both the RX 6800 XT and RX 6800 designs are nearly identical aside from the fact that the RX 6800 is a dual-slot while the RX 6800 XT is a 2.5 slot cooler. Our first look at the cards up close shows the three large fans with an “R” on the hub as the primary feature. The shroud is black and grey with angled chrome accents around the center fan. Both GPUs have a red pinstripe detailing the outside edge and the word RADEON which illuminates when powered up.

RX 6800 XT

RX 6800

 

A Closer Look

Both of the RX 6800 GPUs have the same I/O layout with two DisplayPorts (1.4a), one HDMI (2.1) output which will support 4K @ 144 Hz with a maximum resolution of 7680×4320 (8K) @ 60 Hz, and a USB Type-C output. The I/O plate is solid with the RADEON logo stamped into it. Power is handled by two 8-pin PCIe power leads for a total of 375 W available power, including the PCIe slot.

The heatsink itself isn’t the typical heat-pipe design.  AMD has completely overhauled its reference design incorporating a full-length vapor chamber connected to the fin array. The GPU core, memory, and the MOSFETs make direct contact with the heatsink using ultra soft thermal pads for optimal heat transfer. These are all cooled by their custom-designed triple axial fans for outstanding cooling and quiet performance.

Taking the backplate off the card exposes the two PWM controllers: one is an Infineon XDPE132G5D 16-phase for the core and the second is an IR 35217 from International Rectifier for controlling the memory voltages. Turning the RX 6800 XT over, we have the GPU core as you can see it’s covered in what AMD calls a Graphic thermal interface material which has some thickness to it so we chose not to remove it. Around the core, we have the Samsung GDDR6 memory, and to the right of the upper bank, we find the Cypress Semiconductor CYPD5137-40LQXIT USB-c port controller. We have included close-ups of the various IC’s in the slideshow below.

Below is a closeup of the power bits, Samsung GDDR6, and the control ICs.

This slideshow requires JavaScript.

 

AMD Radeon RX 6800

AMD Radeon RX 6800 XT

As you can see from the GPUz shots above, the RX 6800 has a game clock of 1815 MHz with a 2105 MHz Boost. The full boost speed from this sample managed to hit 2285 MHz under load. The RX 6800 XT has a game clock of 2015 MHz with a 2250 MHz Boost. AMD wasn’t kidding when they touted 2.5 GHz and beyond boost speeds with the 6800 XT reaching 2475 MHz under load. This boost speed was fairly consistent in lighter game loads and 2300-2350 MHz when really working hard.

AMD Radeon RX 6800 XT on the test bench…

Test System and Benchmark Methods

Test System Components
Motherboard ASUS ROG Maximus XII Extreme, ASRock Z490 PG Velocita
CPU Intel i9-10900K @ stock
CPU Cooler EVGA CLC 240
Memory 2×8 GB G.Skill Royal 3600 MHz CL16-16-16-36
SSD Gigabyte Aorus 2 TB NVMe Gen4 (OS + Applications)
Power Supply EVGA 750 W G3
Video Card AMD Radeon RX 6800 XT and RX 6800,  Adrenalin 20.45.01.12-11.6 Beta

 

Our test system is based on the latest mainstream Intel z490 platform and uses the i9-10900K 10/20t CPU. The CPU is overclocked to 4.9 GHz on all cores/threads with cache set to 4.3 GHz. The clock speeds used provides a good base to minimize any limitations the CPU may have on our titles, particularly when using the lower resolutions, and should be attainable with a good air cooler or better. The DRAM is in a 2×8 GB configuration at 3600 MHz with CL16-16-16-36-2T timings, a middle of the road option balancing performance and cost.

We have made some significant changes since the last update adding a few new titles and dropping some of the older games. More details can be found in the 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)
  • Unigine Superposition – Performance, 1080p High
  • Shadow of the Tomb Raider – DX12, “Highest” preset (will add RTX when it has been patched)
  • The Division 2 – DX12, Ultra preset, VSync Off
  • Assassin’s Creed: Odyssey – Ultra High preset, VSync Off
  • Far Cry New Dawn – Ultra defaults
  • F1 2020 – DX12, Very High defaults, TAA, and x16 AF, Australia track, do not show FPS counter
  • Metro: Exodus – DX12, Ultra defaults

Synthetic Benchmarks

Our first set of benchmarks hail from Underwriters Laboratories, who acquired Futuremark back in 2014. Earlier in 2018, a rebrand occurred, and since that time, Futuremark is now UL. The benchmarks have not changed, just the name. We chose to stick with 3DMark Fire Strike (Extreme) and 3DMark Time Spy as these tests give users a good idea of performance on modern titles.

