SAPPHIRE Toxic RX 6900 XT Limited Edition Review

1. Introduction2. Sapphire Toxic RX 6900 XT3. Testing Methodology4. 3DMark Fire Strike & Time Spy5. 3DMark DirectX Raytracing6. Borderlands 37. Control8. Death Stranding9. The Division 210. F1 202011. Far Cry New Dawn12. Gears 513. Ghost Recon: Breakpoint14. Metro Exodus15. Middle Earth: Shadow of War16. Red Dead Redemption 217. Shadow of the Tomb Raider18. Total War Saga: Troy19. Watch Dogs: Legion20. Clock Speed21. Temperatures22. Acoustics23. Noise-Normalised Temperatures24. Power Consumption25. Overclocking26. Closing Thoughts27. View All Pages

It's been some years since we last looked at a graphics card from Sapphire's Toxic brand, but that changes today. Toxic is back, and it's back with a bang, as we review the RX 6900 XT Toxic Limited Edition, featuring a 360mm all-in-one liquid cooler. Sapphire claims this is the fastest RX 6900 XT on the market, with a rated boost clock of 2660MHz when using the card's ‘Toxic Boost' mode. Let's see if that claim holds true.

As far as I can tell, the last Sapphire Toxic graphics card to hit the market was the R9 390, all the way back in May 2016, so it's been almost five years without a proper flagship card from Sapphire. The Toxic RX 6900 XT Limited Edition is the first Toxic card to come with an integrated all-in-one liquid cooling solution which cools the GPU, while a fan and two heatpipes deal with the VRM and memory.

With its Toxic brand geared for all-out performance, Sapphire has also implemented what it calls ‘Toxic Boost', a one-click overclock which is accessible via the TriXX software. This raises the GPU's boost clock from 2365MHz up to 2660MHz, bumps up power draw to 400W total board power and also increases memory speeds by 0.8Gbps.

There's also the choice between the Performance and Quiet BIOS, plenty of RGB lighting as well as a beefed-up PCB compared to reference. Let's see what this card is really about.

	RX 6900 XT	RX 6800 XT	RX 6800	RX 5700 XT	RX 5700
Architecture	RDNA 2	RDNA 2	RDNA 2	RDNA	RDNA
Manufacturing Process	7nm	7nm	7nm	7nm	7nm
Transistor Count	26.8 billion	26.8 billion	26.8 billion	10.3 billion	10.3 billion
Die Size	519 mm²	519 mm²	519 mm²	251 mm²	251mm²
Ray Accelerators	80	72	60	n/a	n/a
Compute Units	80	72	60	40	36
Stream Processors	5120	4608	3840	2560	2304
Game GPU Clock	Up to 2015MHz	Up to 2015MHz	Up to 1815MHz	Up to 1755MHz	Up to 1625MHz
Boost GPU Clock	Up to 2250MHz	Up to 2250MHz	Up to 2105MHz	Up to 1905MHz	Up to 1725MHz
Peak SP Performance	Up to 23.04 TFLOPS	Up to 20.74 TFLOPS	Up to 16.17 TFLOPS	Up to 9.75 TFLOPS	Up to 7.95 TFLOPS
Peak Half Precision Performance	Up to 46.08 TFLOPS	Up to 41.47 TFLOPS	Up to 32.33 TFLOPS	Up to 19.5 TFLOPS	Up to 15.9 TFLOPS
Peak Texture Fill-Rate	Up to 720 GT/s	Up to 648.0 GT/s	Up to 505.2 GT/s	Up to 304.8 GT/s	Up to 248.4 GT/s
ROPs	128	128	96	64	64
Peak Pixel Fill-Rate	Up to 288.0 GP/s	Up to 288.0 GP/s	Up to 202.1 GP/s	Up to 121.9 GP/s	Up to 110.4 GP/s
AMD Infinity Cache	128MB	128MB	128MB	n/a	n/a
Memory	16GB GDDR6	16GB GDDR6	16GB GDDR6	8GB GDDR6	8GB GDDR6
Memory Bandwidth	512 GB/s	512 GB/s	512 GB/s	448 GB/s	448 GB/s
Memory Interface	256-bit	256-bit	256-bit	256-bit	256-bit
Board Power	300W	300W	250W	225W	185W

The make-up of the RX 6900 XT GPU has many similarities to the RX 6800 XT. Still fabricated on TSMC’s 7nm process node, it uses the same Navi 21 GPU, measuring 519 mm². The key difference is that the RX 6900 XT silicon is fully populated, with 80 Compute Units (CUs) compared to 72 for the 6800 XT. This means the flagship chip has an extra 512 Stream Processors, giving a grand total of 5120 shaders.

Rated clock speed is also identical between these two RDNA 2 GPUs. Reference RX 6900 XT has a rated game clock of 2015MHz and a rated boost clock of up to 2250MHz, but Sapphire has pushed the rated boost clock up to 2365MHz for the Performance BIOS. Enabling Toxic Boost sees this increase by almost 300MHz, to 2660MHz.

AMD has also increased the memory capacity, with 16GB of GDDR6 memory found on every RX 6000 SKU (so far). The memory is clocked at 16Gbps (or 16.8Gbps using Toxic Boost), and operates over a 256-bit memory interface for total bandwidth of 512 GB/s. However, AMD has also implemented a 128MB Infinity Cache on the Navi 21 GPU, to significantly increase effective memory bandwidth without excess power draw.

Speaking of power, the reference RX 6900 XT has a rated total board power (TBP) of 300W. Sapphire has increased this for both the Quiet and Performance BIOS, to 350W and 360W respectively, but we look at power draw in more detail later on in the review.

The Sapphire Toxic RX 6900 XT ships in a large black box, with prominent Toxic branding visible on the front. On the back, there's more branding as well as some key specs of the GPU.

Inside, the graphics card and AIO cooler come very well packaged in a thick foam surround. My sample shipped direct from Taiwan and arrived in perfect condition.

Sapphire includes a nifty screwdriver toolkit, which is a nice touch, while we also get a small assortment of screws used for mounting the radiator to the chassis.

The card itself is quite striking, and it's quite hard to miss with the 360mm AIO attached to it. Sapphire has gone for a hybrid approach, with the AIO responsible for the cooling of the GPU die, while an internal heatsink with two heatpipes – plus a 100mm fan – deal with the VRM and memory modules.

In terms of the shroud, Sapphire is using a combination of aluminium and plastic, and it feels very solid in the hand. The 100mm fan is covered by a series of vents which certainly draws the eye.

As for dimensions, the graphics card measures 268.77 x 130.75 x 44.07mm, so by modern standards it is actually pretty compact. That is, until you consider you also need room for a 360mm radiator in your chassis…

On the front side, the two radiator tubes extend away from the card. We can also note the BIOS switch towards the I/O bracket – by default, the Toxic uses the Performance BIOS, but users can switch to the Quiet BIOS as well. A third position on the switch lets you change BIOS modes directly from the TriXX software.

Display outputs consist of 3x DisplayPort 1.4 and 1x HDMI 2.1. Power connectors consist of 2x 8-pins and 1x 6-pin.

The AIO is an Asetek-branded unit, measuring 394 x 120 x 52.48mm (including the fans). The sleeved tubing measures 470mm long, while Sapphire is using three FirstDO 120mm RGB fans on the radiator, with model code ‘FD1225H12D'.

Out of the box, these fans are installed to the underside (or tube-side) of the radiator, with all three fan cables contained within a single sleeved cable running alongside one of the AIO tubes to keep things tidy. This area of installation implies the unit is designed to be installed in the roof of the chassis, with the fans acting as exhausts.