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 compute, explicit multi-adapter, multi-threading, and runs at 2560×1440 resolution. 3DMark Port Royal is the first Ray Tracing benchmark designed for Windows 10 PCs and graphics cards with Microsoft DirectX Raytracing capabilities.

Looking at the results above we can see that the new AMD Radeon RX 6800 XT and RX 6800 did very well in all of our synthetic benchmarks. Surprisingly the RX 6800 XT once overclocked even bested the RTX 3090 in Fire Strike Extreme and the RX 6800 did the same to the RTX 3080. The Time Spy results weren’t quite as impressive but still outstanding trading blows with the RTX 3080 and RTX 2080 Ti respectively. These GPUs are a far cry from the original Navi, nearly doubling the RX 5700 XT scores.

The Big Navi GPUs even did well in Port Royal keeping up with the RTX 2080s which is impressive given that this is AMD’s first attempt at hardware-accelerated Ray Tracing.

Gaming Benchmarks

For gaming benchmarks, we have updated our testing suite to bring more modern titles into the mix. Gone are Battlefield V, F1 2018, Far Cry 5, AOTS:e, and World of Tanks, which were replaced with Metro Exodus, Assassin’s Creed: Odyssey, F1 2020, and Far Cry: New Dawn. We kept The Division 2 and Shadow of the Tomb Raider. The games should provide a good view of the overall performance of the card. Many of these are DX12 games.

1920×1080 (1080p) Results

The gaming benchmark results are a bit of a mixed bag. As we’ve been seeing with these new cards at lower resolutions like 1080p and even 1440p the results don’t seem to represent the horsepower these GPUs have under the hood. Looking at Far Cry New Dawn and Assassin’s Creed Odyssey the results were the same with both GPUs and similar results at 1440p. As you’ll see below in our higher resolution results, we didn’t see any change until we hit 4K. The AMD RX 6000 GPUs really seemed to shine in F1 2020 as the RX 6800 XT again topped the RTX 3090 and tied the RTX 3080 in Shadow of the Tombraider once overclocked. The remaining results were still well into the running but slightly behind the RTX GPUs.

AMD PCIe 4.0 Results

Since the new RDNA2 cards are PCIe 4.0 compliant, we wanted to test them on a PCIe 4.0 capable system, so we swapped out the CPU for a Ryzen 5 5600X at 4.8 GHz all cores, and the motherboard was switched to the MSI MEG X570 Godlike all other components remained the same as the original test system. The CPU has fewer cores than the i9-10900K which affected some of the scores. We tested the RX 6800 XT in RAGE mode which is an AMD preset overclock that is only available on the 6900 XT and 6800 XT graphics cards and it’s not as aggressive as it sounds. Here’s a GPUz screenshot for comparison.

Smart Access Memory

We also wanted to test AMD’s “Smart Access Memory” feature. This isn’t new and involves a sharing of the GPU Vram with the CPU. Up until now, this has been limited to 256 MB but Windows 10 has changed that and removed all the limits. For this feature to work it’s going to require a new Ryzen 5000 CPU, an RX 6000 GPU, and a 500 series chipset motherboard equipped with the updated BIOS. It took some time to locate the proper switch in the BIOS but this is it. Re-Size BAR Support must be enabled for this to work.

At this time I was unable to either get this to work or our game titles were unaffected by the change. Possibly the BIOS or feature still needs a bit of work. AMD claims this can give up to a 10% boost in gaming performance but our results were within a frame or two of having it disabled so we’re not including them at this time.

Title 1080p 1440p 4K
Assassins Creed: Odyssey 80 80 72
The Division 2 187 128 68
F1 2020 260 213 126
Far Cry: New Dawn 131 124 101
Metro: Exodus 122 100 66
Shadow of the Tomb Raider 182 149 81

If you compare the results above, you’ll find that the majority fall between our stock results and the overclocked results in the graphs above thanks to having RAGE mode activated.

2560×1440 and 4K UHD Results

Below are the higher resolution results starting with 2560×1440 and the gaining in popularity 3840×2160 (4K UHD). These resolutions prove to be a bit of a stretch for lower-end cards, but both of the AMD Radeon RX 6800 GPUs handled them well, even at 4K UHD.

Moving up in resolution we can see the gap tighten between the Green and Red teams and the RX 6800 XT is performing right where AMD said trading blows with the RTX 3080 @ 4K. The RX 6800 does quite well also being very close to or above 60 FPS in all of the 4K titles

Ray Tracing and DLSS Testing

Below we did some testing in Metro Exodus and Shadow of the Tomb Raider with ray tracing on at 1440p and 4K. As you can see, both the RX 6800 XT and RX 6800 produced very playable results above 60 FPS at 1440p in both Metro and SOTR. Moving up to 4K was a bit of a stretch for AMD’s first attempt at ray tracing. They didn’t manage to keep up with team Green, but as was said these are decent results for a first attempt and for the most part better than an RTX 2080 Ti.