If this doesn't work for you, it is possible to remove the fans and put them on the opposite side of the radiator, allowing the radiator to be front-mounted with the fans acting as intakes. The only slight hiccup here is that by removing the fans, the cables do get slightly more untidy – by default, they are neatly routed along the frames of the fans themselves, but this isn't possible when installing the fans on the reverse side of the radiator.

Images above, of the PCB and cooling solution, provided by Sapphire.

Coming to the PCB, Sapphire asked us not to disassemble the Toxic graphics card as we would usually do, likely because review units are limited (and expensive) and we could potentially affect thermal performance or rip thermal pads etc.. Instead, Sapphire provided its own photos of the PCB and cooling solution.

We can see there's a total of 18 phases for the VRM, of which Sapphire claims 14 phases make up the GPU power delivery, and then two phases for the memory and two for SOC. Sapphire is using an Infineon XDPE132G5D controller for the GPU VRM, and the International Rectifier IR35217 controller for the memory.

For the cooling, Sapphire utilises a die-cast heatsink with two heatpipes, and this cools the memory and VRM, with airflow from the 100mm fan on the card itself.

Driver Notes

• All Nvidia GPUs (except RTX 3060 Ti) were benchmarked with the 457.09 driver.
• RTX 3060 Ti was benchmarked with the 457.40 driver supplied to press.
• All AMD GPUs (except RX 6000) were benchmarked with the Adrenalin 20.11.1 driver.
• RX 6800 and RX 6800 XT were benchmarked with the Adrenalin 20.45.12.1 driver supplied to press.
• RX 6900 XT was benchmarked with the Adrenalin 20.45.01.14 driver supplied to press.

We did all of our game testing using Toxic Boost, the one-click overclock built into TriXX. We test Toxic Boost alongside the Performance and Quiet BIOS when looking at power, thermals and noise.

Test System

We test using the a custom built system from PCSpecialist, based on Intel's latest Comet Lake-S platform. You can read more about it over HERE, and configure your own system from PCSpecialist HERE.

CPU	Intel Core i9-10900K Overclocked to 5.1GHz on all cores
Motherboard	ASUS ROG Maximus XII Hero Wi-Fi
Memory	Corsair Vengeance DDR4 3600MHz (4 X 8GB) CL 18-22-22-42
Graphics Card	Varies
System Drive	500GB Samsung 970 Evo Plus M.2
Games Drive	2TB Samsung 860 QVO 2.5″ SSD
Chassis	Fractal Meshify S2 Blackout Tempered Glass
CPU Cooler	Corsair H115i RGB Platinum Hydro Series
Power Supply	Corsair 1200W HX Series Modular 80 Plus Platinum
Operating System	Windows 10 2004

Comparison Graphics Cards List

• Gigabyte RTX 3090 Eagle OC 24GB
• Nvidia RTX 3080 FE 10GB
• Nvidia RTX 3070 FE 8GB
• Nvidia RTX 3060 Ti FE 8GB
• Nvidia RTX 2080 Ti FE 11GB
• Nvidia RTX 2080 Super FE 8GB
• Nvidia RTX 2080 FE 8GB
• Nvidia RTX 2070 Super FE 8GB
• Nvidia RTX 2060 Super FE 8GB
• AMD Radeon VII 16GB
• AMD RX 6900 XT 16GB
• Gigabyte RX 6900 XT Gaming OC
• AMD RX 6800 XT 16GB
• AMD RX 6800 16GB
• AMD RX 5700 XT 8GB
• AMD RX Vega 64 8GB

Software and Games List

• 3DMark Fire Strike & Fire Strike Ultra (DX11 Synthetic)
• 3DMark Time Spy (DX12 Synthetic)
• 3DMark Raytracing Feature Test (DXR Synthetic)
• Borderlands 3 (DX12)
• Control (DX12)
• Death Stranding (DX12)
• The Division 2 (DX12)
• F1 2020 (DX12)
• Far Cry New Dawn (DX11)
• Gears 5 (DX12)
• Ghost Recon: Breakpoint (Vulkan)
• Metro: Exodus (DX12)
• Middle Earth: Shadow of War (DX11)
• Red Dead Redemption 2 (Vulkan)
• Shadow of the Tomb Raider (DX12)
• Total War Saga: Troy (DX11)
• Watch Dogs: Legion (DX12)

We run each benchmark/game three times, and present mean averages in our graphs. We use OCAT to measure average frame rates as well as 1% low values across our three runs.