We didn’t have results to compare with Nvidia’s DLSS since AMD’s version, FidelityFX Super Resolution is still in development but is suppose to accomplish similar results.

Overclocking

For overclocking, we used AMD’s Radeon software. The interface is very similar but the approach to overclocking the new Navi 21 GPUs is a bit different. Instead of a fixed maximum clock for the GPU core, you need to set a lower and upper range. This took a bit of testing as setting either range too high would diminish results or cause the benchmark to fail. I did notice that the AMD driver never crashed once during testing the overclocks. As you can see below the results are quite astounding with both GPUs hitting some impressive speeds. The RX 6800 managed almost 2500 MHz consistently settling in at 2490 MHz while the RX 6800 XT blew right past that and landed at 2585 MHz, the fastest speed I have ever seen out of a GPU that wasn’t running on exotic cooling. We also added 100 MHz to the memory speed to balance out our overclock. Overall this helped both cards improve their scores but the RX 6800 saw the most benefit. Below you can see the setting used for the RX 6800 XT.

 

Temperatures and Power Use

We test power consumption by running through the game benchmarks of Shadow of the Tomb Raider and F1 2020 at stock speeds and while overclocked. We monitor temperatures throughout this testing, with the peak temperature being what is listed in the data below. To more accurately simulate real gaming conditions, the benchmarks are extended (time) to allow the card to settle.

Temperatures on the Radeon RX 6000 GPUs during testing were well-controlled, peaking at 78 °C while overclocked and 76°C at stock. The fans ramped up slightly but were inaudible using the auto setting. These temperatures are a long way from the RDNA2 maximum of 110° C.

Power use for 300 W RX 6800 XT peaked at 509 W (system) running stock and 548 W while overclocked. A quality 750 W PSU will be adequate, including overclocking both CPU and GPU and still allowing for headroom and quiet operation. The RX 6800 on the other hand has a TBP of 250 W and peaked at 441 W while overclocked and 426 W at stock. AMD is recommending a 650 W PSU for this one.

Conclusion

AMD has really made some outstanding improvements in its RDNA2 architecture when compared to its predecessor. Making a leap in performance and efficiency this big while still within the same 7 nm node is almost unheard of. The engineering teams over at AMD should be proud of themselves for their hard work and thinking out of the box when it comes to problem-solving. The infinity cache appears to work very well and was a gamble on something a bit different but has really paid off for them.

Next, we have the prices – $650 for an AMD RX 6800 XT seems quite steep, but is still less than the Nvidia RTX 3080 with relatively the same performance, making it a good deal and worthy competitor at this price and performance point. The RX 6800 comes in slightly lower with a price tag of $580 and still a capable 4K gaming card. It is nearly a clone of the RX 6800 XT but with fewer Compute Units, a reduced TBP, and a 2-slot cooler instead of the 2.5-slot cooler found on the RX 6800 XT. It is still an improved and quiet three-fan cooler along with a robust 15-phase power delivery system. These aren’t the “screaming meanies” of a blower-type cooler but more in line with your typical AIB offering. They’re also quite sharp looking.

Overall, AMD has pulled a rabbit out of the hat. These are great GPUs with exceptional performance. and gets us quite excited for the upcoming Radeon RX 6900 XT release which is right around the corner. The battle for the performance crown between Red and Green isn’t over yet and it’s still anybody’s game. AMD is hitting on all cylinders offering performance, power efficiency, and competitive pricing. We can easily give “Big Navi” the Overclockers stamp of approval!

Click to find out what this means

– Shawn Jennings (Johan45)

Related Articles

Discussion

Leave a Reply

Your email address will not be published.


*


  1. BugFreak
    Great write-up. I really like that you tested it with the Intel rig and look forward to the comparison when/if you get an AMD 5000 series.

    There was testing done with a 5600X below the 1080p results. Not sure if it was the CPU or BIOS or what exactly but I tried testing the SAM (smart access memory) and the results didn't change in some of the titles we use I should have see a 6% uptick in FPS
    Johan45
    There was testing done with a 5600X below the 1080p results. Not sure if it was the CPU or BIOS or what exactly but I tried testing the SAM (smart access memory) and the results didn't change in some of the titles we use I should have see a 6% uptick in FPS

    Yea I saw that but with the minimal change I would assume something was not totally "right" yet. I'll wait to see what comes out in the future for sure though. If that ends up being the end result then AMD making a big deal about makes no sense so I figure it needs tweaking and time.
    Very good read, thank you! I think though with these cards it is going to hurt them in the long run if they do not get an equivalent to DLSS. The ray tracing performance doesn't seem to be as strong as the 3000 series so it will need something to compete.