Fire Strike is a showcase DirectX 11 benchmark for modern gaming PCs. Its ambitious real-time graphics are rendered with detail and complexity far beyond other DirectX 11 benchmarks and games. Fire Strike includes two graphics tests, a physics test and a combined test that stresses the CPU and GPU. (UL).

3DMark Time Spy is a DirectX 12 benchmark test for Windows 10 gaming PCs. Time Spy is one of the first DirectX 12 apps to be built the right way from the ground up to fully realize the performance gains that the new API offers. With its pure DirectX 12 engine, which supports new API features like asynchronous compute, explicit multi-adapter, and multi-threading, Time Spy is the ideal test for benchmarking the latest graphics cards. (UL).

Kicking off with 3DMark, we can immediately see sizable performance gains for the Toxic when compared to AMD's reference card. It's 10% faster in both Fire Strike and Fire Strike Ultra, and 11% faster in Time Spy. It's also faster than the RTX 3090 in all three benchmarks.

Real-time ray tracing is incredibly demanding. The latest graphics cards have dedicated hardware that’s optimized for ray-tracing. The 3DMark DirectX Raytracing feature test measures the performance of this dedicated hardware. Instead of using traditional rendering techniques, the whole scene is ray-traced and drawn in one pass. The result of the test depends entirely on ray-tracing performance. (UL).

Ray tracing is an area where Nvidia has the upper hand, and we can see that in the 3DMark DXR feature test. Despite outperforming the reference 6900 XT by 11%, the Toxic is still 43% slower than the RTX 3090 and 31% slower than RTX 3080.

Borderlands 3 is an action role-playing first-person shooter video game developed by Gearbox Software and published by 2K Games. It is a sequel to 2012's Borderlands 2, and the fourth entry in the main Borderlands series. Borderlands 3 was released on 13 September 2019 for Microsoft Windows, PlayStation 4, Xbox One. (Wikipedia).

Engine: Unreal Engine 4. We test using the Badass preset, DX12 API.

Our first game of the day is Borderlands 3, an AMD-sponsored title using Unreal Engine 4. Borderlands is one of the best case scenarios for the Toxic, as it outperforms the reference 6900 XT by 10% at 4K, giving it an 8% advantage over RTX 3090.

Control is an action-adventure video game developed by Remedy Entertainment and published by 505 Games. Control was released on 27 August 2019 for Microsoft Windows, PlayStation 4, and Xbox One. (Wikipedia).

Engine: Northlight Engine. We test using the High preset, with 4x MSAA, DX12 API.

Next we come to Control, a game which favours Nvidia hardware. Here, the Toxic manages to outperform the reference 6900 XT by 8% at 4K, but it still comes in 12% slower than the RTX 3090. It has closed the gap on Nvidia significantly though, as the reference 6900 XT is 8% slower than the RTX 3080, which the Toxic is able to match.

Death Stranding is an action game developed by Kojima Productions. It is the first game from director Hideo Kojima and Kojima Productions after their split from Konami in 2015. It was released by Sony Interactive Entertainment for the PlayStation 4 in November 2019 and by 505 Games for Windows in July 2020. (Wikipedia).

Engine: Decima. We test using the Very High preset, with TAA, DX12 API.

 

While Death Stranding isn't an AMD-sponsored title, Team Red does well in this game and that is on show here. The Toxic comes in 9% ahead of both AMD's reference 6900 XT and RTX 3090 when testing at 4K.

Tom Clancy's The Division 2 is an online action role-playing video game developed by Massive Entertainment and published by Ubisoft. The sequel to Tom Clancy's The Division (2016), it is set in a near-future Washington, D.C. in the aftermath of a smallpox pandemic, and follows an agent of the Strategic Homeland Division as they try to rebuild the city. (Wikipedia).

Engine: Snowdrop. We test using the Ultra preset, but with V-Sync disabled, DX12 API.

Next is The Division 2. This game saw the Toxic outperforming reference 6900 XT by 11%, the joint-biggest margin of difference between the two cards we will see today. It's still slower than RTX 3090, but by 8% now – whereas the reference card is 17% slower.

F1 2020 is the official video game of the 2020 Formula 1 and Formula 2 Championships developed and published by Codemasters. It is the thirteenth title in the Formula 1 series developed by the studio and was released on 7 July 2020 for pre-orders of the Michael Schumacher Edition and 10 July 2020 for the Seventy Edition on Microsoft Windows, PlayStation 4, Xbox One. (Wikipedia).

Engine: EGO. We test using the Ultra High preset, DX12 API.

F1 2020 is the other game where Sapphire's Toxic outperforms the reference design by 11%. That's a big enough difference to bring the Toxic level with the RTX 3090 at 4K, where the reference card is 9% slower.

Far Cry New Dawn is an action-adventure first-person shooter developed by Ubisoft Montreal and published by Ubisoft. The game is a spin-off of the Far Cry series and a narrative sequel to Far Cry 5. It was released for Microsoft Windows, PlayStation 4 and Xbox One on February 15, 2019. (Wikipedia).

Engine: Dunia 2. We test using the Ultra preset, with the HD Textures pack, DX11 API.

Far Cry New Dawn is heavily CPU-limited, even at 1440p, so the data isn't particularly useful. Even at 4K, the Toxic is just 4% faster than the reference 6900 XT and 3% faster than RTX 3090.

Gears 5 is a third-person shooter video game developed by The Coalition and published by Xbox Game Studios for Xbox One, Microsoft Windows and Xbox Series X. It is the fifth installment of the Gears of War series and the sequel to Gears of War 4. The ultimate edition was released on September 6, 2019, while the standard edition of the game was released worldwide on September 10, 2019. (Wikipedia).

Engine: Unreal Engine 4. We test using the Ultra preset, with Best Animation Quality (instead of Auto), DX12 API.

Gears 5 is another AMD title, but at 4K we can see the reference 6900 XT is still 3% slower than the RTX 3090. Thanks to the Toxic's increased clock speed, it comes in 6% faster than the 3090.

Tom Clancy's Ghost Recon Breakpoint is an online tactical shooter video game developed by Ubisoft Paris and published by Ubisoft. The game was released worldwide on 4 October 2019, for Microsoft Windows, PlayStation 4 and Xbox One, (Wikipedia).

Engine: AnvilNext 2.0. We test using the Ultra preset, with AA disabled, Vulkan API.

 

Coming to Ghost Recon: Breakpoint, this is an interesting title where AMD's Big Navi GPUs fall away from the Nvidia competition as we increase in resolution. Despite being 8% faster than the reference 6900 XT at 4K, the Toxic is still 12% slower than the RTX 3090. At a guess, I'd say memory bandwidth could be the limiting factor here.

Metro Exodus is a first-person shooter video game developed by 4A Games and published by Deep Silver in 2019. It is the third instalment in the Metro video game series based on Dmitry Glukhovsky's novels, following the events of Metro 2033 and Metro: Last Light. (Wikipedia).

Engine: 4A Engine. We test using the Ultra preset, but with Hairworks and Advanced PhysX turned off, DX12 API.

As for Metro Exodus, the Toxic gets an 8% boost compared to the reference 6900 XT here. It's still slower than the RTX 3090 at 4K, but by a significantly reduced margin of 7%.

Middle-earth: Shadow of War is an action role-playing video game developed by Monolith Productions and published by Warner Bros. Interactive Entertainment. It is the sequel to 2014’s Middle-earth: Shadow of Mordor, and was released worldwide for Microsoft Windows, PlayStation 4, and Xbox One on October 10, 2017. (Wikipedia).

Engine: LithTech Firebird. We test using the Very High preset, DX11 API.

Shadow of War doesn't scale as well as some of the other games we've tested, with the Toxic outperforming AMD's reference 6900 XT by 7% at 4K. That puts it just 5% slower than the RTX 3090.

Red Dead Redemption 2 is a 2018 action-adventure game developed and published by Rockstar Games. The game is the third entry in the Red Dead series and is a prequel to the 2010 game Red Dead Redemption. Red Dead Redemption 2 was released for the PlayStation 4 and Xbox One in October 2018, and for Microsoft Windows and Stadia in November 2019. (Wikipedia).

Engine: Rockstar Advance Game Engine (RAGE). We test by manually selecting Ultra settings (or High where Ultra is not available), TAA, Vulkan API.

 

Red Dead Redemption 2 meanwhile, sees the Toxic fall just 3% behind the RTX 3090 at 4K. It's 5 FPS, or 8% faster, than AMD's reference card in this title.

Shadow of the Tomb Raider is an action-adventure video game developed by Eidos Montréal in conjunction with Crystal Dynamics and published by Square Enix. It continues the narrative from the 2013 game Tomb Raider and its sequel Rise of the Tomb Raider, and is the twelfth mainline entry in the Tomb Raider series. The game released worldwide on 14 September 2018 for Microsoft Windows, PlayStation 4 and Xbox One. (Wikipedia).

Engine: Foundation Engine. We test using the Highest preset, with TAA, DX12 API.

Shadow of the Tomb Raider is another good showcase for the Toxic, despite it being an Nvidia title. It's just 3% slower than the RTX 3090 at 4K, and is actually slightly faster than the Nvidia flagship at 1080p and 1440p.

Total War Saga: Troy is a 2020 turn-based strategy video game developed by Creative Assembly Sofia and published by Sega. The game was released for Windows on 13 August 2020 as the second installment in the Total War Saga subseries, succeeding Thrones of Britannia (2018). (Wikipedia).

Engine: TW 3 Engine. We test using the Ultra preset, DX11 API.

As for Total War Saga: Troy, this is actually the game where we see the 6900 XT perform at its worst, relatively speaking, when compared to the RTX 3090. The Toxic is 15% slower than the RTX 3090 here, so it's a significant margin. The Toxic is also just 4% slower than the reference card, so the added clock speed doesn't make much difference in this title.

Watch Dogs: Legion is a 2020 action-adventure game published by Ubisoft and developed by its Toronto studio. It is the third instalment in the Watch Dogs series, and the sequel to 2016's Watch Dogs 2. Legion was released on October 29, 2020 for Microsoft Windows, PlayStation 4, Xbox One, and Stadia. (Wikipedia).

Engine: Disrupt. We test using the Ultra preset, DX12 API.

Finally we come to Watch Dogs: Legion. The Toxic is able to outperform the RTX 3090 by 13% in this title, but only at 1080p. Once more, as we scale up the resolution, the Big Navi GPU falls away, and it is 5% slower at 4K.

Here we present the average clock speed for each graphics card while running the 3DMark Time Spy stress test for 30 minutes. We use GPU-Z to record the GPU core frequency during the Time Spy runs. We calculate the average core frequency during the 30 minute run to present here.

Looking at the Toxic's clock speed while testing Toxic Boost, the Performance BIOS and the Quiet BIOS, we can see a clear advantage to Toxic Boost in that it operates the core at a significantly higher frequency. Both the Quiet and Performance BIOS are still faster than the reference 6900 XT, and even Gigabyte's Gaming OC, but Toxic Boost brings average clock speed to nearly 2600MHz, which is highly impressive.

For our temperature testing, we measure the peak GPU core temperature under load. A reading under load comes from running the 3DMark Time Spy stress test for 30 minutes.

In terms of thermals, it's hardly a surprise to see the Toxic offering sub-70C operating temperatures regardless of the BIOS used. The Performance BIOS gives the lowest temperatures, as a result of its increased fan speed, while the Quiet BIOS is the hottest-running option as it significantly decreases fan speed. Toxic Boost is still hotter than the Performance BIOS as a result of its increased power limit and operating frequencies.

We take our noise measurements with the sound meter positioned 1 foot from the graphics card. I measured the noise floor to be 32 dBA, thus anything above this level can be attributed to the graphics cards. The power supply is passive for the entire power output range we tested all graphics cards in, while all CPU and system fans were disabled. A reading under load comes from running the 3DMark Time Spy stress test for 30 minutes.

Noise levels is an interesting area to look at. There's no doubt about it, the Quiet BIOS is impressively quiet, recording just 38dBa while the fans spun at 900rpm (37%.) The Performance mode is significantly louder however, as the fans spin up to 1290rpm, or 53%. Toxic Boost takes that a step further and pushes fan speed to 1410rpm (58%), producing 46dBa.

The latter two modes are certainly quite audible, and I would say Toxic Boost does make the fans overall reasonably loud. Let's see what happens when we noise-normalise…

Following on from our stock thermal and acoustic testing, here we re-test the operating temperature of the GPU, but with noise levels normalised to 40dBa. This allows us to measure the efficiency of the overall cooling solution as varying noise levels as a result of more aggressive fan curves are no longer a factor.

As a result of the different power targets, we don't get identical results when testing noise-normalised thermals with the Toxic. The Quiet BIOS has the lowest power target, and therefore offers the lowest temperatures. Even Toxic Boost, with its 400W target, hit just 75C Junction temperature, which is 12C cooler than the reference 6900 XT. Of course, we'd expect these temperatures to be lower considering the Toxic uses a 360mm liquid cooler, but it's still good to see the AIO is more than a match for this 400W GPU.

We measure system-wide power draw from the wall while running the 3DMark Time Spy stress test for 30 minutes.

We also use Nvidia PCAT to measure power draw of the graphics card only, with readings from both the PCIe slot and the PCIe power cables combined into a single figure. This provides us with significantly more accurate data to work with as it is measuring only the GPU power, and not total system power which is a fundamentally imprecise measurement.

Power draw for Toxic Boost is obviously the highest of all three modes tested, averaging 404W over our stress test. There's not much between the Quiet and Performance BIOS though, with the former drawing 342W compared to 352W for the latter. Both are still comfortably more power-hungry than reference however.

After testing with Toxic Boost enabled, we couldn't get much more out of the Toxic. We maximised the power limit (+15%) and tweaked the core to 2700MHz, while we added another 20MHz to the memory.

This manual overclock barely delivered any extra performance compared to Toxic Boost. In fact, with just a 2FPS difference in F1 2020, and then an improvement of just a single frame for Gears 5 and Watch Dogs: Legion, I'd say there is no actual difference with this overclock applied.

Power draw didn't really change either, with a negligible increase of just 2W based on our testing. Sapphire has clearly tweaked Toxic Boost pretty aggressively, as I was unable to extract anything more from the GPU.

Today marks the launch of Sapphire's Toxic RX 6900 XT Limited Edition, the first Toxic-branded card to hit the market in almost five years. Harnessing the power of AMD's flagship RX 6900 XT, we can think of the Toxic as the best case scenario for this GPU – it's been unshackled from its power constraints, while a 360mm AIO takes care of any cooling worries we might have around a 400W graphics card.

In all honesty, I wasn't expecting the performance difference versus the reference 6900 XT to be as big as it is. Granted, we tested using Sapphire's Toxic Boost mode, which is a one-click overclock found within the TriXX software, but realistically who is going to buy this kind of GPU and not want to run it in its fastest possible mode?

Toxic Boost enables the Toxic 6900 XT to outperform the reference card by up to 11% at 4K, but by 8% on average. This means the Toxic can really close the gap on RTX 3090. Our data puts the reference 6900 XT 10% slower than the RTX 3090, again at 4K. For the Sapphire Toxic, this gap is reduced to just 3%, close enough to call it a tie. In fact, at 1440p, there's on average 0% difference between the two GPUs, and the Toxic is even 3% faster at 1080p, though I can't imagine anyone spending this sort of cash for 1080p gaming.

The Toxic is also able to run out as a much more convincing winner against the RTX 3080. Where the reference 6900 XT is just 2% faster on average, the Toxic increases that to 11%. It's also 18% faster than the reference RX 6800 XT at 4K.

It's not a surprise to learn that strapping a 360mm AIO to this GPU results in pretty impressive thermals. In fact, even when using Toxic Boost – which raises total board power to 400W – we didn't see the GPU run hotter than 64C. When noise-normalised, it is comfortably cooler than either AMD's reference 6900 XT, or Gigabyte's Gaming OC model that we reviewed last month.

Of course, adding 100W of power negatively affects overall efficiency. Our data shows that the Toxic 6900 XT, when using Toxic Boost, draws 36% more power for its 8% performance gain. I don't see this as a problem though –  if you're spending this much money on a GPU, you're obviously not worried about the financial side, and it's pretty much always been the case that power has to increase exponentially to get as much out of a GPU as you can.

In actual fact, Sapphire's Toxic Boost one-click overclock works so well that I couldn't get anything more from the GPU when manually overclocking. There is of course sample-to-sample variance at play, so it could be that my card is just a bit limited, but Sapphire has pushed the GPU pretty hard with Toxic Boost.

So, should you buy the Sapphire Toxic RX 6900 XT? With an MSRP of £1299, it's only going to be for the most cash-rich of enthusiasts, and given the current state of affairs where Sapphire's own Nitro+ 6900 XT is currently listed at £1399.99, I'd say this card will be priced closer to £2000 when it hits the retail channel.

The obvious question to ask would be, ‘why not just buy an RTX 3090?' and I think that is a valid question. For as close as the Toxic comes to the RTX 3090, it is 3% slower at 4K, and that's when the GPU is at it's limit, whereas there is still scope to overclock the RTX 3090 and stretch that lead further. With an air cooled card, you obviously don't have to find room for a 360mm radiator, either.

Then again, I think there is a niche audience of enthusiasts out there who would buy a GPU like this, just because it's something fun and something different. For 99% of buyers, it's not going to make any sense. But for that cash-rich group of enthusiasts, the overclockers who want to extract every last megahertz from the GPU, and for those where every single point matters with that 3DMark score… that's who Toxic is for. On a technical level, what Sapphire has done here is very impressive indeed.

The Toxic 6900 XT is available from Overclockers UK HERE.

Discuss on our Facebook page HERE.

Pros

• On average 8% faster than reference, when testing at 4K.
• More-or-less as fast as RTX 3090 when using Toxic Boost.
• 400W power limit.
• Clean aesthetic.
• 360mm AIO means thermals are not a concern, even when using Toxic Boost.

Cons

• Significantly less efficient than reference.
• Fans can get a bit raucous under load.
• Likely going to cost significantly more than the £1299 MSRP.

KitGuru says: It's clearly not going to appeal to everyone, but the Sapphire Toxic RX 6900 XT is a highly impressive piece of engineering designed for serious enthusiasts.

Become a Patron!